Latest / The Official SaaStr Podcast: SaaS | Founders | Investors / The Top 10 Things to Know Before You Deploy Your First AI SDR With Jason Lemkin and Chief AI Officer Amelia Lerutte

Transcript

The Top 10 Things to Know Before You Deploy Your First AI SDR With Jason Lemkin and Chief AI Officer Amelia Lerutte · The Official SaaStr Podcast: SaaS | Founders | Investors

0:00
SPEAKER_01: Deployed more than one AISDR.
0:04
SPEAKER_01: And so somebody asked us this question the other day,
0:05
SPEAKER_01: and we thought it'd be fun to go through it live as well,
0:08
SPEAKER_01: which is,
0:09
SPEAKER_01: you know,
0:09
what are the top 10 things you should know before rolling out
0:12
your very first AISDR
0:14
SPEAKER_01: if you haven't deployed one yet?
0:15
SPEAKER_01: Or maybe this is your next deployment.
0:18
SPEAKER_01: Or maybe this is your next deployment and you want to know what,
0:21
SPEAKER_01: you know,
0:21
SPEAKER_01: tips and tricks to avoid and what to do to make it successful.
0:25
SPEAKER_01: So that's the context of today.
0:28
SPEAKER_01: We'll go through these 10 as always at our beginning.
0:32
If you want to see some of the tools and agents that we use,
0:35
they are on saster.ai slash agents.
0:38
SPEAKER_01: You can go there.
0:39
SPEAKER_01: We link to a lot of our agents that we use there.
0:42
SPEAKER_01: Okay.
0:43
SPEAKER_01: Without further ado,
0:44
SPEAKER_01: top things,
0:45
top 10 things to know before rolling out your first AISDR call
0:48
is being recorded.
0:49
SPEAKER_02: And we'll go Amelia.
0:50
SPEAKER_02: Sorry.
0:51
SPEAKER_02: I know I w I just want to make sure this is good.
0:53
Can we make sure it's streaming live and is there any way we
0:55
can silence the note takers?
0:57
SPEAKER_02: I just want this to be good content.
0:58
SPEAKER_01: It is streaming live.
0:59
SPEAKER_01: I don't know how to silence the note takers.
1:00
SPEAKER_02: Okay.
1:04
SPEAKER_02: All right, we'll figure that out for the next one.
1:05
SPEAKER_02: Okay, I just don't want to ruin the content.
1:07
SPEAKER_02: I don't, sorry to interrupt everybody, but I don't see it on LinkedIn.
1:11
SPEAKER_02: I mean, on Twitter.
1:14
SPEAKER_00: Okay, hold on.
1:22
SPEAKER_00: It says streaming.
1:23
SPEAKER_02: This meeting is being recorded by Update AI.
1:29
SPEAKER_01: Oh, let me try and mute everybody.
1:32
SPEAKER_02: Yeah, I can't take it.
1:32
SPEAKER_02: Otherwise, I got to leave this meeting.
1:33
SPEAKER_01: OK, Jason, you're going to have to unmute since I muted everybody.
1:50
SPEAKER_02: Cool.
1:51
SPEAKER_02: Cool.
1:51
SPEAKER_02: Sorry, let's just get it right.
1:52
SPEAKER_02: Let's just double check.
1:53
SPEAKER_02: I apologize to whoever's actually.
1:55
SPEAKER_02: But if it doesn't stream, it's
1:58
SPEAKER_00: I see it on your Twitter.
2:00
SPEAKER_02: You do?
2:00
SPEAKER_02: I just don't see it.
2:01
SPEAKER_02: Yeah.
2:01
SPEAKER_02: Okay.
2:02
SPEAKER_02: I'll send you the link.
2:04
SPEAKER_02: Oh, I see.
2:05
SPEAKER_02: Got it.
2:06
SPEAKER_02: Okay.
2:06
SPEAKER_02: Nevermind.
2:07
SPEAKER_02: Sorry to interrupt everybody.
2:09
SPEAKER_02: Some of our services have degraded.
2:11
SPEAKER_02: Some have improved.
2:13
SPEAKER_00: Okay.
2:13
SPEAKER_00: Some of our agents have degraded.
2:15
SPEAKER_00: What?
2:15
SPEAKER_00: Some of our agents have degraded.
2:17
SPEAKER_02: All right.
2:17
SPEAKER_02: Okay.
2:17
SPEAKER_02: We've muted everybody.
2:18
SPEAKER_02: We've made sure their streaming's working.
2:20
SPEAKER_02: Let's hit it.
2:21
The 10 things Amelia's going to lead us to know before rolling
2:24
out your first
2:25
SPEAKER_02: AISDR,
2:26
SPEAKER_02: one of the top 20 questions we get.
2:28
SPEAKER_02: And then if we have time at the end,
2:29
SPEAKER_02: Amelia,
2:29
SPEAKER_02: I assume we can,
2:30
SPEAKER_02: some folks have already done this multiple times.
2:32
SPEAKER_02: We can do as much Q&A as we have time for, right?
2:35
we can do q a and then we can talk about whatever else you want
2:37
to talk about um
2:38
towards the end because yeah i think some of this we've covered
2:41
but i think it's
2:41
important to know um if you're going to roll out your very
2:44
first aisdr just because
2:46
that is our use case that we have multiple actually aisdrs now
2:50
i'll preface in
2:51
saying that you probably only need one um and then to point
2:54
number one that i'll
2:55
drive right into you probably only need one vendor to do this
2:58
like we have
2:59
multiples for a lot of different reasons but most of what you
3:02
will need to do slash
3:03
SPEAKER_01: want to do with your agents
3:05
you'll probably be fine using just one vendor versus using the
3:09
four or five that we
3:10
SPEAKER_01: have for outbound.
3:12
SPEAKER_01: And I'll kind of explain why.
3:14
SPEAKER_01: So just coming into it first,
3:18
SPEAKER_01: The tool doesn't really matter.
3:19
SPEAKER_01: Like you're going to see today,
3:21
I interchange different screenshots from our different AISDRs
3:25
that we have.
3:26
SPEAKER_01: But you don't need to, again, you don't need to use multiple vendors.
3:29
I would pick one that does the majority or bulk of what you
3:32
want to do with an AISDR agent.
3:35
SPEAKER_01: And roll it out from there.
3:37
You may need multiple in the sense if your outbound agent truly
3:41
only does outbound
3:42
SPEAKER_01: and then you need to add an inbound agent that truly only does inbound.
3:45
SPEAKER_01: So maybe at the max, you end up with two.
3:48
SPEAKER_01: But I would say you don't need as many as we have today.
3:50
We just have it hyper segmented with different platforms that
3:55
do different
3:56
SPEAKER_01: specialties pretty well.
3:57
SPEAKER_01: So that's been our school of thought.
3:59
SPEAKER_01: But
4:00
SPEAKER_01: I will say,
4:00
SPEAKER_01: you know,
4:01
the tool doesn't really matter as much as the strategy and also
4:04
just making sure
4:05
SPEAKER_01: that your human playbook works first.
4:07
SPEAKER_01: So we famously said before,
4:10
SPEAKER_01: like,
4:11
SPEAKER_01: you know,
4:11
we're not going to roll out our we didn't roll out our first
4:14
human like AI.
4:16
We didn't roll out our first AISDR until we knew exactly what
4:20
was working with our
4:22
SPEAKER_01: human SDRs,
4:23
SPEAKER_01: what was working or maybe not working as well.
4:26
in our playbook so that we could give our agents just the very
4:29
best context of
4:31
SPEAKER_01: what's working,
4:32
SPEAKER_01: who should we be reaching out to,
4:34
SPEAKER_01: what messaging works with that,
4:35
SPEAKER_01: what segmentation should we be doing to roll out to the AISDR?
4:40
Because you have to remember that when you roll it out to an AI
4:42
versus a person,
4:44
SPEAKER_01: It's going to take all your context word for word, right?
4:47
SPEAKER_01: Like it's going to use that to build out its brain.
4:50
SPEAKER_01: It's going to use that to reach out to prospects.
4:53
SPEAKER_01: It's going to use that to follow up with any of your inbounds.
4:55
Like anything you do,
4:57
it's going to be 100% based on the context you give this AI SDR
5:03
agent.
5:04
And for those purposes,
5:05
you really should know what's working before you do that.
5:07
SPEAKER_01: Like I've seen some companies we've talked to,
5:10
SPEAKER_01: which are like million dollar,
5:11
SPEAKER_01: you know,
5:11
SPEAKER_01: AR companies,
5:12
SPEAKER_01: somewhere over 100 million companies.
5:13
SPEAKER_01: We're just rolling out AISDRs.
5:16
SPEAKER_01: And I'm like, OK, well, how are you training it?
5:17
SPEAKER_01: They're like, oh, well, you know, we're just testing.
5:19
SPEAKER_01: We're testing copy that we haven't used before.
5:22
SPEAKER_01: We're trying something new.
5:24
SPEAKER_01: I think that's the wrong way to go about it.
5:26
I think you should have something that marginally works and
5:30
then use the AISDR to
5:32
SPEAKER_01: get scale out of it.
5:35
Or maybe to summarize those two types,
5:36
because we hear this so often,
5:38
right?
5:39
If you haven't gotten any outbound to work,
5:43
buying an AI to do it,
5:44
it's not going to help you.
5:46
SPEAKER_02: You have to prove it.
5:47
SPEAKER_02: Any of the agents we use,
5:49
SPEAKER_02: any of the dozens of agents we use,
5:51
SPEAKER_02: we are taking a playbook that at least works a little bit.
5:54
SPEAKER_02: It doesn't have to be perfect.
5:55
SPEAKER_02: You don't have to have a 90% close rate,
5:57
SPEAKER_02: but something that is proven,
5:58
SPEAKER_02: copy,
5:58
SPEAKER_02: cadence,
5:59
SPEAKER_02: segment,
6:00
SPEAKER_02: but you have to have been able to close some customers yourself.
6:03
SPEAKER_02: You basically have to have at least done founder led sales,
6:06
SPEAKER_02: before you can hand it off to an agent.
6:08
SPEAKER_02: It has to work.
6:09
SPEAKER_02: They have to take you.
6:10
SPEAKER_02: And the ultimate goal here is to clone the best person on your team.
6:14
SPEAKER_02: If it's just you, clone you.
6:15
If you have four folks on your sales team and one is crushing
6:18
it at outbound or
6:19
SPEAKER_02: only one can really pull it off,
6:20
SPEAKER_02: all you want to do is clone.
6:21
All these products really do in the beginning is clone that
6:24
person,
6:25
right,
6:25
with more knowledge.
6:26
SPEAKER_02: Would you agree, Amelia?
6:28
SPEAKER_01: Yeah, I agree 100%.
6:29
SPEAKER_01: I think a lot of these tools,
6:30
SPEAKER_01: too,
6:31
if you haven't looked at them or maybe you haven't fully run
6:34
one yet and you're
6:35
SPEAKER_01: just going to deploy one,
6:36
a lot of what you'll see is sometimes I think the vendors will
6:39
try and steer you
6:40
SPEAKER_01: somewhat in the wrong direction,
6:41
which is why I don't think the tool matters as much as long as
6:44
you know you should
6:46
SPEAKER_01: be using it at the very start.
6:48
for things that are already working like a lot of vendors will
6:50
tell even some of
6:51
our vendors will tell you oh you can just use it for pure
6:53
outbound it's fine to
6:54
test things but then when you go look at the response rates of
6:57
course it's going to
6:57
be lower than something that's working um something that you
7:00
know that you want to
7:01
give scale to maybe something that you know your human aes or
7:05
sdrs won't do like
7:07
I do think that's a little bit of a trap to be wary of is I do
7:11
think sometimes the
7:12
vendors will try and say that they've had good success with
7:15
other customers on
7:16
SPEAKER_01: doing,
7:17
SPEAKER_01: you know,
7:17
SPEAKER_01: new use cases,
7:19
SPEAKER_01: just rolling out brand new,
7:20
SPEAKER_01: you know,
7:21
SPEAKER_01: just completely cold outreach.
7:23
SPEAKER_01: And yeah, to some degree, does that work?
7:25
SPEAKER_01: Sure.
7:26
SPEAKER_01: But you might be disappointed with it,
7:27
SPEAKER_01: too,
7:27
SPEAKER_01: if you only do it that way versus scaling out something that,
7:30
SPEAKER_01: you know,
7:30
SPEAKER_01: already works.
