Latest / Contractor Business Builder Podcast / Stop Playing With ChatGPT, Use It Like This Instead For Contractors with Michelle Hamilton

Transcript

Stop Playing With ChatGPT, Use It Like This Instead For Contractors with Michelle Hamilton · Contractor Business Builder Podcast

0:03
Welcome to another exciting episode of the Contractor
0:05
Business Builder podcast. Today we're joined by Michelle
0:09
Hamilton, who spent 30 years in construction and now leads
0:11
global AI adoption at Answer Rocket.
0:14
She worked with enterprise companies to actually implement
0:17
AI into real workflows. If you've tried chat TBT, felt
0:22
overwhelmed, or wondered how AI actually fits into estimating
0:25
project management and construction, SOP's episode is
0:28
going to be for you to give you practical ways to start using it
0:32
immediately without breaking your business.
0:34
So Michelle Hamilton, AI Adoption Change management
0:37
practice lead at Answer Rocket. How you doing today?
0:40
Michelle, Derek, I'm doing great.
0:42
Thank you. I really appreciate you having
0:43
me. Perfect.
0:45
So definitely, I'm sure there's a lot of story there behind, you
0:48
know, 30 years in the architecture, engineering,
0:50
construction space and now, you know, moving into AI adoption.
0:55
What have you pulled from your experience and how how did that
0:58
transition come about into the leading the way with I?
1:02
You know, I have to tell you, I really lucked out in that I
1:06
received a beta copy of Chachi PT about six months before the
1:11
rest of the world had ever heard of it, and had about a six month
1:15
lead time to iterate actually back and forth with my son, who
1:19
was in college at the time. And we began working through
1:23
some of our own approaches, workflows, and iterations and
1:26
understanding what does natural language processing mean and how
1:31
can humans, everyday users, actually use technology as an
1:39
actual choreographer, Someone that is actually building on
1:44
their own technological use cases and not waiting for
1:47
someone else in technology to hand them their solutions.
1:52
And it immediately took off. As soon as everybody knows
1:56
Chachi, PT dropped it. It certainly became incredibly
2:00
enticing. By then I had already been
2:04
working with quite a few of my colleagues that were in the
2:09
construction space, architecture and engineering, development,
2:14
real estate and honestly just coaching friends and and
2:19
business owners privately and then eventually built out a
2:23
business practice behind it. Started my own AI companies
2:28
about two years ago and then this year Answer Rocket, which
2:33
is a global AI company. They've been building AI systems
2:37
since 2013. Found me and said this is one of
2:41
the hottest topics is understanding adoption and not
2:45
overwhelming people with new language.
2:47
So I opted to sell my companies and and join join them.
2:53
Wonderful. That's great.
2:55
So as far as you know, I think a lot of contractors have
2:58
definitely experimenting if not investing pretty heavily in AI
3:02
knowing how good the results and efficiencies can be.
3:06
But I think a lot are struggling to actually realize those
3:09
results or the efficiencies that you know are promised with with
3:13
AI. Where do you see the disconnect
3:15
or or what's the difference between those utilizing it
3:18
effectively and those that are just trying it but not getting
3:22
results? Actually think it begins with
3:26
truly understanding what you can ask AI to do alongside you and
3:31
not thinking of it as just hyper Google.
3:34
Not thinking of it as just a way to answer your e-mail, but
3:38
really actually utilizing it as a thinking partner.
3:43
Not to replace you, not to replace your employees, but to
3:47
allow them to create efficiencies in their work with
3:51
the information and the knowledge and expertise that
3:55
they already have. But now they have this
3:58
unbelievable personal assistant that can be working alongside
4:03
them doing research, beginning to build documentation,
4:07
understanding their clients, or even understanding new
4:12
permitting and, and codes that are are dropping and, and having
4:17
an ability to still do your job. But now you've got something
4:23
that you can have a conversation with and ask it to do some of
4:27
the research that might have taken you a day, a week, or a
4:32
month to do. Now you can get that done so
4:36
much more expeditiously. But where I'm finding the
4:40
biggest problem is, is that people feel like that they have
4:43
to understand a lot of IT language and don't truly
4:49
understand that. If you can explain what you're
4:52
trying to do and you can have a conversation with a colleague,
4:57
you already have the skills available to you to be an AI
5:01
super user because it wants you to iterate and talk through what
5:07
you are ultimately looking for it to partner with you on.
