Latest / The STANDOUT Agency Podcast / EPISODE 6: AI Governance for Agencies
Transcript
- 0:04SPEAKER_00: Welcome to The STANDOUT Agency Podcast, for agency owners who want to grow, not just survive.
- 0:09SPEAKER_01: I'm Gareth Healey.
- 0:10SPEAKER_01: I'm an ex-agency owner and author of STANDOUT OR DIE.
- 0:14SPEAKER_01: I help agency founders see where they're still a bottleneck in their own business
- 0:17SPEAKER_01: and make sure they do something about it.
- 0:19SPEAKER_00: And I'm Callum Healey.
- 0:20SPEAKER_00: I find where agencies are losing hours to manual work and then build AI systems
- 0:24SPEAKER_00: that get the time back.
- 0:27SPEAKER_01: Each week, one of us interviews the other on a subject related to agency growth.
- 0:31SPEAKER_01: Everything we get into comes straight out of our own experience and client work.
- 0:34SPEAKER_01: Occasionally we bring in some guests to add their expert perspective on a topic.
- 0:40SPEAKER_00: Welcome back to The STANDOUT Agency Podcast,
- 0:42SPEAKER_00: Episode 6, AI Governance, How to Keep Your Agency Safe.
- 0:46SPEAKER_00: We've got to Episode 6 already.
- 0:49SPEAKER_00: Yeah, absolutely.
- 0:49SPEAKER_00: It's flown, hasn't it?
- 0:50SPEAKER_00: Time is flying by, but I'm all here for it.
- 0:52SPEAKER_00: We've got a routine now.
- 0:54SPEAKER_00: We've had so many lessons we've learned along the way,
- 0:56SPEAKER_00: or mandatory checks of audio, visuals, all sorts.
- 0:59SPEAKER_01: Don't jinx it.
- 1:00SPEAKER_00: Yeah, well, that's the thing.
- 1:02SPEAKER_00: Hopefully, everything is sounding and looking fine for our listeners here today.
- 1:07SPEAKER_01: So, Callum, AI governance, how to keep your agency safe.
- 1:11SPEAKER_01: Interesting topic you've chosen today, of course, very relevant.
- 1:16SPEAKER_01: I think perhaps we should start with AI governance.
- 1:19SPEAKER_01: What's your understanding?
- 1:20SPEAKER_01: What do you actually mean when we say that?
- 1:22SPEAKER_00: In general, it's safety.
- 1:25SPEAKER_00: AI, with its power, there is great responsibility that is needed.
- 1:30SPEAKER_00: I've become somewhat of a data protection specialist alongside an AI consultant
- 1:36SPEAKER_00: because it is ultimately the first thing we always look at.
- 1:40SPEAKER_00: It's simply how do we use AI to benefit agencies while mitigating as much risk
- 1:49SPEAKER_00: as possible,
- 1:50SPEAKER_00: hopefully it's no risk at all, making it a safe space for Cool,
- 1:56SPEAKER_01: sounds good.
- 1:57SPEAKER_01: And well,
- 1:58SPEAKER_01: I guess with that summary,
- 2:02SPEAKER_01: what are the things that you're seeing at the moment when it comes
- 2:04SPEAKER_01: to agency businesses in particular?
- 2:07SPEAKER_01: What would you say the current state of the market is with regard to governance?
- 2:12SPEAKER_01: Are we well covered?
- 2:14SPEAKER_01: Are we not covered at all?
- 2:16SPEAKER_01: Is it somewhere in the middle, I guess?
- 2:17SPEAKER_00: Yeah, well, I've been doing this for about a year now.
- 2:20SPEAKER_00: So I would say that Definitely towards the start,
- 2:24SPEAKER_00: and it still occurs, Shadow AI is a massive deal.
- 2:29SPEAKER_00: And I mentioned the year gap because it's such an evolving landscape.
- 2:35SPEAKER_00: Agencies are becoming more and more familiar with ways of protecting.
- 2:39SPEAKER_00: Ultimately it's an education thing to start off with.
- 2:42SPEAKER_00: And what I mean by shadow AI is people in agency teams or any business that use
- 2:49SPEAKER_00: the tools in the background without leadership knowing.