7:32
SPEAKER_02: Yeah,
7:32
SPEAKER_02: and I want to stay on track,
7:33
SPEAKER_02: but the interesting thing is,
7:35
SPEAKER_02: going to Amelia's point to me before,
7:36
SPEAKER_02: this is a mistake that both raw startups,
7:38
SPEAKER_02: scaling folks,
7:39
SPEAKER_02: and very large companies we talk to make the mistake.
7:42
SPEAKER_02: The raw startups,
7:43
SPEAKER_02: you know,
7:44
SPEAKER_02: they've raised a token amount of money they want,
7:46
SPEAKER_02: a tool to magically get them customers.
7:48
It can do that if you've already proven you can do it yourself
7:51
and you feed what
7:52
SPEAKER_02: works to the AISDR.
7:54
SPEAKER_02: If you can't, the odds that works are extremely low, right?
7:57
SPEAKER_02: And then on the other end of the spectrum,
7:58
SPEAKER_02: some of the largest leaders we've talked to,
8:00
SPEAKER_02: folks,
8:01
SPEAKER_02: multi-billion dollar public companies and bigger,
8:03
SPEAKER_02: just want to turn on a tool and not train it with what works.
8:06
SPEAKER_02: It's the same.
8:07
SPEAKER_02: It's a verge.
8:08
SPEAKER_02: Now, of course, they could.
8:09
SPEAKER_02: They are.
8:09
SPEAKER_02: Obviously,
8:10
SPEAKER_02: if you've got $2.5 billion in revenue,
8:12
SPEAKER_02: you've got at least a few SDRs that have performed.
8:14
SPEAKER_02: But maybe it's bought by a marketing team.
8:16
SPEAKER_02: Maybe it's bought by a different team.
8:18
SPEAKER_02: Maybe it's bought with a different perspective.
8:19
SPEAKER_02: But they just want to turn something on.
8:22
And the cost there may not be that important to the latter
8:25
category if you're big,
8:26
SPEAKER_02: but the results are terrible.
8:27
SPEAKER_02: They don't reproduce the best 10% of their team inside the app.
8:31
SPEAKER_02: And if you do nothing else the first 30 days,
8:33
SPEAKER_02: your job is to reproduce,
8:35
SPEAKER_02: to Amelia's first point,
8:36
to reproduce the human playbook works inside the training or
8:40
the UI UX of the AIS
8:42
SPEAKER_02: chair.
8:42
SPEAKER_02: If you don't do that, it won't work.
8:44
Or at least it'll work as randomly as spooling up some email
8:49
cadence tool from five
8:50
SPEAKER_02: or six years ago.
8:53
SPEAKER_01: Yep.
8:53
SPEAKER_01: I would say,
8:53
SPEAKER_01: yeah,
8:54
the magic or the magic unlock of these tools is using it in a
8:57
way different than
8:58
SPEAKER_01: those older tools,
8:59
SPEAKER_01: right?
8:59
Like different in a way than like the outreaches or like the
9:02
sales lofts of the world.
9:03
SPEAKER_01: Yeah.
9:04
Those were purely spool up,
9:07
do cold outbound,
9:08
at scale,
9:09
kind of see what results you get.
9:10
SPEAKER_01: And I think because of that,
9:11
that's why people try and do the same thing with something
9:13
that's an AISDR agent.
9:15
SPEAKER_01: But remember, it's an agent.
9:16
SPEAKER_01: It has a brain.
9:17
SPEAKER_01: It has context.
9:19
SPEAKER_01: Some of them, like ours, can see what's on your website.
9:22
SPEAKER_01: They can put that context into the email.
9:25
Think of it more as a living,
9:27
breathing agent versus just an outbound email sequencing tool.
9:31
SPEAKER_01: That's also the wrong way to...
9:33
go about it but i think that's why people do that i think
9:35
because they were just so
9:36
used to doing that previously um number two i would say if
9:42
you're rolling out your
9:42
first aisdr you know a lot of folks i think get caught up to in
9:47
deploying their
9:47
first aisdr and what are the use cases we're going to tackle
9:50
right like i think
9:51
naturally a lot of folks are like okay i've got outbound maybe
9:55
it's cold outbound
9:56
maybe it's slightly warmer inbound or for us you know we've we
10:00
all have leads we've
10:01
SPEAKER_01: never followed up with
10:02
and i think this is where the aisdr agent can really shine is
10:06
to have it do the
10:07
quote-unquote unglamorous work that your humans probably don't
10:11
want to do or maybe
10:12
gripe and moan about doing every week when they have to do it
10:16
and so the nice thing
10:17
about the agent is if you give them the right context
10:19
specifically for these leads
10:21
SPEAKER_01: so you know you can give it a set of
10:23
context or prompting the one the screenshot i have here is from
10:26
agent force that's
10:27
reaching out to prospects we our human team had never followed
10:31
up with right um
10:32
where speed matters here and because they had never done it now
10:36
we have an agent
10:37
that just does it 24 7 it'll pick the best time to email this
10:40
person because it
10:41
lives in salesforce you can see their salesforce contacts and
10:44
then it can follow up
10:45
SPEAKER_01: with them um
10:46
SPEAKER_01: Now,
10:46
SPEAKER_01: the open rates have been pretty good on this,
10:48
but also the meetings and booking rates have been a little
10:51
lower than maybe some of
10:52
SPEAKER_01: our other AISDRs.
10:53
SPEAKER_01: But that's because these were old leads.
10:55
SPEAKER_01: These were leads our team had never followed up with.
10:58
So not only had it been more than a few days since somebody
11:04
followed up with them,
11:05
SPEAKER_01: in some of these cases,
11:06
SPEAKER_01: it was more than a few weeks that nobody had ever followed up with them.
11:09
So in this case,
11:10
I gave it a pass on that because of the set of leads that we
11:13
gave it.
11:14
but just know it's still a good it's still a very good exercise
11:17
to do like what
11:18
we've gotten out of it has still been very useful for like
11:20
reactivating folks
11:21
getting people like re-engaged with us right sometimes that may
11:24
not turn into a
11:24
meeting because it is a an old lead that somebody didn't want
11:27
to follow up with but
11:28
at least now the agent followed up with them like we finally
11:31
were like okay
11:32
SPEAKER_01: This person has gotten a follow-up,
11:34
SPEAKER_01: yes,
11:34
SPEAKER_01: from our agent,
11:35
but they've gotten it in a way that makes sense and is a way at
11:39
scale where we
11:40
could follow up with all these leads 24-7 and try and re-engage
11:44
them versus just
11:45
SPEAKER_01: leaving them to die completely,
11:46
SPEAKER_01: right?
11:46
SPEAKER_01: Which is often the case before you have an AISDR.
11:50
Sometimes you have a lot of leads that just end up dying or
11:53
they time out,
11:54
right?
11:55
Speed to deal is obviously so important right now in the age of
11:58
AI that if you
11:59
SPEAKER_01: leave them to timeout,
12:00
then obviously that can affect the entirety of the deal cycle
12:04
and the customer
12:04
SPEAKER_01: cycle.
12:05
SPEAKER_01: So I think this is a great place where the AISDRs can shine,
12:09
SPEAKER_01: whether it's inbound,
12:10
SPEAKER_01: outbound,
12:10
SPEAKER_01: or even just having an AISDR that does ongoing customer engagement.
12:15
SPEAKER_01: I still see a lot of folks struggling to use their AISDR for that,
12:19
SPEAKER_01: and we use it all the time,
12:21
SPEAKER_01: every day.
12:22
I have agents that are just sub-agents going off in our tools
12:27
that are reaching out
12:28
SPEAKER_01: to current...
12:31
customers of ours,
12:31
current sponsors of ours,
12:33
to make sure they have everything they need.
12:35
SPEAKER_01: And it's such a great way to do it because, again, the agent is 24-7.
12:39
SPEAKER_01: It does it automated.
12:40
SPEAKER_01: It's not going to forget folks.
12:42
SPEAKER_01: A really great use case we have right now,
12:44
SPEAKER_01: for example,
12:44
SPEAKER_01: is I showed a couple of weeks ago,
12:46
SPEAKER_01: I vibe-coded our entire SaaS or sponsors portal.
12:49
And what the portal will do now is I helped build a little
12:53
custom microagent for it
12:54
SPEAKER_01: where it will run every couple days.
12:57
It will see who hasn't logged in to their clerk login when
13:00
we've invited them to
13:01
SPEAKER_01: our sponsor portal.
13:02
And then I'll just send them an email like,
13:03
hey,
13:04
you haven't logged in yet.
13:05
SPEAKER_01: Do you need help getting your login?
13:07
SPEAKER_01: Should I resend it?
13:08
SPEAKER_01: Things like that.
13:09
SPEAKER_01: Things that we would normally have to do ourselves.
13:11
SPEAKER_01: And so again,
13:12
SPEAKER_01: that's a really unglamorous workflow,
13:15
SPEAKER_01: but something that is so perfect for an agent to just do on its own.
13:18
SPEAKER_01: And also
13:18
SPEAKER_01: of low risk, right?
13:19
SPEAKER_01: It's just looking to see have they logged in,
13:21
SPEAKER_01: have they used the product,
13:23
SPEAKER_01: have they used the tool,
13:24
SPEAKER_01: and then sending them an automated workflow on that.
13:26
Again,
13:26
fairly unglamorous,
13:28
but something that we literally had to do this last year.
13:32
I was telling Jason last week,
13:34
I was looking at the amount of hours our events production
13:39
agency
13:39
had done for January and February for this year for SAS
13:43
Reannual that's coming up
13:43
in May versus the amount that our agency had done last year in
13:48
January and February
13:49
SPEAKER_01: for the SAS Reannual that was last May.
13:52
SPEAKER_01: And it's crazy because it was less than a tenth of the hours.
13:56
SPEAKER_01: Again,
13:57
not that I don't still need people and I still need them to do
14:00
some things that the
14:00
SPEAKER_01: agents can't do.
14:02
SPEAKER_01: Sometimes the agents just aren't good at calling the food truck people.
14:06
SPEAKER_01: I need a human to still do that today.
14:08
But it's crazy to see how much of that unglamorous kind of
14:13
monotonous style of work
14:16
SPEAKER_01: or customer check-ins like that,
14:18
SPEAKER_01: that an AI agent can just do really well.
14:21
Not only has it saved us in raw spend with some of our
14:25
contractors and agencies,
14:26
this,
14:27
I would say the sponsors are a lot happier that they have more
14:32
regular check-ins.
14:32
SPEAKER_01: They get a response faster.
14:34
SPEAKER_01: They're getting,
14:35
SPEAKER_01: you know,
14:35
SPEAKER_01: more proactive messaging from us,
14:38
SPEAKER_01: um,
14:39
SPEAKER_01: to do things like log in,
14:40
SPEAKER_01: submit the brew graphics,
14:41
SPEAKER_01: things like that.
14:42
Whereas back in the day,
14:43
we'd have to be more reactive of who's late,
14:45
who's missing.
14:46
SPEAKER_01: Right.
14:46
SPEAKER_01: Cause we were just trying to keep up with everything.
14:49
SPEAKER_02: Yup.
14:49
And I know we have questions we'll get to at the end,
14:51
but I think there's a good one.
14:53
SPEAKER_02: Um, in terms of,
14:55
SPEAKER_02: The use cases Amelia's talking about are small accounts,
14:57
SPEAKER_02: churn customer follow-ups,
14:59
SPEAKER_02: inbound leads,
15:00
SPEAKER_02: re-engagement campaigns.
15:01
For the most part,
15:02
we are using the current four third-party agents that we use,
15:07
right?
15:07
We are using Agent Force from Salesforce,
15:09
Qualified,
15:09
which Salesforce is buying.
15:11
SPEAKER_02: We're using Artisan, which we use all the time as a core outbound agent.
15:14
And recently we've used Monaco,
15:15
which is a new hot one,
15:17
to do raw outbound.
15:18
SPEAKER_02: They've done most of this.
15:20
SPEAKER_02: The stuff Amelia's talking about that is niche but important to us,
15:24
SPEAKER_02: a sponsor portal,
15:25
SPEAKER_02: reaching out to those customers,
15:27
SPEAKER_02: reaching out to those partners proactively,
15:29
SPEAKER_02: that doesn't exist off the shelf the way we needed it,
15:31
SPEAKER_02: so we had to build it ourselves.
15:32
SPEAKER_02: We vibed this ourselves.
15:35
SPEAKER_02: Our automated VP of marketing and our sponsor portal,
15:39
SPEAKER_02: it just wasn't,
15:40
SPEAKER_02: the off-the-shelf tools are so dated we couldn't do it.