5:10
The less you tell it, the less it will perform.
5:14
The more you tell it, the more background information you give
5:17
it, the better it is going to be for you.
5:21
There's a lot of strategies that go into it as far as protecting
5:24
you, ensuring that everything is cited and correct and not
5:29
hallucinated, but just starting with the idea that you can
5:34
actually just simply talk to it. And I actually have most of my
5:38
clients use their AI in voice mode or in transcription mode
5:43
50% of the day so that they can just hit that button while
5:47
they're at a job site and start talking about what they're
5:50
working on and what they're going to need help with.
5:54
Yeah, that's great. Lots to unpack there.
5:57
And you mentioned a whole bunch of good stuff.
5:59
I I think the bit of a mindset shift as far as the AI is not
6:03
there to necessarily replace or fully take over jobs.
6:06
They're really there to help support people and make them
6:08
more knowledgeable, efficient. Essentially everyone now can
6:11
have an assistant and you can have assistance for all your
6:15
employees that just help them be better at, at what they're
6:18
doing. You've you've mentioned one
6:22
thing while there's a bunch of stuff there, but you mentioned
6:24
super user. What, what would you consider
6:26
like an AI super user? What is?
6:27
What is that that mean? Or what?
6:29
What's the? What's the difference there?
6:31
Sure. Somebody that what I would
6:33
consider to be a super user is first of all, someone that
6:36
understands the need to use multiple models that also are
6:41
building out workflows where the AI is acting on their behalf
6:47
while they're away from their desk and bringing back
6:51
reporting. Either through scheduled tasks,
6:54
through building out more advanced agents, or truly
6:58
understanding how to take often decades of data and turn that
7:04
into information that will allow you to understand how to act
7:09
moving forward regarding projects, project types, cost of
7:13
materials, supply chain, whatever that might be.
7:18
A super user is someone that reaches beyond just asking, you
7:24
know, I need you to help me write this e-mail.
7:27
It's someone that sees the possibilities of connecting
7:31
multiple tools together. Both they're invested third
7:36
party tools like a pro core, for instance, or some of their
7:40
foundational tools like a Microsoft or a Google product.
7:45
How are you connecting those tools together in order to lean
7:49
into the knowledge that you already have access to?
7:53
It feeds the engine and allows it to truly work far faster and
7:59
better to get you the results that you need.
8:04
Perfect. Thanks for clarifying what that
8:07
is. And it's it's interesting to me
8:11
that really the strong skill that organizations need more of
8:19
and has been known for so much as communication, right?
8:21
It's communication within your teams of people.
8:24
But now it's also communication with AI, as you mentioned, if
8:27
you provide inadequate information or you know, it's
8:31
getting better and better and it can into it a lot of different
8:33
things, But the better that you learn to communicate with AI,
8:37
the better, you know, output that you'll essentially get.
8:40
Now you've said AI is really a people problem.
8:42
It's it's not a technology problem.
8:44
Can you kind of elaborate or clean what you know?
8:48
I actually, one of the mistakes that I see a lot of owners or
8:52
key stakeholders making is assuming that if they invest in
8:56
AI, that the people will just figure out how to use it.
9:00
And my favorite analogy is if you had a job site and all of
9:07
your trades arrived or your Subs arrived and you didn't know if
9:13
they knew how to use the tools, perform the acts that you had
9:17
hired #4 and just said, I need you to build that building for
9:21
me. Well, they might be able to
9:23
coordinate themselves and build something.
9:26
Is it going to be a building? Is it going to be a structure
9:30
that is both safe, inhabitable? Is it going to be, you know, the
9:36
foundation going to be appropriately and properly
9:39
scoped out? No, because you didn't
9:42
understand the tools. You handed somebody a
9:45
screwdriver to do the job of a hammer because you just assumed
9:49
they knew what they were doing. An AI is exactly the same way.
9:53
For ownership. To assume that your employees
9:56
just know how to use it is oftentimes A costly mistake
10:02
because they will inadvertently put in information that does not
10:09
belong in there because you absolutely need to protect data
10:14
privacy. PII, HIPAA, you've got to be
10:18
thinking about all of those things, but also the ability to
10:23
teach everyone how to use this tool for their job and their
10:28
role to build the building or to build the RFP.
10:34
On it and let's let's say I have a construction company, you
10:37
know, we've got, I don't know, 20-50 people or whatever it may
10:40
be and we want to start utilizing AI in a bigger way.
10:46
How would you approach like training or getting people up to
10:50
speed on on AI? The way that we approach our
10:54
training is first we do meet with the key stakeholders to
10:57
understand what is the current tech stack, what are you
11:00
currently using, what have you currently invested in, and what
11:05
are some of your areas where you'd really like to see growth.
11:10
Then we begin to map out your most likely use cases in users.
11:16
We provide an AI101 training, if you will, to the entire team so
11:21
that everybody is leveled up simultaneously with use cases,
11:26
clear understanding of how to speak to the AI, acknowledgement
11:31
of what is the governance that an organization is put forth.
11:35
And then from there, I always recommend then we go department
11:39
by department. We talked to the estimators, we
11:42
talked to the site, Supers marketing, Bizdev, whatever it
11:46
might be. It could be finance, but really
11:49
then get hyper focused on what are their Cpis, what are their
11:55
deliverables, what is their job look like.
11:58
When I train a firm, they do not receive the same training as
12:02
anybody else. It is only focused on who they
12:06
are, their client type and their most likely work type so that we
12:11
can develop very customized delivery.
12:15
Now from there. The rest of my team then is also
12:18
talking to those owners to understand how are you utilizing
12:23
your data? How are you building out
12:26
internal predictive models to help you determine how to grow,
12:31
how to either expand or contract, how to even predict
12:36
shifts in the economy? And they're going to work
12:40
directly with those key data holders within an organization
12:45
to build out custom models for your company's data, only not
12:51
putting it out into an enterprise model, but building
12:54
something that is your company's proprietary data AI engine.
13:00
And my whole team can come in and serve the entire company, or
13:05
it can be as simple as some of the the training that I was
13:09
mentioning. But I will tell you in
13:11
construction, data does seem to be king.
13:15
And understanding how to move and how to predict oftentimes is
13:21
the measure of success. And having something that can
13:25
read literally thousands of data points simultaneously, infer
13:31
from those, and then begin to predict operational measures is
13:36
an astonishing use of the tool. Yeah, very, you know, powerful
13:43
abilities to sift through information and and extrapolate,
13:48
you know, useful ideas and feedback and things like that.
13:52
With construction, there's, there's so much communication,
13:55
there's so many plans, specification codes, emails,
13:59
just a lot going back and forth, lots of stakeholders,
14:04
regulations and, and so much more that being able to make
14:09
that make sense, which an AI can can help you is, is really
14:13
powerful. Now as far as you know, maybe
14:18
mistakes or or pitfalls to avoid or maybe things contractors
14:23
aren't seeing, what what do you see sort of commonly happening
14:27
as far as mistakes maybe by the business or even by individual
14:31
employees of the business using AI?
14:34
Believing what the AI tells you. If you #1 have not properly told
14:39
the AI what you're looking for and given it all of the
14:42
information that it is going to need to act on your behalf, a
14:47
big mistake is not remaining the human in the loop.
14:51
And AI makes everything sound unbelievably perfect.
14:57
And it is very easy to fall into the trap of saying, I don't
15:01
know, that all looks pretty good to me without reading every
15:05
single word. But you can also force AI to
15:09
cite its sources. Go show you a full bibliography
15:14
of where it got every single fact that it is sharing with
15:18
you. Almost show the math, if you
15:21
will. How did you come to this price
15:23
point? How did you help determine this
15:27
investment? You can ask to show its work and
15:31
hyperlink its own citations and results, but one of my favorite
15:37
tips and tricks is always ask for that.