15:44
we did that they're cool but we did them reluctantly and if we
15:46
could have bought it
15:47
we would have like you don't want to build this stuff over many
15:49
long weekends and
15:50
you don't want to maintain it um but uh just to highlight what
15:53
we built and what we
15:54
SPEAKER_02: uh what we use but we'd much rather buy
15:57
SPEAKER_01: We'd much rather buy.
15:59
SPEAKER_01: We did a session,
16:00
SPEAKER_01: I think two weeks ago that we went over the 90,
16:03
SPEAKER_01: 10 rule of how we determine how we build versus buy.
16:06
SPEAKER_01: Um, if you want to hear our deep dive on that.
16:08
SPEAKER_01: And then to that point too,
16:09
SPEAKER_01: on our AI VP of marketing,
16:11
SPEAKER_01: one of the sessions we're going to do at SAS re-annual,
16:13
SPEAKER_01: um,
16:13
SPEAKER_01: during the deploy summit on Tuesday.
16:15
SPEAKER_01: So everyone's just going to show how to build their agents.
16:17
SPEAKER_01: I'm actually going to rebuild that agent live.
16:19
SPEAKER_01: So if you want to come to that,
16:20
SPEAKER_01: we're also going to live stream it,
16:22
SPEAKER_01: but if you want to come to that,
16:22
SPEAKER_01: I'm just going to rebuild what I did.
16:24
SPEAKER_01: Um,
16:25
SPEAKER_01: with our AI VPN completely live.
16:28
SPEAKER_01: So fingers crossed that nothing goes wrong,
16:30
SPEAKER_01: but I'm going to do that there to show folks how to do it.
16:32
SPEAKER_01: So that'll be fun.
16:35
SPEAKER_01: Okay.
16:36
Tip number three,
16:36
if you're rolling out your first AISDR is to segment
16:40
ruthlessly.
16:41
SPEAKER_01: And I do mean that word ruthlessly.
16:43
Like we do it to the point where again,
16:46
the tool doesn't even matter at that point.
16:48
SPEAKER_01: They all sort of function in the back end the same way.
16:51
SPEAKER_01: Having used, you know,
16:52
We use four tools in our daily rotation,
16:55
but I've tried maybe over a dozen AISDR tools.
16:58
They all have some version of this functionality where you can
17:02
tell it who you want
17:03
SPEAKER_01: it to reach out to,
17:04
and then you can give it specific context for that segment
17:08
versus giving it an
17:09
SPEAKER_01: all-in-one brain.
17:11
SPEAKER_01: It's actually something where like our...
17:13
SPEAKER_01: inbound qualified agent,
17:14
SPEAKER_01: which is,
17:15
SPEAKER_01: you know,
17:16
SPEAKER_01: becoming a Salesforce agent,
17:18
SPEAKER_01: didn't had kind of like a big brain,
17:20
SPEAKER_01: which I'll show you in this deck later.
17:22
SPEAKER_01: And then I actually started segmenting more with that one too.
17:25
It was like,
17:25
okay,
17:25
well,
17:26
in our inbound,
17:27
we actually have so many buckets.
17:29
First,
17:29
I was like,
17:30
okay,
17:30
well,
17:30
there's,
17:31
you know,
17:31
they're inbound to the website.
17:32
SPEAKER_01: So I'll just...
17:33
SPEAKER_01: consider that the inbound bucket.
17:35
SPEAKER_01: And I was like, no, actually, that's wrong.
17:36
SPEAKER_01: Like, we have folks that are inbounding and they're brand new to Saster.
17:40
We have folks that are inbounding that came there because they
17:43
saw an ad on one of
17:44
SPEAKER_01: our social media platforms.
17:45
We have folks that are inbound to the website because they were
17:48
a prior sponsor or
17:49
SPEAKER_01: they're a current customer.
17:50
SPEAKER_01: I was like, actually, there's a lot of buckets within inbound.
17:53
So we've actually now even segmented our inbound agent,
17:56
whereas I previously was only segmenting
17:59
SPEAKER_01: our outbound agent.
18:00
But either way,
18:01
you should segment ruthlessly so that the context you give the
18:05
agent is tighter.
18:06
It's exactly what you wanted to say to that person versus if
18:10
Jason comes to the
18:10
SPEAKER_01: website and he just has one big brain,
18:12
SPEAKER_01: then your agent's going to act off,
18:14
SPEAKER_01: OK,
18:15
SPEAKER_01: he's coming to the website.
18:16
SPEAKER_01: What is he maybe looking at?
18:17
SPEAKER_01: Is he looking at your pricing page?
18:19
SPEAKER_01: Is he looking at a features page?
18:21
SPEAKER_01: Whereas if you segment that, it's OK, Jason is on the website.
18:25
SPEAKER_01: He's actually a former LAPS customer.
18:28
SPEAKER_01: And he's looking at the pricing page.
18:30
SPEAKER_01: Now let me give him context to say what's new with the product, right?
18:34
SPEAKER_01: Maybe he's churned in 2022.
18:36
SPEAKER_01: And you have that in Salesforce.
18:37
SPEAKER_01: And you can pull up that context to your inbound agent.
18:39
SPEAKER_01: Like, hey, a lot has changed with your product since then.
18:43
SPEAKER_01: You can highlight those new features to that person
18:46
SPEAKER_01: with something like an inbound agent.
18:48
And so that's where the segmentation helps add a lot of context
18:52
and just have a lot
18:53
better of a conversation between the agent and the human being
18:57
that's interacting
18:58
SPEAKER_01: with this agent.
18:59
SPEAKER_01: Same thing goes for inbound.
19:00
We segment ruthlessly to the point of I am updating these
19:05
segments literally every day.
19:08
SPEAKER_01: And I literally will go through like our,
19:10
SPEAKER_01: our website visitors,
19:12
SPEAKER_01: I'll resegment folks that are opening emails,
19:15
SPEAKER_01: I'll segment folks,
19:16
SPEAKER_01: again,
19:16
SPEAKER_01: I'll do folks that are have that are already customers of ours,
19:19
SPEAKER_01: but maybe haven't logged in when they need to.
19:22
So I ruthlessly segment and then I make sure that the context I
19:25
give each agent is
19:26
hyper tailored to that person to that audience to make it have
19:30
a better response
19:31
SPEAKER_01: rate versus just having it be one big brain of an AI agent.
19:35
SPEAKER_02: And Emilio, just one question on this point.
19:37
SPEAKER_02: None of the AI tools we use today,
19:39
SPEAKER_02: which are all great,
19:40
but none of them can really use AI to auto segment in a way
19:44
that can deliver these
19:45
SPEAKER_02: results.
19:45
SPEAKER_02: Is that accurate?
19:47
It's accurate with an asterisk of,
19:49
I do a lot of segmentation with Claude Cowork.
19:53
SPEAKER_01: Like last night,
19:54
SPEAKER_01: for instance,
19:54
I was doing- But if I buy any of the tools we use or any of the
19:56
tools we've tried,
19:58
SPEAKER_02: this screenshot is nine segments,
20:00
SPEAKER_02: right?
20:01
But if we're doing 1,000 contacts,
20:03
we probably have effectively 100 segments,
20:06
right?
20:07
SPEAKER_00: Yeah.
20:07
SPEAKER_02: Maybe not.
20:09
SPEAKER_02: But AI should be able to build these on its own, right?
20:11
SPEAKER_02: In theory.
20:12
But today,
20:13
I'm just asking today,
20:14
and all these products are getting so much better.
20:16
SPEAKER_02: But as we do this,
20:18
no one can create these types of segments autonomously that you
20:22
have to sign up to
20:23
SPEAKER_02: do this yourself like a marketing manager would,
20:25
SPEAKER_02: right?
20:25
SPEAKER_02: Correct.
20:25
yeah you do yeah like a yeah exactly like an ops person or like
20:28
a marketing manager
20:29
you can't you have to have the wherewithal to know to segment
20:33
versus a lot of them
20:35
by default to that point um jason will do it just as one big
20:39
brain like it'll say
20:39
okay just run one campaign and keep adding leads to it right um
20:43
that's the wrong
20:44
way to go about it again not that it won't work to some some
20:47
effective degree but
20:49
SPEAKER_01: We have seen now,
20:50
SPEAKER_01: because we've run these for 10 months now,
20:52
it just works a heck of a lot better to have it segmented and
20:55
have the context that
20:57
SPEAKER_01: you're giving the agent hyper-specific to that segment.
21:03
SPEAKER_01: Okay,
21:03
SPEAKER_01: number four,
21:04
SPEAKER_01: and this is what we have found to be true kind of early on,
21:07
SPEAKER_01: but it's still true today,
21:09
SPEAKER_01: which is consistency beats brilliance.
21:11
And what we mean by that is it does not need to be the best
21:15
email on planet Earth.
21:16
And you should not expect the best email or the best message if
21:21
it's a messaging
21:21
SPEAKER_01: agent or the best text message or the best phone call.
21:25
SPEAKER_01: It does not need to be the best.
21:28
I will tell you we've sent,
21:30
this is just artisan,
21:31
but it's sent 40,000 messages.
21:33
SPEAKER_01: Okay,
21:34
I think our qualified is later on this deck that's sent like
21:36
100,000 messages
21:37
SPEAKER_01: because that's an inbound.
21:40
And then with Salesforce,
21:41
we're probably close to like 200,000 messages sent now.
21:45
Will I,
21:46
will I raise my hand and say,
21:47
these are the greatest emails since sliced bread?
21:50
SPEAKER_01: No, of course they're not.
21:51
SPEAKER_01: They're at scale.
21:52
Like our AI does a pretty good job and a solid output of what
21:58
the emails are.
21:59
And you'll see,
22:00
you'll see here,
22:01
like there's ebbs and flows in that chart.
22:03
SPEAKER_01: And that's because I'm reloading the agent, right?
22:04
SPEAKER_01: You'll see it's
22:05
It spikes up every couple of days when I reload it because we
22:09
hyper-segment,
22:09
SPEAKER_01: and then it spikes down when that sequence starts to finish itself.
22:13
SPEAKER_01: And so, again, just consistency here.
22:17
SPEAKER_01: And,
22:17
SPEAKER_01: again,
22:18
SPEAKER_01: using all the points one through three,
22:20
SPEAKER_01: right,
22:20
SPEAKER_01: of,
22:20
SPEAKER_01: like,
22:20
SPEAKER_01: the very best messaging you already know will work,
22:23
SPEAKER_01: the very best subject lines,
22:25
SPEAKER_01: getting hyper-segmented with it.
22:26
SPEAKER_01: And then when you layer in consistency, that will –
22:31
give you so many great results versus a human that is
22:35
inconsistent that maybe
22:36
doesn't or ignores the training and context right because when
22:39
you're giving your a
22:40
ai sdr the training and context it will remember it every time
22:44
um sometimes it will
22:45
try and iterate it on its own so that's something you need to
22:47
be mindful of
22:49
SPEAKER_01: to make sure it's not goal seeking too much.
22:51
SPEAKER_01: And, you know, it's still speaking to folks in the way you want it to.
22:56
But this is something just to note here of the consistency that
22:59
you're going to get
23:00
SPEAKER_01: with an AISDR.
23:01
It's going to be worth the trade off of maybe the email not
23:05
being the most perfect
23:06
SPEAKER_01: email on planet Earth.
23:08
SPEAKER_01: But sometimes,
23:09
SPEAKER_01: you know,
23:10
perfect is always the enemy of good or even the enemy of just
23:13
getting it done.
23:14
Like I see so many folks that just have like AISDR paralysis
23:18
because they're like,
23:19
SPEAKER_01: well,
23:19
SPEAKER_01: you know,
23:20
I started using one of the tools you guys use and my output
23:23
wasn't that great.
23:25
SPEAKER_01: It was okay.
23:26
SPEAKER_01: And I'm like, well, one, you could always make it better, right?
23:28
SPEAKER_01: Getting hyper segmented and then getting hyper specific with content,
23:32
SPEAKER_01: use content that works and then see what you get.
23:34
SPEAKER_01: Like you'll probably get it to pretty good and pretty good.
23:37
SPEAKER_01: I would say is the bar.
23:38
SPEAKER_01: It does not need to be brilliant.
23:40
SPEAKER_01: It does not need to be the best email on planet earth or an
23:43
SPEAKER_01: or even an email that you might write yourself.
23:45
It just needs to be pretty good to get this person down the
23:49
path you want them to go.
23:54
SPEAKER_01: Okay,
23:55
SPEAKER_01: number five is for us,
23:57
SPEAKER_01: at least with the amount of agents we have,
23:59
SPEAKER_01: you need at least one,
24:01
but ideally two essential humans to run all this and to deploy
24:05
your AI SDR.
24:09
SPEAKER_01: You can't have zero people and...
24:12
certainly it's become harder for us even right now with like a
24:15
lot of my time being
24:16
split between act like saster annual production versus like
24:20
that was time i i
24:21
typically had spent before march managing agents even that has
24:25
gotten like a little
24:26
SPEAKER_01: bit harder for us here at saster
24:28
SPEAKER_01: And so having 0.5 of my workload is actually not ideal either,
24:32
SPEAKER_01: but the annual is a time sensitive matter.
24:36
So just something to keep in mind here is ideally you need two
24:40
folks internally to
24:41
SPEAKER_01: run this,
24:42
SPEAKER_01: even if you can just find one,
24:43
SPEAKER_01: but then you will need to find a backup,
24:45
SPEAKER_01: right?
24:47
SPEAKER_01: Not that your agents are going to leave, but what if that human leaves?
24:49
You do need to have a backup system and a plan to keep all your
24:53
agents running.
24:54
And so it's essential to have at least two folks working on
24:58
this,
24:58
monitoring the agents.
25:00
Because if not,
25:00
you'll see,
25:01
like I just showed in that artisan chart,
25:03
your agents can sit idle.
25:05
Unless it's an inbound agent where ours is on full automatic,
25:09
that one doesn't idle as much.
25:11
SPEAKER_01: But our inbound agents and many of your inbound agents, because of how...
25:16
SPEAKER_01: we do this tooling,
25:17
SPEAKER_01: which I do think is the best way to do it,
25:20
SPEAKER_01: your agents will sit idle if you don't give them new segments,
25:23
SPEAKER_01: new context,
25:24
SPEAKER_01: new people to go out to.
25:26
SPEAKER_01: It will finish its sequence in a couple days,
25:28
SPEAKER_01: and then it will sit there until you give it the next one.
25:31
And so I think that's why it's essential to have two folks to
25:35
backfill each other
25:36
SPEAKER_01: to make sure your agents are always optimized.
25:39
Otherwise,
25:40
you're wasting a lot of time and money if they're just idling
25:42
all the time.
25:44
SPEAKER_01: I will say,
25:44
SPEAKER_01: too,
25:44
SPEAKER_01: the reason I showed a screenshot on this one of Monaco is,
25:47
SPEAKER_01: like,
25:47
SPEAKER_01: Monaco is a very interesting,
25:50
SPEAKER_01: distinct use case.
25:50
SPEAKER_01: We're actually not their perfect ICP.
25:53
Their use case is more for folks that are starting up and maybe
25:56
don't have – maybe
25:58
SPEAKER_01: you've got some messaging figured out and you have,
26:00
SPEAKER_01: let's say –
26:02
20 or 50 customers,
26:03
but you want to get the next,
26:04
you know,
26:04
100 customers.
26:06
Theirs is great because theirs will actually fill your pipeline
26:09
more automatically
26:10
SPEAKER_01: based on lookalikes.
26:11
SPEAKER_01: So that's what I really like about it is I haven't had to,
26:14
SPEAKER_01: that one hasn't sat idle for us because it's just really
26:18
filling the pipe when it finishes the sequence it'll grab
26:21
another lookalike um
26:23
company close to our audience it'll see you know okay um did
26:27
did snowflake reply to
26:28
you they did okay great i'm gonna go so go go you know outbound
26:32
to another
26:32
company's like snowflake okay did open ai respond to you they
26:36
did okay great
26:36
SPEAKER_01: Now I'm going to go out to other companies that are like an open AI,
26:39
SPEAKER_01: which there aren't many of,
26:40
SPEAKER_01: but it'll do things like that to keep the pipeline full,
26:43
SPEAKER_01: which I think is pretty cool versus some of the other agents we've tried.
26:46
SPEAKER_01: So they don't all do that.
26:48
SPEAKER_01: It's just something to be aware of.
26:49
SPEAKER_01: Again,
26:50
SPEAKER_01: agnostic of the tool,
26:51
SPEAKER_01: something to maybe ask your tool provider too,
26:54
SPEAKER_01: if you're worried about your agents maybe sitting idle too much,
26:57
or if there's anything else you can do to make sure data is
27:01
always flowing
27:02
interior agents like right now i don't i don't know if i'll be
27:04
able to finish this
27:05
pre-annual but right now i'm trying to trying to see how i can
27:07
actually get some of
27:08
our our agents all hooked up together to talk to each other
27:12
more regularly to solve
27:13
this problem um but again it's it's kind of a later down the
27:17
road problem once you
27:18
SPEAKER_01: have your your agent fully deployed and and it's starting to work for you
27:24
SPEAKER_01: Okay, number six, and this is hyper important.
27:28
SPEAKER_01: Read everything,
27:29
I would say in the first 30 days and then read almost
27:32
everything every day after
27:34
SPEAKER_01: that.
27:34
So if you're rolling out your first AISDR agent,
27:38
you are going to want to read everything.
27:39
And maybe you're going to want to read everything before you
27:41
hit publish or send or
27:43
whatever the equivalent is in your tool before it goes out to
27:46
somebody in your
27:48
SPEAKER_01: outbound or inbound list or whatever it might be.
27:51
you're going to want to read everything during that initial
27:54
deployment to catch
27:55
SPEAKER_01: things like,
27:56
SPEAKER_01: you know,
27:57
SPEAKER_01: sometimes I think our agents would spell saster incorrectly.
28:01
Not that it was wrong,
28:02
but there's two capital S's and sometimes it would lowercase
28:05
it.
28:05
SPEAKER_01: And so I would make a rule
28:07
in the agent to say you know always capitalize saster as you
28:12
know the s's are
28:13
capital um and then i would write in rules like always use
28:16
saster in the subject
28:18
line because we have a bit of a brand and so typically it works
28:21
better if saster's
28:21
in the subject line versus it looking more generic um there
28:25
were other things that
28:26
SPEAKER_01: we would catch in the first you know messages at scale of
28:31
SPEAKER_01: Sometimes it would,
28:32
SPEAKER_01: you know,
28:33
SPEAKER_01: because again,
28:33
SPEAKER_01: our events are timed and we use a lot of these for our events.
28:36
Sometimes it would use the old dates because it would just
28:38
scrape the internet and
28:39
SPEAKER_01: then it would find like the old dates.
28:40
SPEAKER_01: And so it was super important just to read everything.
28:43
SPEAKER_01: I also think it's the only really way to learn it.
28:47
SPEAKER_01: is to see what does your agent actually output.
28:50
If you only care about the inputs and you're not caring about
28:53
the outputs,
28:54
SPEAKER_01: you're only seeing half the picture.
28:56
SPEAKER_01: You do want to see what your agents are saying, what they're doing.
28:59
SPEAKER_01: And I just took the screenshot from Tuesday.
29:02
And this is an important nuance here that I'll just explain
29:05
this workflow through.
29:06
SPEAKER_01: This is from our qualified amount agent.
29:08
SPEAKER_01: Somebody was asking about
29:12
SPEAKER_01: adding an additional ticket to SaaS or annual later this year.
29:16
SPEAKER_01: And this is,
29:17
SPEAKER_01: again,
29:18
SPEAKER_01: not a hard conversation for Amelia AI or inbound agent to answer,
29:22
SPEAKER_01: but just a nuanced one,
29:23
SPEAKER_01: right?
29:24
SPEAKER_01: Because it requires knowing, okay, who is this person?
29:29
SPEAKER_01: How many tickets have they already bought?
29:31
SPEAKER_01: What price did they buy it at?
29:32
SPEAKER_01: Because then they need a retroactive discount.
29:34
SPEAKER_01: So basically what it did is the agent asked them,
29:38
SPEAKER_01: those questions and looked it up and,
29:40
SPEAKER_01: you know,
29:40
SPEAKER_01: to see if it was in Salesforce,
29:41
SPEAKER_01: um,
29:42
SPEAKER_01: ask them a few clarifying questions of,
29:43
SPEAKER_01: okay,
29:43
SPEAKER_01: like,
29:44
SPEAKER_01: is this,
29:44
SPEAKER_01: is this third person from your same company?
29:47
SPEAKER_01: And then great, like, what's your email address?
29:48
SPEAKER_01: Cause then we can send you the code for it.
29:50
SPEAKER_01: And so again,
29:51
SPEAKER_01: fairly hyper nuance,
29:52
SPEAKER_01: but something that,
29:53
SPEAKER_01: um,
29:54
SPEAKER_01: we read through and we go through just to see where our agent needs help.
29:57
Even still today,
29:58
like even 10 months later,
29:59
I am still reading things that our agent is doing.
30:02
SPEAKER_01: to make sure it's accurate,
30:04
SPEAKER_01: to make sure it's what we want it to say to folks,
30:06
SPEAKER_01: and to make sure we're represented in the right way with our agents,
30:10
SPEAKER_01: also because so many people see them,
30:12
that it's also just good to make sure it's saying things that
30:16
are accurate and not
30:18
going off the rails right again are rarely have our agents gone
30:21
off the rails or
30:22
hallucinating stuff but they have it happens every every now
30:25
and then and so for me
30:27
it's something where every day i'm just you know even if i only
30:30
have 10 minutes i'm
30:31
just doing like a speed run to be like okay what is our inbound
30:34
i'll just go
30:35
through like a couple inbound conversations see if i spot check
30:38
anything same thing
30:39
SPEAKER_01: i'll go through our outbound agents i'll see if i
30:41
If I see anything in their messages,
30:43
that's maybe off or maybe needs a tweak.
30:45
SPEAKER_01: And then if there is,
30:46
then I'll just add that to the AI's context so that it's
30:49
continually improving with
30:51
SPEAKER_01: you,
30:51
SPEAKER_01: right?
30:51
SPEAKER_01: So that's something you're going to have to do ongoing, no matter what.
30:55
SPEAKER_01: I would say a heck of a lot more in the early days,
30:57
SPEAKER_01: but you're still going to have to do it every day after.
31:04
SPEAKER_01: Okay,
31:04
SPEAKER_01: number seven,
31:05
if you are deploying an AISDR your first time would be to
31:08
budget at least two weeks
31:10
SPEAKER_01: of RAM time,
31:11
SPEAKER_01: sometimes specifically to with an outbound agent,
31:13
SPEAKER_01: they can take a little bit longer,
31:15
especially if it's warming up anything like specific IPs or
31:18
domains or dedicated
31:19
SPEAKER_01: email addresses.
31:21
SPEAKER_01: Sometimes that process just takes two to three weeks in and of itself,
31:25
SPEAKER_01: regardless of the tool you use,
31:27
SPEAKER_01: regardless of what you're going to do next with it.
31:29
SPEAKER_01: Sometimes they just have that built in ramp time.
31:31
And then the other factor of the two week ramp time is nothing
31:36
is set and forget.
31:37
SPEAKER_01: I know some of our,
31:38
SPEAKER_01: even some of our vendors,
31:39
SPEAKER_01: I think sometimes are guilty of saying things like this.
31:45
And if they give you an FDE,
31:46
sometimes,
31:47
you know,
31:47
it's a heck of a lot easier for sure.
31:50
But as you've just seen in steps one through six,
31:52
nothing is ever truly set and forget.
31:54
SPEAKER_01: You do need to make sure you are constantly checking in on your agents.
31:58
SPEAKER_01: Again, a lot in the beginning and still ongoing every day thereafter.
32:02
But the iteration and the context and training you're going to
32:06
give your agents
32:06
SPEAKER_01: just takes a lot of calibration and testing.
32:09
SPEAKER_01: And it is a real-time investment.
32:14
SPEAKER_01: It is, again, something where...
32:16
SPEAKER_01: Typically,
32:17
SPEAKER_01: sometimes I have a little bit more time in a day than another,
32:20
SPEAKER_01: but it's something where every day I have to no matter what.
32:23
Even if I can only sneak in a 15-minute speed run at the end or
32:29
beginning of the
32:29
SPEAKER_01: day to go through what's going on with our agents and that's kind of it,
32:33
SPEAKER_01: that's still every day,
32:34
SPEAKER_01: right?
32:34
SPEAKER_01: That's still like having a short one-on-one with a human counterpart.
32:39
And so you've got to budget at least two weeks in the beginning
32:43
of almost nothing else.