15:41
But then take everything that it tells you and upload it to a
15:45
second model and ask that second model to prove the first model
15:50
wrong. Ask the second model to find all
15:53
of the mistakes, and then you've got 2 bits of information that
15:57
you're working with here. It doesn't mean you do not still
16:01
need to read every single word, but it does give you a head
16:05
start in finding and catching any of the hallucinations and
16:09
inconsistencies. I think the other mistake I'm
16:12
seeing is that organizations are not realizing that they have
16:17
built their reputation on their brand and on their voice and
16:22
that their clients have trusted them oftentimes for more than a,
16:26
you know, 100 years in some of my clients cases.
16:30
That brand and that voice needs to stay front and centre when
16:35
you are using AI and learning how to train AI models and how
16:41
to be a proper representative of your organization is a really
16:47
important lesson. So that it is reflective of how
16:51
you talk about your work, how you talk as an individual
16:55
contributor, and what is the language that has supported the
17:00
success of your company being around.
17:03
So that it aligns as an employee.
17:06
Not just as this AI that makes everything sound great might
17:11
have mistakes and is filled with hyperbole if there's no point.
17:16
Yeah, yeah, fantastic points and definitely encountered both of
17:21
those. Where on the 1st pass it often
17:24
sounds really good, but you really do have to kind of
17:27
question it sometimes if it's giving you specific things and
17:30
they're a little bit off. And then also having different
17:32
models that you can sort of play off each other does just really
17:35
make the output that much better, especially if you want
17:38
something refined and, and like double checked.
17:40
I, I know I've done that in, in a number of cases.
17:43
And you find certain models have certain strengths or
17:45
personalities, if you will, and are just better suited for, for
17:49
different things. But one thing with, with AI, I
17:52
know that I do more and more of now, especially if I want really
17:55
good output, is you've got to be a little bit hard on it and just
17:59
and ask it like, you know, 8 out of 10, is this the best that you
18:03
think this can be? Or how can we improve this?
18:05
Because the first go, it's really, it's a large language
18:09
model. It's going to use percentages of
18:11
what it expects to be the most relevant or active or the thing
18:14
you want to hear, but it may not actually be the best, you know,
18:19
whatever answer or output that that you need.
18:23
So asking it to, yeah, cited sources to improve upon itself.
18:29
What are you missing? How can you make the sound more
18:32
human? You know how if you were
18:35
considering this from my stakeholders viewpoint, how
18:38
would the output change? Like really diving into it and
18:42
asking it some kind of deeper questions.
18:43
We'll, we'll definitely set you on the right track.
18:46
And then what you said about your voice.
18:48
I know I'm guilty of this. Our company has as well where,
18:51
you know, we put our marketing materials or different things
18:53
and, and maybe a little quickly and it eventually ends up
18:56
starting to sound the same as a lot of the other stuff out
19:00
there. Even though we're inputting, you
19:01
know, our, our ideal clients and our, our brand messaging and
19:05
everything. You do have to work with it and
19:07
really refine that to make sure that you're, you're not losing
19:10
that, that human touch and things are coming off as just
19:13
another, you know, AI output or, or what have you.
19:15
So now as far as specific use cases or strategies for, you
19:21
know, contractors, what have you seen kind of make the biggest
19:25
improvements or or shifts inside an organization as as far as
19:29
specific areas that that AI is helping?
19:34
One of the most common use cases I get asked to demonstrate is
19:38
how do you answer an RFP, determine whether it's a go, no,
19:43
go, and then begin to fill that RFP out with known language and
19:50
documentation. And how do you take, let's say
19:53
this 60 page RFP and immediately find the top 10 things that you
19:59
would ask of every RFP's, key dates, key stakeholders,
20:04
deliveries, what kinds? Of permitting or we or
20:07
licensure, do we need to have you know, what are the
20:12
expectations being able to upload a 60 page document and
20:18
within you know 10 to 20 seconds now you can have a conversation
20:22
with that document and ask it. Tell me what you know.