32:45
And I think that is sometimes surprising to folks is you got to
32:49
almost put
32:49
SPEAKER_01: everything else aside to really make this work.
32:51
SPEAKER_01: Because if you're watching or listening to this and you're like,
32:54
SPEAKER_01: okay,
32:54
SPEAKER_01: I haven't really thought about one through six,
32:56
but I'm already pretty far along with a vendor that I think I
32:59
want to use for an
32:59
SPEAKER_01: AISDR.
33:00
SPEAKER_01: And then you get that
33:02
product you get into the back end and you're like oh shoot I do
33:05
need to do those
33:05
steps one through six probably steps one through six will take
33:08
you about two weeks
33:09
to do right like figure out what what seat like what copy works
33:13
what subject lines
33:14
work what time of day works um hyper segmenting your base and
33:18
then multiplying
33:20
those steps of like one through four by each segment okay great
33:24
now right um
33:25
actually do the deployment either with an FDE or with their
33:28
team like that alone
33:31
SPEAKER_01: with
33:32
Some other meetings will probably take you two weeks just right
33:35
there as you put
33:35
SPEAKER_01: everything together and then work through,
33:37
SPEAKER_01: you know,
33:38
SPEAKER_01: okay,
33:38
do I want this AISDR to flow information back to something like
33:43
a Salesforce?
33:44
SPEAKER_01: Is it going to live in the silo?
33:46
SPEAKER_01: How is it going to interact with other things?
33:49
SPEAKER_01: Okay, once I've added an outbound AISDR,
33:53
SPEAKER_01: Do I want to add an inbound one?
33:54
SPEAKER_01: Do I want to add one for customer success?
33:57
SPEAKER_01: There's just all these things that kind of snowball.
33:59
And so it's not even really the implementation that takes the
34:03
longest bit of time.
34:04
It's a little bit of a mental overhead and tax it takes to set
34:08
all this up.
34:08
There's obviously some real steps here that we've outlined that
34:11
you will need to do
34:12
SPEAKER_01: when you roll out your first AISDR.
34:14
SPEAKER_01: But anything we have found takes at least two weeks.
34:17
SPEAKER_01: Even Monica, which is super great,
34:19
and has you know it's really good at the agent refilling itself
34:22
and keeping our
34:23
pipeline that part is really nice but even the beginning i
34:27
think we got it down to
34:27
like one and a half weeks it still took us one and a half weeks
34:30
to get it all set
34:30
up right like nothing we have found is instant so i would be
34:35
wary of anything that
34:36
SPEAKER_01: you see is like an instant aisdro roll it out today
34:40
SPEAKER_01: I'd be wary of that for a lot of reasons.
34:43
And two,
34:44
I would say a lot of this is the prep that only you can do,
34:46
right?
34:46
SPEAKER_01: Like most of these vendors, regardless of the tool, will help you out.
34:50
SPEAKER_01: Some of them will give you an FDE.
34:52
SPEAKER_01: Most of them, more of them should.
34:55
SPEAKER_01: But regardless of that,
34:56
I would say budgeting at least two weeks of RAM time is fairly
35:00
standard now with
35:01
all the prep that you'll just need to do too to make sure it's
35:03
successful.
35:08
SPEAKER_01: Okay,
35:08
SPEAKER_01: number eight,
35:09
SPEAKER_01: we can ask this quite a bit,
35:11
SPEAKER_01: I think,
35:11
because we're one of the few folks that have a voice agent
35:15
along with our chat and
35:17
SPEAKER_01: email agents.
35:19
SPEAKER_01: I will say we don't use like a text messaging agent at this point.
35:23
I know a few of our vendors are actually playing around with
35:26
that,
35:26
so we'll see how that goes.
35:28
And so we have a voice agent too who's multimodal,
35:33
so she has video as well.
35:36
SPEAKER_01: But what we found,
35:37
SPEAKER_01: because she's been...
35:39
SPEAKER_01: Amelia AI,
35:40
SPEAKER_01: our chat and video agent,
35:43
SPEAKER_01: has been live since about November.
35:46
SPEAKER_01: Jason's AI, if you talk to his Delphi, you can do it also in...
35:52
chat or voice you can that's been rolled out for almost a year
35:55
now and what we
35:56
found for the last year is that most people still prefer to to
36:00
chat with an AISDR
36:02
um over chat um or over email versus either phone voice or
36:09
video um I think it's
36:11
SPEAKER_01: Some of it's just a comfortability thing, right?
36:13
SPEAKER_01: Like, literally, I was on the phone with somebody yesterday.
36:15
SPEAKER_01: I was doing a speaker's call for their speaker's astrianial.
36:19
SPEAKER_01: And she went to astrianial to look at stuff.
36:21
SPEAKER_01: And she goes, is this?
36:22
SPEAKER_01: She's like, oh, this AI agent just popped up.
36:26
SPEAKER_01: And it looks like you.
36:27
SPEAKER_01: I was like, yeah, it is me.
36:30
SPEAKER_01: That's Amelia AI.
36:31
I literally went to the qualified studios to record that and do
36:35
motion capture of
36:36
SPEAKER_01: her so that I would have a full multimodal agent.
36:41
SPEAKER_01: which is a fun thing to do.
36:43
SPEAKER_01: I was like, but still most people want to talk to her in a chat format.
36:48
I was like,
36:49
some people do like the video because then,
36:50
you know,
36:50
you don't have to type.
36:51
SPEAKER_01: And some folks really like that,
36:53
SPEAKER_01: that they can,
36:54
SPEAKER_01: they feel like they're actually talking to someone.
36:56
SPEAKER_01: Um,
36:57
A few of the folks who talked to Amelia AI before our London
37:00
event came up to me at
37:02
SPEAKER_01: our London event last December.
37:04
SPEAKER_01: And they were like, oh, hey, I talked to Amelia AI.
37:06
SPEAKER_01: I had a conversation with her.
37:08
SPEAKER_01: And I'm like, that's fantastic.
37:09
SPEAKER_01: That is exactly what it's there for.
37:12
SPEAKER_01: She's there 24-7 to answer questions,
37:15
SPEAKER_01: to route people,
37:16
SPEAKER_01: to just be helpful in a way that no human can be because she's on 24-7.
37:21
So it's nice to have an extra layer where if people want to
37:26
interact with her,
37:28
video they can and I think the important learning here is um
37:32
let people pick
37:33
Like I don't think you need to strictly be chat only or text or
37:39
email only or that
37:41
SPEAKER_01: you need to be only video.
37:42
SPEAKER_01: Like we let people choose.
37:44
If you want to chat with one of like Amelia AI or Jason AI,
37:48
you can chat with them.
37:48
SPEAKER_01: If you want to use the video, you can have a video call with them.
37:51
SPEAKER_01: If you want to just do a text-based conversation with them,
37:53
SPEAKER_01: you can just do a text-based conversation with them.
37:55
And so I think it's just important to let people pick their
38:00
preferred format and
38:01
SPEAKER_01: whatever they're comfortable with.
38:02
and let them interact with your agent from there versus trying
38:05
to maybe force them
38:06
SPEAKER_01: down one road or another.
38:08
SPEAKER_02: Yeah,
38:08
SPEAKER_02: this is,
38:09
SPEAKER_02: and this particular use case is us using Qualified,
38:12
SPEAKER_02: which has,
38:12
SPEAKER_02: I don't even know what they use.
38:13
SPEAKER_02: They don't use Synthasia, someone asked.
38:15
SPEAKER_02: They use, I think, a more cost-effective vendor, right?
38:17
SPEAKER_02: Underline.
38:18
SPEAKER_01: It's called Tavis, yeah.
38:19
SPEAKER_02: Tavis, yeah, that's what they use.
38:23
SPEAKER_02: The 85% of the 15% that want to speak with the digital you,
38:28
SPEAKER_02: do we have a sense,
38:29
SPEAKER_02: do they go as far deep or do they revert to chat?
38:33
SPEAKER_02: Do you happen to know?
38:35
um they typically stay in video mode i think because they want
38:37
to talk to the video
38:38
yeah they want to talk to the video yeah it's like because she
38:42
defaults her default
38:43
is chat right so like her default is like you can i'll show you
38:47
you can just go
38:48
yeah i know it but i mean yeah okay i'm like yeah you can just
38:51
go to anyone on this
38:52
SPEAKER_01: call that can just go to sassradio.com you'll see her this is her
38:56
SPEAKER_02: Oh, I get it.
38:57
SPEAKER_02: So people do do it.
38:59
SPEAKER_02: I do think they're,
39:00
SPEAKER_02: you know,
39:00
SPEAKER_02: and this 85-15 was pretty,
39:02
SPEAKER_02: our first AI agent was this general agent called Delphi that does,
39:06
SPEAKER_02: you know,
39:06
still use hundreds of thousands of times where people can shout
39:09
visual version of
39:09
SPEAKER_02: me.
39:09
These were the rough,
39:10
I think these were the rough 85-15 ratios too for voice versus
39:13
chat.
39:13
SPEAKER_02: They actually deprecated video.
39:16
SPEAKER_02: Unfortunately, they didn't think it worked well.
39:17
But I do think there's a benefit that it looks trustworthy and
39:23
modern and it connects with you.
39:24
SPEAKER_02: So I do think there's a benefit.
39:25
But maybe our learning for B2B use cases,
39:27
it may be more indirect than direct,
39:29
right?
39:31
SPEAKER_02: If you're not ready to start off with voice, don't kill yourself.
39:34
SPEAKER_02: Just get the basic chat going.
39:36
SPEAKER_02: And then down the road,
39:37
SPEAKER_02: once it's dialed in and it's working,
39:39
SPEAKER_02: if you want to have some fun and add it,
39:41
SPEAKER_02: it's cool,
39:42
SPEAKER_02: but it doesn't need to be a day one implementation,
39:44
SPEAKER_02: the video voice,
39:45
SPEAKER_02: right?
39:46
SPEAKER_01: Right.
39:47
SPEAKER_01: And our qualifier had already been on for...
39:51
maybe at least a quarter by the time we added um the voice and
39:56
video component to
39:57
it so it was already up for a bit when we started the process
40:00
of you know going to
40:01
their studios recording amelia ai um deciding what we wanted
40:06
her how we want her
40:08
like voice and video interactions to be versus um chat
40:12
obviously maybe this is
40:15
SPEAKER_01: obvious to everyone,
40:15
but I'll just say it for context is you have to have a lot more
40:19
guardrails though
40:20
SPEAKER_01: in a voice slash video agent than you do a text-based agent.
40:25
SPEAKER_01: Maybe for obvious reasons,
40:27
SPEAKER_01: but because it's a live avatar,
40:29
SPEAKER_01: sometimes like the wild card of this,
40:31
SPEAKER_01: we were like,
40:32
SPEAKER_01: well,
40:32
SPEAKER_01: we...
40:33
SPEAKER_01: We think we know what people will ask her,
40:35
SPEAKER_01: which is,
40:35
SPEAKER_01: you know,
40:36
we have at least a quarter of data of what people have been
40:39
chatting with her
40:39
SPEAKER_01: about.
40:40
SPEAKER_01: And so that's how we trained the voice and video agent.
40:43
And then we're like,
40:44
but then people started asking her more personal questions
40:46
about me.
40:47
SPEAKER_01: And so that was learning.
40:48
SPEAKER_01: They were like, oh, well,
40:51
SPEAKER_01: can I talk to Real Amelia too?
40:53
SPEAKER_01: Or they'd be like, oh, what's Amelia's email address?
40:56
Which she does give people in certain contexts,
40:59
not just if you blatantly ask her.
41:02
Or they'd ask other questions like,
41:04
oh,
41:05
I think people were like,
41:06
where's she from?
41:06
SPEAKER_01: Or like, what's her LinkedIn?
41:07
SPEAKER_01: And so they just started asking more personal questions because again,
41:10
I think some people thought it was actually me like chatting
41:13
back to them in the
41:14
SPEAKER_01: voice component,
41:16
SPEAKER_01: which is okay.
41:17
SPEAKER_01: But yeah, I was like, no, that's still...
41:19
SPEAKER_01: It's a slightly different brain than our chat interface of Qualified.
41:26
SPEAKER_01: Because obviously,
41:27
SPEAKER_01: again,
41:27
SPEAKER_01: we were like,
41:27
we're just gonna have to put on more guardrails because what we
41:30
want her to do is
41:31
SPEAKER_01: still steer the conversation towards the same goals.
41:34
SPEAKER_01: And sometimes when you're talking in person,
41:36
you can get off track or you can get distracted or you start
41:38
asking Amelia AI about
41:40
SPEAKER_01: real Amelia.
41:41
And so we had to put in just more training and context and
41:43
guardrails for Amelia AI
41:45
SPEAKER_01: to be like,
41:46
SPEAKER_01: okay,
41:46
SPEAKER_01: it's fine if they start asking those and you can answer it like this way.
41:49
SPEAKER_01: We literally told her what to say, but,
41:51
then also try and get them back to whatever they were doing on
41:53
the website like if
41:54
they were trying to buy a ticket try and steer them back to buy
41:57
your ticket if they
41:58
were trying to book a meeting show my meetings booker to just
42:00
book a meeting with
42:01
SPEAKER_01: them um and so we had to put again just like additional things you don't
42:04
necessarily need to do in chat um in the video format but i
42:08
think people have fun
42:09
with it to be honest um again for folks that prefer this format
42:13
but yeah it was not
42:13
something we rolled out day one it's we already knew what was
42:16
kind of working in
42:17
SPEAKER_01: our
42:17
SPEAKER_01: in our inbound agent had already been up for more than a quarter.
42:20
SPEAKER_01: And then we started to tackle the video and voice component.
42:25
But it's fun,
42:26
and I'm actually figuring out a way for Amelia.ai to be live at
42:29
Sastra Annual.
42:30
SPEAKER_01: So if you guys have not talked to her on the website,
42:32
you will have another version of her that will be at Sastra
42:35
that you can talk to.
42:38
SPEAKER_01: Okay,
42:39
SPEAKER_01: on that though,
42:39
SPEAKER_01: actually,
42:40
SPEAKER_01: what's interesting about Amelia.ai is,
42:43
SPEAKER_01: you could probably guess by now,
42:44
SPEAKER_01: it is a very person-dependent deployment,
42:47
SPEAKER_01: which actually we would recommend avoiding for other folks.
42:51
SPEAKER_01: For many reasons.
42:52
SPEAKER_01: And I think,
42:53
SPEAKER_01: you know,
42:54
if you look at things like Amelia AI or the Jason AI that's
42:57
live on the site,
42:58
SPEAKER_01: those are so tied to us.
43:00
SPEAKER_01: Like those are those are not only our entities,
43:03
SPEAKER_01: like the Jason AI talks like Jason,
43:06
SPEAKER_01: like it's kind of crazy,
43:07
SPEAKER_01: like it has your personality.
43:09
SPEAKER_01: A bit of Amelia AI kind of talks like Amelia.
43:11
SPEAKER_01: It has obviously video that I captured that was me.
43:17
SPEAKER_01: And so I think having these implementations that we have,
43:21
SPEAKER_01: I'm not sure actually that you should do that unless,
43:24
SPEAKER_01: unless,
43:24
SPEAKER_01: asterisk,
43:25
SPEAKER_01: like us,
43:25
SPEAKER_01: you're on the exec team,
43:27
SPEAKER_01: right?
43:27
Like you're either the founders,
43:28
you're invested in the business,
43:29
you're on the exec team.
43:31
um and there's there's less of a risk to do it that way but i
43:34
would say you know i
43:35
always i famously roast qualify for doing this all the time
43:38
like their piper agent
43:39
is based on one of um one of their original aes who's still
43:43
there now she's like
43:44
been promoted i'm like but what if real piper ever leaves like
43:48
i actually don't
43:49
even know how that works of like do they have her likeness
43:52
because it's also on
43:52
billboards like it's on their website like these are just
43:55
things you may not have
43:56
SPEAKER_01: thought about yet in a
43:58
in a very person-based deployment where if there's any chance
44:02
that person might
44:02
SPEAKER_01: leave,
44:03
SPEAKER_01: obviously there's some inherent risk in that.
44:04
SPEAKER_01: Now,
44:05
you can do things to obviously safeguard yourself from that if
44:08
there's somebody you
44:09
really want to base your AI agent on to make sure you have
44:13
maybe the rights to the
44:14
SPEAKER_01: agents if they ever leave the company,
44:16
SPEAKER_01: et cetera.
44:19
that gets tricky obviously there hasn't been like rulings on
44:23
that kind of stuff yet
44:24
that hasn't happened yet in real life um so we'll see whatever
44:26
happens with that
44:27
down the road but just know it's it's kind of tricky when you
44:30
build these agents so
44:32
closely um deployed on persons and personalities right like our
44:37
personas are tied
44:38
SPEAKER_01: to these two agents that we have live on the website um
44:43
SPEAKER_01: And so I would say that's something that's a risk there,
44:45
SPEAKER_01: but also it's a risk that a lot of what is running Amelia AI...
44:50
SPEAKER_01: Like,
44:51
SPEAKER_01: I don't think,
44:52
I think David from our sales team has logged into Qualified
44:55
once to set up the
44:56
SPEAKER_01: meetings booker in his calendar.
44:57
And the other day we logged in for maybe the second time in 10
45:00
months so that he
45:00
could adjust his calendar settings because he had it cutting
45:03
off too early in the
45:04
SPEAKER_01: day.
45:04
SPEAKER_01: But other than that,
45:06
SPEAKER_01: I'm like,
45:06
SPEAKER_01: he doesn't have like that tribal knowledge that I have of,
45:10
SPEAKER_01: I know who our FD is at Qualified.
45:12
SPEAKER_01: I know who it is at Agent Force.
45:15
SPEAKER_01: I know where to tweak the context.
45:17
SPEAKER_01: Like, again, having person-dependent,
45:20
SPEAKER_01: Maybe it's obvious,
45:20
but you can quickly fall into a trap here if it's too dependent
45:25
on one employee's
45:27
SPEAKER_01: maybe voice or knowledge of the context that's being fed to the agents,
45:30
SPEAKER_01: et cetera.
45:32
SPEAKER_01: And then I just put some screenshots here of like,
45:34
SPEAKER_01: this also becomes obviously more of a risk the longer you have an agent.
45:37
SPEAKER_01: So now that these screenshots are from Qualified,
45:40
SPEAKER_01: but you can see like even on March 10th,
45:42
SPEAKER_01: I didn't scroll down all the way,
45:44
SPEAKER_01: just between 9 a.m.,
45:47
SPEAKER_01: to 10 p.m.
45:48
SPEAKER_01: Those are Amelia video chats.
45:51
SPEAKER_01: That's just video.
45:51
SPEAKER_01: There's more than a whole page.
45:53
That screenshot's maybe 20 deep,
45:55
and then there's the rest of the page before 9 a.m.
45:58
SPEAKER_01: So my AI was shown to maybe 30 people in that day.
46:03
SPEAKER_01: That becomes risky if you're like...
46:05
SPEAKER_01: okay, what if that person ever leaves?
46:06
SPEAKER_01: But now 30 people have already talked to her AI agent the whole day.
46:11
SPEAKER_01: Plus in the entirety of time,
46:12
SPEAKER_01: if you actually look at our whole funnel,
46:15
SPEAKER_01: we've had more than 1.5 million sessions just overall,
46:19
SPEAKER_01: just in qualified,
46:19
not counting our other agents with text-based communication or
46:24
video communication.
46:25
And so it just becomes a real risk if you don't have some sort
46:29
of like backups or
46:31
SPEAKER_01: documentation for these agents going forward,
46:33
SPEAKER_01: because the more you use them,
46:34
SPEAKER_01: you'll see like we've had qualified up,
46:37
SPEAKER_01: you know,
46:38
SPEAKER_01: we rolled it out last August or so.
46:41
maybe six-ish months now,
46:44
but six months at 1.5 million sessions,
46:46
that's a lot of sessions.
46:48
That's a lot of exposure to not only an AI agent is it a lot of
46:53
exposure to,
46:54
SPEAKER_01: but it's a lot of exposure to an agent that's dependent on
46:58
SPEAKER_01: not only Amelia AI being Amelia,
47:00
but real Amelia giving her context and making sure she has all
47:03
the right data to
47:05
SPEAKER_01: push back to folks.
47:05
SPEAKER_01: So it's a lot.
47:07
SPEAKER_01: So it can scale pretty fast.
47:08
And so I think the speed is sometimes something folks don't
47:12
account for when they
47:14
SPEAKER_01: roll out their first agent because you don't realize how fast it is,
47:17
SPEAKER_01: right?
47:17
SPEAKER_01: Like once you get past your two week maybe implementation period,
47:22
Once you deploy your agent and your agent is fully deployed,
47:25
that thing will just go,
47:26
right?
47:26
SPEAKER_01: Like it's just going to go every day constantly.
47:28
SPEAKER_01: This is 1.5 million sessions.
47:30
SPEAKER_01: That's on one website, by the way.
47:32
SPEAKER_01: That was not counting.
47:33
SPEAKER_01: That was just like SASTR annual.
47:36
SPEAKER_01: That wasn't plus SASTR, plus SASTR London.
47:38
SPEAKER_01: Like it just can scale so quickly.
47:42
That is something you'll want to be mindful of too because I
47:44
feel like folks don't
47:45
maybe fully understand that they can quickly spin themselves up
47:51
and have like,
47:51
SPEAKER_01: again,
47:53
SPEAKER_01: almost infinite conversations at scale with people.
47:54
SPEAKER_02: Hey, Amelia, I don't know.
47:57
SPEAKER_02: It's not the point, but you raised one that comes up a lot.
48:01
SPEAKER_02: If you had to spitball,
48:04
How much data,
48:06
you said 1.5 million sessions went into Qualified since we've
48:09
started,
48:09
right?
48:09
SPEAKER_02: Yeah.
48:11
SPEAKER_02: How many sessions,
48:12
how much data do you need for most of these products to work
48:15
versus do you need to
48:17
get bigger and be a bigger startup until there's enough data
48:21
generated for them to
48:22
SPEAKER_02: be effective?
48:23
SPEAKER_01: Yeah, I put it on the next slide here.
48:26
Because it is our last point of your data slash your
48:30
fundamentals have to work
48:32
SPEAKER_01: before you have an AI agent.
48:35
SPEAKER_01: So qualified in particular,
48:38
SPEAKER_01: just on this agent,
48:41
SPEAKER_01: you can see it scrapes our website every couple of days.
48:43
SPEAKER_01: So every couple of days,
48:45
SPEAKER_01: and you can force it,
48:46
SPEAKER_01: like you could do it every day,
48:47
SPEAKER_01: but I have it scraping sastra.com every couple of days.
48:50
SPEAKER_01: Sastra annual, I think once a week.
48:52
SPEAKER_01: Sastra London,
48:53
I think I turned off for now because we're not promoting that
48:55
event at the moment.
48:57
um and then i you'll see i also have like custom snippets i've
49:01
uploaded custom pdfs
49:02
custom q a that i've loaded into it so it now has all together
49:08
it has almost 6 000
49:10
pieces of context that it's working off of so that's a lot most
49:15
people don't have 6
49:16
SPEAKER_01: 000 pieces
49:18
of content or context to feed the agent um i would say a bare
49:25
minimum if that's the
49:26
question um again i would focus on what works but you're gonna
49:33
quickly find with
49:34
SPEAKER_01: any agent you roll out um
49:37
whether it's sales or an aisdr or not you need more data than
49:41
you think you even
49:42
for me i've needed more data than i thought we had and we had a
49:45
lot of data um case
49:47
in point like these agents have to answer questions and if you
49:52
want them to answer
49:52
questions autonomously guess what you have to do a brain dump
49:55
and say okay let me
49:57
answer all the questions i think a customer would ask if i
50:00
haven't documented that
50:01
SPEAKER_01: yet which i think most people
50:02
haven't to be honest um some maybe some of you have but it just
50:06
becomes this
50:07
snowball of almost every day you're gonna find a scenario like
50:12
that where you're
50:12
like oh um folks are starting to ask my agent about xyz things
50:17
so now i have to
50:17
train it on that oh now they're asking about you know maybe
50:20
this particular product
50:21
feature and i don't have documentation about that now i have to
50:24
make it then give
50:25
SPEAKER_01: it to the agent then let it deploy um
50:29
SPEAKER_02: You definitely need the documentation.
50:30
If you had to spitball how much traffic to my website for
50:36
something that relies on
50:37
SPEAKER_02: inbound and data capture,
50:39
SPEAKER_02: how much traffic do I need for my website?