20:26
Give me the top three key deadlines.
20:29
What are the deliverables? What do we need to provide in
20:33
both writing and electronically? Who are the key stakeholders?
20:38
Go out and do research on who the ultimate client is so that I
20:43
can be understanding a little bit more about what is important
20:47
to them. Why are they let's say building
20:50
this building? What do I need to know in order
20:54
to help my team perform better in winning or determining you
21:00
know what this RFP is not for us, Reading 60 pages takes a
21:05
long time. Finding that information takes
21:07
even longer. If you have a head start like
21:11
that. First of all, it's a huge time
21:13
saver. But then you can also ask that
21:16
RFP to go ahead and write emails to everybody internally in my
21:20
company asking for deliverables, project examples, data sets,
21:27
resumes, whatever that might be, all in the same conversation.
21:31
It can begin to build out communication internally for you
21:36
so that you can get that out to everybody on time.
21:40
You start getting that information brought in and then
21:44
at the same time you might upload a couple of example RFP's
21:48
or maybe you're working in a project folder where you've got
21:51
RFPs already loaded. Now you can say, I want you to
21:55
begin to fill this out for me with all of the information.
21:58
That's the same information we generally provide.
22:01
Tell me all of the detailed questions that I am really going
22:05
to need to finesse research and work with my team to make a
22:10
determination on. It cuts down the time of of
22:15
answering an RFP of two to four weeks 2.
22:20
It can be done easily in a week with the right prompts, the
22:24
right questions, and you save a lot of frustration if you
22:30
realize that it's a no go for you.
22:32
You don't have all of the correct licensure, you don't
22:35
have the correct team members, or it's just not your ideal
22:38
client profile. It sounds great, but it's not
22:41
for us. It ends up saving your team a
22:44
lot of time so that they can get about the business of moving on
22:49
to the next one or to another internal project.
22:54
My other favorite use cases for business development teams at
22:58
conferences or networking events where they are out really
23:03
seeking to make connections to build their opportunities and
23:09
pipeline using AI and mobile and telling it while you're in the
23:14
car or while you're at the airport.
23:15
Hey, I'm going to this conference, I'm going to this
23:18
networking event. Here are the people that I'm
23:21
looking for. Here's what I'm hoping to get
23:23
out of it. Setting it up so it knows what
23:25
you're doing. Then you're at the networking
23:28
event every time you talk to someone that you need to
23:32
remember to follow up with, and maybe you had a very interesting
23:35
conversation with them where they included maybe some
23:38
personal interesting details. You're going to step away for 20
23:43
seconds and just hit that same conversation, hit the transcribe
23:48
mode and say, well, I just met with Derek Isaac and he's got
23:51
this fantastic new podcast. And gosh, I have to tell you, he
23:55
went on this great vacation and I want to make sure I send him
23:58
an e-mail tomorrow saying how much I enjoyed it and that I
24:02
think we have some opportunities to really connect, offers an
24:06
opportunity of some sort. And you just do that throughout
24:10
the night or when you're in the car or at the airport on your
24:13
way home. Finish that thought out and have
24:17
it. Go ahead, start doing all the
24:19
research, drafting all the emails and getting the work
24:23
ready so that when you are back at your desk, all of that mobile
24:27
work has now been turned into actionable plans and drafts of
24:32
emails that you can get out so much more efficiently.
24:37
Wonderful. Yeah, great use cases.
24:41
And yeah, I hadn't thought about using it as a, you know, a note
24:45
taker so much at like a networking event or something
24:48
like that. But yeah, really great use cases
24:51
and and so much time savings. Now as far as the future goes, I
24:55
mean, AI improving so much and it continues to improve more and
24:59
more. I think the newer thing right
25:03
now is, you know, genetic AI and taking care of these tasks for
25:06
you automatically sometimes, you know, on your computer opening
25:09
programs and and doing things for you.