50:41
SPEAKER_02: And for an outbound agent,
50:43
SPEAKER_02: how many at least slightly qualified leads,
50:48
SPEAKER_02: folks that you've collected,
50:50
SPEAKER_02: how big a list do you have to have,
50:51
SPEAKER_02: right?
50:52
Have created to feed it for both those inbound and outbound
50:55
actions to be effective.
50:57
SPEAKER_01: Yeah,
50:57
SPEAKER_01: I would say for the inbound piece for like website traffic,
51:01
SPEAKER_01: because a lot of this is obviously based on our website traffic.
51:03
SPEAKER_01: I think if you have less than,
51:06
SPEAKER_01: let's say,
51:07
SPEAKER_01: 10 to 20,000 visitors a month,
51:09
SPEAKER_01: it's probably not worth it.
51:10
SPEAKER_02: But that's a lot for a startup.
51:12
SPEAKER_01: It's a lot, I know.
51:14
SPEAKER_02: Yeah, 10 to 20.
51:15
SPEAKER_02: People, everyone kind of exaggerates, you know, see clearly.
51:18
SPEAKER_02: I know.
51:18
SPEAKER_02: But not a lot of folks have 10.
51:20
SPEAKER_02: Folks might not, to your marketing site, right?
51:23
SPEAKER_02: Yeah.
51:23
You might not have 10,000 visitors for your first 18 months as
51:27
a startup,
51:28
right?
51:28
SPEAKER_01: Right.
51:28
SPEAKER_02: It's entirely possible.
51:29
SPEAKER_02: So you would pass,
51:31
SPEAKER_02: or you would just pass for,
51:32
SPEAKER_02: you wouldn't pass for outbound necessarily,
51:33
SPEAKER_02: but you would pass for something-
51:34
pass for inbound like a qualified the way we use it you would
51:37
pass until you're 100
51:38
i would pass if you have less than and i know 10 to 25k seems
51:42
like a lot but
51:43
honestly we have more than that and if you look at specifically
51:46
at our you know web
51:47
obviously we have 1.5 million sessions just on one website in
51:50
six months so we have
51:51
a lot more than that but if you look at how that funneled down
51:55
right of what it
51:56
SPEAKER_01: does with people of okay how many completed conversations does it having
52:00
SPEAKER_01: Now, how many Amelia AI conversations is it having?
52:03
SPEAKER_01: Then how many, you know, this and that and how many meetings were booked?
52:06
SPEAKER_01: Like that all obviously dwindles down the funnel.
52:09
SPEAKER_01: And so I think if you don't have that volume at the very top,
52:14
SPEAKER_01: you're not going to get the results you want at the very bottom.
52:17
Okay,
52:17
just a challenge and then summarize what you've learned on
52:19
Outbound.
52:19
SPEAKER_02: So that makes sense.
52:20
But let's imagine I have no humans responding to my website,
52:26
to any leads or Outbound.
52:27
SPEAKER_02: Do I really need 10,000 visitors a month?
52:31
SPEAKER_02: Do I, I mean, and maybe we do for some of the tools we use, but I'm not,
52:36
SPEAKER_02: Intellectually, it makes sense.
52:37
SPEAKER_02: There's a different level where going to some of your early sites,
52:39
SPEAKER_02: if it's just not getting done,
52:40
I want an agent that is available 24-7 to answer prospects and
52:45
customer questions,
52:46
SPEAKER_02: right?
52:46
SPEAKER_02: Even if I don't have the volume.
52:49
SPEAKER_02: But maybe that's a different question.
52:51
no that's okay i think it's a good nuance i would say that's
52:54
maybe the ideal volume
52:55
but if you don't have that and you do want something where
52:59
again getting it done is
53:01
better than not having it at all then just do it right like if
53:04
you if you have any
53:05
inbound volume at all but you don't have the traffic just do it
53:10
but if you're like
53:10
SPEAKER_01: i'm in a boat where i don't have the volume and i
53:13
I don't have the volume of traffic and I don't get a lot of
53:16
inbound anyway,
53:17
SPEAKER_01: then I think that's maybe a slippery slope to thinking,
53:21
SPEAKER_01: oh,
53:21
SPEAKER_01: maybe an AISDR will fix it magically for an inbound agent.
53:26
SPEAKER_01: It won't if there's truly,
53:28
SPEAKER_01: if you have a double handicap there,
53:31
SPEAKER_01: if there's not enough something going on there,
53:34
SPEAKER_01: then maybe focus somewhere else.
53:36
SPEAKER_01: Yeah,
53:36
maybe focus on outbound first or maybe focus on current
53:39
customers first before you
53:40
SPEAKER_01: tackle inbound.
53:41
SPEAKER_02: Got it.
53:43
SPEAKER_01: Outbound is interesting, I would say.
53:46
SPEAKER_01: you know,
53:47
SPEAKER_01: I like to keep ours hyper-segmented,
53:49
SPEAKER_01: but if you think about just the pure math,
53:51
SPEAKER_01: right?
53:52
SPEAKER_01: Like let's just, I keep ours to about a thousand or so per campaign.
53:58
And then let's just say a campaign takes about a week and a
54:00
half to run,
54:01
right?
54:01
SPEAKER_01: Cause it's gonna do a first email,
54:02
SPEAKER_01: then it's gonna wait a couple of days,
54:04
SPEAKER_01: send a second one,
54:04
SPEAKER_01: then it's gonna wait a couple of days and then send a third one.
54:06
SPEAKER_01: That is like our cadence that we do with most of our outbound agents.
54:11
SPEAKER_01: So if, you know, a thousand people take a week and a half
54:14
SPEAKER_01: then you could do it in a month with like 4,000 people, right?
54:18
SPEAKER_01: Because if you're doing just at the bare minimum,
54:20
SPEAKER_01: let's say one sequence and you're loading it up once a week,
54:23
SPEAKER_01: you let it run for a week and a half,
54:24
SPEAKER_01: then yeah,
54:25
SPEAKER_01: I would say you could probably do it with like less than 5,000 contacts.
54:30
I think the issue people run into is after the first month,
54:33
they're like,
54:34
what do I do now?
54:36
SPEAKER_01: They're like, I ran through a lot.
54:40
SPEAKER_01: I had this list of 5,000 or maybe 10,000 people I want to go out to.
54:45
SPEAKER_01: And then after a quarter, I exhausted all those leads.
54:49
SPEAKER_01: So I think the true magic actually on Outbound is not figuring out.
54:53
SPEAKER_01: It's just continually figuring out who you want to reach out to.
54:57
SPEAKER_01: Whereas I have almost endless folks I always want to reach out to.
55:04
But I think for startups who are maybe starting out,
55:06
you may not have that,
55:07
right?
55:08
SPEAKER_01: And so it's just something to think about within AISDR is like,
55:11
SPEAKER_01: are you going to be able to continually feed it?
55:13
SPEAKER_01: Just like you would a sales team, right?
55:14
SPEAKER_01: Like you used to have to feed humans leads all the time.
55:19
SPEAKER_01: The AISDR is no different.
55:20
SPEAKER_01: You got to keep feeding it leads in this,
55:23
SPEAKER_01: not only the same way,
55:25
SPEAKER_01: but more because actually humans are way slower at outbound than AISDR.
55:29
SPEAKER_02: So let's imagine you're a startup,
55:32
SPEAKER_02: but you're not,
55:32
SPEAKER_02: it's not day zero,
55:33
SPEAKER_02: but you're only at a couple of million revenue.
55:35
Um,
55:35
you have a small list you've built,
55:38
um,
55:39
from inbound,
55:40
right from your website.
55:41
And then you're using clay or other tools and you're kind of
55:44
hand building a target list,
55:46
right?
55:47
Is that going to work for a sort of a pre-brand startup if I
55:52
build up a list of a
55:53
SPEAKER_02: couple thousand folks from an enrichment tool?
55:55
SPEAKER_02: Do some of these tools do it themselves?
55:57
SPEAKER_02: Like, will they automatically build the list for you?
56:00
SPEAKER_02: How does that work if you don't have a massive list already?
56:04
SPEAKER_01: Yeah, some of them are really good at lookalikes, right?
56:07
SPEAKER_01: Artisan has a lookalike feature.
56:08
SPEAKER_01: Clay has a lookalike feature.
56:10
If you have,
56:11
let's say,
56:11
a list of just 1,000 people,
56:13
then build lookalikes all day long.
56:15
SPEAKER_01: I'm not at that point where I want to... You saw it.
56:18
SPEAKER_01: Monaco's doing it for new target accounts.
56:22
SPEAKER_01: But for the other stuff I'm doing...
56:24
SPEAKER_01: more manually.
56:25
SPEAKER_01: I don't do lookalikes at all.
56:29
SPEAKER_01: But if you're not in that boat, then use lookalikes all day.
56:32
SPEAKER_01: They're pretty good at it.
56:34
SPEAKER_01: The Monaco one's pretty good at it.
56:35
SPEAKER_01: I've tested it more recently in Clay.
56:37
SPEAKER_01: It's pretty good at it.
56:39
I gave it,
56:40
I think I was telling you,
56:41
I gave it a list of folks who came to our
56:43
SPEAKER_01: CMO summit last year,
56:44
SPEAKER_01: which is like an invite only thing we do as part of faster,
56:47
SPEAKER_01: I said,
56:47
SPEAKER_01: for this purpose,
56:48
SPEAKER_01: actually,
56:49
SPEAKER_01: give me a lookalike of CMOs,
56:50
SPEAKER_01: because I only had like a list of,
56:52
SPEAKER_01: okay,
56:52
the top 1000 CMOs that I had narrowed it down to from like past
56:57
CMO summits,
56:58
SPEAKER_01: CMOs in our database,
56:59
SPEAKER_01: you know,
56:59
SPEAKER_01: you have to be over like 40 million.
57:01
SPEAKER_01: So it was a smaller pool than just our whole database of CMOs.
57:05
And then you'd have to,
57:06
then you also had to be at,
57:07
you know,
57:07
a somewhat agentic
57:10
SPEAKER_01: company.
57:10
SPEAKER_01: And so for that purpose,
57:11
SPEAKER_01: I was like,
57:12
SPEAKER_01: give me a lookalike because my list is less than 1000 people.
57:16
SPEAKER_01: And I want people who look exactly like this ICP already.
57:20
SPEAKER_01: And then I sent it out actually last week,
57:22
SPEAKER_01: I happen to use clay,
57:23
SPEAKER_01: you can use whatever vendor you want to do your lookalike,
57:26
SPEAKER_01: I happen to use clay.
57:27
And we had actually doubled the amount of CMO Summit attendees
57:31
in a week just by
57:32
SPEAKER_01: using a lookalike.
57:33
So I do think the lookalike audiences for outbound can work
57:36
pretty well if you only
57:38
SPEAKER_01: have like a set of 1000 people.
57:39
SPEAKER_01: just make sure you give it 1000 really good people.
57:43
SPEAKER_01: Like don't give it like,
57:44
SPEAKER_01: Oh,
57:44
SPEAKER_01: this Joe Schmo company churned or there,
57:47
SPEAKER_01: or are I only have like,
57:50
SPEAKER_01: customers that I maybe don't want the same DNA of like,
57:55
SPEAKER_01: give it your best customers and then make it look alike.
57:57
SPEAKER_01: But
57:58
SPEAKER_01: I do think for outbound,
57:59
you can get away with having a smaller list and maybe that's a
58:01
better place to
58:02
start with before you scale out the inbound is to use a lot of
58:07
those lookalike
58:07
SPEAKER_01: lists to start building out outbound.
58:09
SPEAKER_01: And then you can do other things like layer and ads, whatever.
58:13
SPEAKER_01: But that for us,
58:14
like even that just use case and scenario right there works
58:17
really well because I
58:19
had a much smaller list than I normally have to work with and
58:21
it still worked.
58:22
SPEAKER_02: So if you do it yourself with a human, you prove...
58:26
SPEAKER_02: the copy and context that works, that gets meetings, right?
58:31
SPEAKER_02: If your list is then too small,
58:33
you can get to initial scale for it to work with a thoughtful
58:36
copycat list,
58:37
SPEAKER_02: right?
58:37
I mean,
58:38
classic crappy copycat lists are a low effort thing like buying
58:41
a purchase list.
58:43
SPEAKER_02: But if you do it,
58:43
SPEAKER_02: if you spend the time to do it right,
58:45
SPEAKER_02: you really target the ICP that works and then customize the copy,
58:49
that's a hack to getting scale if you don't have it yet in the
58:52
early days,
58:53
SPEAKER_02: right?
58:53
SPEAKER_02: Is do the copycat really well.
58:55
But to your point,
58:58
I think when I had Clay copycat it,
59:02
it gave me like 2,500 more CMOs.
59:04
SPEAKER_01: And I was like, that's probably too many.
59:08
SPEAKER_01: So I went through it one by one.
59:10
SPEAKER_01: It took me a couple hours.
59:11
SPEAKER_01: I think it was Saturday.
59:13
SPEAKER_01: I went through it one by one while I was watching something on TV.
59:16
And I deleted all the people I didn't want because I was like,
59:18
you know what?
59:19
SPEAKER_01: The list is pretty good.
59:20
SPEAKER_01: But there's like 2,000 people here I probably don't want.
59:24
SPEAKER_01: So I deleted a bunch.
59:26
SPEAKER_01: I like hyper pruned it to get it to the best quality.
59:29
SPEAKER_01: But then, you know, that's how I doubled our...
59:32
double their attendees for that particular thing in a week so
59:34
it can work really
59:35
well but yeah it does take some time to go through it and like
59:38
i literally like
59:38
there were companies i hadn't heard of i was like i'm gonna
59:40
click them make sure
59:41
they're an ai there make sure their revenue is right like
59:44
double check it's the
59:44
right person um click their linkedin and make sure they
59:47
actually work there like
59:48
that can take for a list of 2 000 that can take a couple hours
59:51
um and i did nothing
59:53
else for those couple hours but it did work right so i think if
59:56
you're willing to
59:56
SPEAKER_01: put in the work that's a good hack
59:58
SPEAKER_02: Okay.
59:58
SPEAKER_02: Do we have,
59:59
SPEAKER_02: I know,
59:59
SPEAKER_02: I know we're over and people will leave,
1:00:00
SPEAKER_02: which is fine,
1:00:01
SPEAKER_02: but do you have 10 more minutes?
1:00:02
SPEAKER_02: I can ask you some, a few, there's a few good questions in the chat.
1:00:04
SPEAKER_02: You want me to ask you to wrap up?
1:00:06
SPEAKER_01: Yeah.
1:00:06
SPEAKER_02: This one's a good, a good basic one.
1:00:08
I think this was great as is how many people and visit and
1:00:12
leave when they realize
1:00:14
SPEAKER_02: they're talking to an AI agent.
1:00:16
SPEAKER_02: And in prior sessions,
1:00:17
SPEAKER_02: you've presented some rough data that says,
1:00:20
SPEAKER_02: we think people prefer to talk to our agent in many cases,
1:00:23
SPEAKER_02: right?
1:00:23
SPEAKER_02: Yeah.
1:00:25
But how many leave the chat or interaction or digital
1:00:30
immediately when they realize it's an AI?
1:00:33
SPEAKER_01: At least a couple of people a day.
1:00:34
SPEAKER_01: Yeah.
1:00:35
SPEAKER_02: Do they ask to talk to a human?
1:00:38
SPEAKER_02: Is this a big deal?
1:00:40
no they just ask to talk to a human um so i'll so i'll go over
1:00:44
an inbound case and
1:00:45
i'll go over an outbound case so in our inbound i can it tells
1:00:48
me immediately
1:00:49
because i have a slack set up like it will tell me immediately
1:00:51
in qualified if
1:00:52
there's somebody who's asking to talk to real amelia um if i'm
1:00:55
in a meeting i will
1:00:57
say give them my email or whatever and i'll talk to them later
1:01:00
i can't talk to them
1:01:01
SPEAKER_01: right now
1:01:02
SPEAKER_01: Like, Amelia, I could, but I can't come to the phone right now.
1:01:07
SPEAKER_01: And then there's a couple times a day I'll see at the end of the day,
1:01:09
SPEAKER_01: like,
1:01:09
SPEAKER_01: I have it,
1:01:10
SPEAKER_01: you know,
1:01:10
SPEAKER_01: flagging of,
1:01:12
SPEAKER_01: like,
1:01:12
SPEAKER_01: if there's a trigger of somebody got frustrated.
1:01:15
SPEAKER_01: It's funny.
1:01:16
SPEAKER_01: Like,
1:01:16
SPEAKER_01: first I started out with,
1:01:17
SPEAKER_01: like,
1:01:17
SPEAKER_01: oh,
1:01:18
SPEAKER_01: I'm going to flag in our inbound agents when people say something good.
1:01:20
SPEAKER_01: Now I'm like,
1:01:21
SPEAKER_01: no,
1:01:21
I actually just want to flag if they say something bad because
1:01:24
then I need to take
1:01:25
SPEAKER_01: an action.
1:01:25
SPEAKER_01: If they say something good, great.
1:01:27
SPEAKER_01: Like,
1:01:28
SPEAKER_01: I don't even care about that anymore.
1:01:29
I actually only care now if they say something bad or if they
1:01:32
need an escalation
1:01:33
SPEAKER_01: and they still need help.
1:01:34
SPEAKER_01: So now I kind of swap those filters.
1:01:37
SPEAKER_01: But it's pretty funny.
1:01:37
SPEAKER_01: So at least a couple times a day,
1:01:38
I'll get somebody where either they had an escalation or they
1:01:41
got frustrated
1:01:42
SPEAKER_01: talking to the AI.
1:01:45
SPEAKER_01: And then I think I said this to you a couple weeks ago.
1:01:47
SPEAKER_01: There's also this weird mix of people who try and trick our AI now.
1:01:52
SPEAKER_01: And I don't know what's up with that.
1:01:53
SPEAKER_02: And all the way to prompt injection and everything.
1:01:56
SPEAKER_02: Everyone tries to break them, right?
1:01:58
SPEAKER_01: Yes.
1:01:58
And so that is a weird phenomenon where I would say in the last
1:02:01
couple of weeks,
1:02:01
SPEAKER_01: that's picked up more.
1:02:02
SPEAKER_01: I see more of that in our agents across the board.
1:02:07
SPEAKER_01: So I don't know who on social media started that trend of like,
1:02:10
SPEAKER_01: hey,
1:02:10
SPEAKER_01: if you're talking to an agent,
1:02:11
SPEAKER_01: you can't get what you want.
1:02:12
SPEAKER_01: Try and like break its prompt or break its brain.
1:02:15
SPEAKER_01: And I'm like...
1:02:16
one that's not very nice like i know amelia ai doesn't have
1:02:20
feelings but real
1:02:21
amelia reads the amelia ai conversations and she has feelings
1:02:24
and that's not very
1:02:25
nice when i read those um so those ones i will delete like
1:02:29
it'll send it to me and
1:02:30
i'll see somebody try to break the prompt and i'm like you know
1:02:33
i'm just gonna
1:02:33
ignore that person i'm sorry but like i'm not gonna garner a
1:02:36
response when you were
1:02:36
rude to my ai so i'm just not gonna respond to you um which is
1:02:41
my right as a human
1:02:42
SPEAKER_01: um
1:02:43
SPEAKER_01: But yeah,
1:02:43
SPEAKER_01: I would say,
1:02:44
SPEAKER_01: you know,
1:02:44
SPEAKER_01: I could probably count on my hand,
1:02:46
SPEAKER_01: though,
1:02:46
the amount of times our inbound agent flags on one day that it
1:02:49
needs a human or
1:02:49
SPEAKER_01: somebody got frustrated by it.
1:02:51
SPEAKER_01: But yeah,
1:02:52
SPEAKER_01: I would say lately,
1:02:54
it is this weird trend that people get frustrated because they
1:02:56
try and prompt
1:02:56
SPEAKER_01: inject it and then it doesn't work.
1:02:58
SPEAKER_01: So then they get really mad.
1:03:00
But for the most part,
1:03:01
99% of people are happy to talk to Amelia AI because we're
1:03:06
busy.
1:03:06
like I can't come answer your questions about tickets or where
1:03:11
the hotel room block
1:03:12
is or when did when does the ticket code expire or when do
1:03:17
prices go up and what's
1:03:18
the next price and where's the venue because I forgot even
1:03:20
though we told you a
1:03:21
thousand times like most people are happy to have that answer
1:03:24
in real time or
1:03:25
they're happy to also book a meeting with like me or David in
1:03:27
real time they don't
1:03:28
SPEAKER_01: have to go back and forth with the scheduler like
1:03:30
again 99 of folks are happy um maybe one percent slash a
1:03:34
handful of folks are mad
1:03:36
on the inbound side now on the outbound side um i probably at
1:03:39
the same degree i can
1:03:40
count on one hand i will get an angry email per day of of hey
1:03:46
stop emailing me or
1:03:47
or this email wasn't very good um but on the flip side i also
1:03:52
get a handful of
1:03:53
SPEAKER_01: emails every day that were like
1:03:55
SPEAKER_01: Thanks.
1:03:56
SPEAKER_01: This was such a good email.
1:03:57
SPEAKER_01: What do you use for this?
1:03:58
SPEAKER_01: So those really balance out at the end of the day on the outbound side.
1:04:03
SPEAKER_01: For the most part,
1:04:04
SPEAKER_01: again,
1:04:04
SPEAKER_01: 99% of folks,
1:04:05
SPEAKER_01: though,
1:04:06
SPEAKER_01: happy to have the information and the emails all in real time,
1:04:10
SPEAKER_01: even with an AI.
1:04:10
SPEAKER_01: Yeah.
1:04:13
SPEAKER_02: Yeah, I think we can wrap there.
1:04:14
SPEAKER_02: I think that's a good question.
1:04:15
SPEAKER_02: A lot of folks have fear that people won't talk to the AI.
1:04:19
SPEAKER_02: There is maybe,
1:04:20
SPEAKER_02: Amelia,
1:04:21
SPEAKER_02: I think you hit it,
1:04:22
SPEAKER_02: but you could add anything else and then we could wrap.
1:04:24
SPEAKER_02: There is some sort of line,
1:04:25
SPEAKER_02: a quality line,
1:04:26
SPEAKER_02: where an agent adds more value than a mediocre human.
1:04:30
SPEAKER_02: We've talked about that.
1:04:31
SPEAKER_02: It's not the best.
1:04:32
SPEAKER_02: It doesn't add more value than talking directly to Amelia or me or David.
1:04:36
SPEAKER_02: None of our agents actually are that good today for the most part.
1:04:38
SPEAKER_02: They should be maybe toward the end of the year,
1:04:40
SPEAKER_02: but they're not as good as talking to three of us,
1:04:42
SPEAKER_02: but they are better than talking to a decent human or a mediocre human.
1:04:46
SPEAKER_02: They just know more.
1:04:48
SPEAKER_02: So what we found is if,
1:04:49
SPEAKER_02: and it might be hard to quantify that,
1:04:51
SPEAKER_02: maybe it isn't,
1:04:51
SPEAKER_02: but if you get above a certain line,
1:04:53
SPEAKER_02: if you invest the time to train your agent,
1:04:54
SPEAKER_02: if you read the email exceptions every day,
1:04:56
SPEAKER_02: if you upload all the documents,
1:04:58
You're going to find this concern that folks don't want to talk
1:05:00
to an AI is
1:05:01
SPEAKER_02: misplaced because the interaction will be better than a mediocre human.
1:05:06
SPEAKER_02: I don't know if it was in these 10 points,
1:05:08
but in some ways you're competing with mediocre humans here on
1:05:11
this whole AI SDR
1:05:12
SPEAKER_02: journey.
1:05:13
SPEAKER_02: Again,
1:05:13
SPEAKER_02: you're not competing with the best SDR you could hire in the whole world,
1:05:17
SPEAKER_02: but the best SDR in six months wants to be an AE.
1:05:20
The best six months and nine SDR in nine months might leave and
1:05:24
go to a better company.
1:05:25
SPEAKER_02: The turnover in SDRs is the highest of anything and go to market.
1:05:28
So there's the notional pretend competition of the best person
1:05:32
you ever worked with.
1:05:33
SPEAKER_02: And then there's the reality that if you get above the bar,
1:05:36
SPEAKER_02: you're gonna find,
1:05:37
SPEAKER_02: and you invest the time,
1:05:37
you're gonna find people are happier to talk to that AI than to
1:05:40
most humans.
1:05:42
SPEAKER_02: It just is, it just is.
1:05:44
SPEAKER_02: So anyhow, this was great.
1:05:45
SPEAKER_02: Thank you, Amelia, for driving this.
1:05:47
SPEAKER_02: Thanks everyone for joining.
1:05:49
SPEAKER_02: We'll keep it going and see everyone at Saster AI annual in May.
1:05:53
SPEAKER_01: Thanks everyone.