25:13
Is that something that companies are are rolling out or, or like
25:17
is there safety concerns there? Can you maybe just give us a
25:20
little bit of a snapshot of where you sort of see things
25:22
going more on the cutting edge of going more from a prompt
25:26
based, you know, back and forth to an actual agentic AI that's,
25:30
you know, getting tasks done and and doing.
25:33
I love that. I love that that actually is one
25:35
of answer rockets superpowers. They have an entire division
25:39
devoted to agentic AI and building out agentic workflows,
25:44
almost building out agent based employees for companies, but
25:49
they also focus very heavily in teaching and educating companies
25:55
and how to set agents up and have them working in a safe
25:59
environment. What files do you give it access
26:02
to? Where do you stop it from
26:04
working autonomously without checking in with you on every
26:08
single step that it takes? Not sending out emails on your
26:13
behalf. You should always review emails.
26:17
Make sure that it is both correct and the information is
26:21
there and that frankly it's in your voice.
26:24
But really understanding where does Agentic A I belong and
26:29
where doesn't belong and how can you be the orchestrator of that
26:34
so that you are truly standing in front of the orchestra and
26:38
saying now I need you to do this.
26:39
And then now you're gonna come in and then you're going to come
26:42
in and you're going to come in. It does take some coaching and
26:46
it does take, you know, first of all, experimenting with it and
26:51
having fun with it is great. You should be very careful and
26:55
know what you're doing so that it is not deleting files from
27:00
your system, which an agent can easily do if you have not
27:05
permissioned it properly and correctly.
27:08
So, you know, I'm fortunate that I have an entire team of, of you
27:12
know, agent experts, if you will, that help me with, with my
27:16
agents. But there's also some really
27:18
phenomenal tools out there. I happen to be a big cloud user
27:23
and use both Claude cowork and Claude code to build out my own
27:28
agentic workflows that frankly, I will tell you have made my
27:33
work life significantly better. And that when I arrive in the
27:37
morning at 5:00 AM, it has already looked at my calendar
27:41
for the day. It has already read all my
27:43
e-mail for the day. It has researched every person
27:45
I'm meeting with, given me a dossier on what we're going to
27:49
be talking about and what are they interested in.
27:53
It is then checking my Slack channel.
27:56
It is, you know, I mean, it is kind of covering the gamut so
28:01
that when I sit down at my desk, I have what I would call a chief
28:05
of staff handing me that set of paperwork and saying, here's
28:09
your day. I've done all the preparation
28:11
and research for you. What else do you need to know?
28:16
Perfect, Thanks for for clarifying that.
28:19
And again, these things are stuff you want to get right, but
28:25
also don't have an excuse of, you know, not trying some stuff
28:29
out. Start small, start in an
28:30
environment like you said, that's safe, where there's not
28:32
much access to different things and start to use it.
28:36
You know, our company is very involved with this as well as
28:39
far as like our internal workflows and our teams.
28:41
And it just it does make your life so much easier when these
28:45
different agentic AI's are set up or even just, you know,
28:47
having standard prompts and, and procedures documentation, it
28:52
makes everything flow a lot easier, a lot quicker.
28:54
So thanks so much for coming out here, Michelle, if you want to
28:58
learn more about answer Rocket, the work you do getting AI into
29:03
their business to increase efficiency and just solve
29:06
problems for them. What's the best way for them to
29:08
to get in touch with you? You can find me on LinkedIn.
29:12
It's Michelle Hamilton dot AI on LinkedIn and I post a lot.
29:18
I love conversations. I love meeting people like
29:20
yourself and, and really chatting back and forth.
29:23
And they can also visit answerrocket.com to learn more
29:28
about the work that we're doing. They are working with some of
29:31
the world's largest construction companies and providing them
29:35
everything that we've talked about today and so much more.
29:38
So it's it's such an exciting time.
29:41
Wonderful, so we'll put links in the description of this, you
29:44
know, audio video where every year you're watching or
29:46
listening in from go click through check out answer rocket.
29:49
Get in touch with Michelle here to learn more and get, you know,
29:52
AI working for you. Thanks so much, Michelle.
29:54
Thanks, Derek. Appreciate it.
29:55
Have a great day.