- 2:53SPEAKER_00: And this can have some quite horrible implications.
- 2:56SPEAKER_00: Of course, a lot of these tools that people are using are free plans.
- 3:01SPEAKER_00: Personal plans aren't as bad,
- 3:03SPEAKER_00: but in general, I'd say free and personal, both of them should be abolished.
- 3:08SPEAKER_00: For use in business, of course.
- 3:09SPEAKER_00: For use in business, yeah.
- 3:11SPEAKER_01: For people bringing in or installing or using on their company laptops
- 3:17SPEAKER_01: for company work,
- 3:19SPEAKER_01: asking questions of their own AI platform that they've subscribed to personally.
- 3:25SPEAKER_00: Exactly, yes.
- 3:26SPEAKER_00: And the problem,
- 3:28SPEAKER_00: especially with free plans,
- 3:29SPEAKER_00: that there's that old saying, if the product is free, then you are the product.
- 3:35SPEAKER_00: AI companies get a lot of value from using the data that people give it in chat histories.
- 3:42SPEAKER_00: If you don't explicitly turn off things like model training or you're not
- 3:46SPEAKER_00: on commercial plans like some of these,
- 3:50SPEAKER_00: for example,
- 3:51SPEAKER_00: team and enterprise plans that AI subscriptions,
- 3:54SPEAKER_00: these are tiers that they offer, they will take that data and help it.
- 3:59SPEAKER_00: They can help improve their models and When there isn't any UK GDPR
- 4:04SPEAKER_00: or data protection agreements and these free plans,
- 4:08SPEAKER_00: they can theoretically do what they want with the data.
- 4:10SPEAKER_00: So it's very important to protect your private, either client data, things like regulated data.
- 4:18SPEAKER_00: And I've heard horror stories.
- 4:20SPEAKER_00: I spoke to an AI lawyer or one who is now deviating, specializing in AI.
- 4:26SPEAKER_00: He told me of a story where A agency was hired for a marketing campaign for such
- 4:32SPEAKER_00: a business and this business knew that they only gave this data
- 4:39SPEAKER_00: for their new campaign to the agency.
- 4:42SPEAKER_00: Someone somewhere found that data in AI chats and so they knew,
- 4:48SPEAKER_00: the business knew that the marketing agency was responsible for that data breach
- 4:52SPEAKER_00: and got hit with a massive insurance claim and The conclusion was
- 4:57SPEAKER_00: that someone was using a free plan and uploaded these sensitive documents in and
- 5:01SPEAKER_00: then it siphoned the data.
- 5:03SPEAKER_00: Wow.
- 5:04SPEAKER_01: So, I mean,
- 5:05SPEAKER_01: that's a horror story in itself,
- 5:06SPEAKER_01: but I guess the background or what's sat behind that is the principle,
- 5:12SPEAKER_01: isn't it,
- 5:12SPEAKER_01: of if people are employees,
- 5:15SPEAKER_01: for want of a better word,
- 5:16SPEAKER_01: using their own personal plans within a business,
- 5:19SPEAKER_01: an agency in our context,
- 5:22SPEAKER_01: they might be doing that with the best will in the world, might they?
- 5:26SPEAKER_01: To help them in their job,
- 5:27SPEAKER_01: to help the client, to help the business, just to get the job done, etc.
- 5:31SPEAKER_01: But of course,
- 5:32SPEAKER_01: if something goes wrong,
- 5:33SPEAKER_01: then the client company in this case,
- 5:36SPEAKER_01: or somebody that's grieved or impacted by it,
- 5:39SPEAKER_01: won't have any issue whether it's why the personal plan is used.
- 5:46SPEAKER_01: They'll come to the door of the company, the agency, won't they?
- 5:50SPEAKER_01: Ununderstandably, rightly so.
- 5:53SPEAKER_01: Well it was a personal plan, that's absolutely no get out for the business at all.
- 5:58SPEAKER_00: Yeah and people can have great incentives,
- 6:02SPEAKER_00: moral incentives,
- 6:04SPEAKER_00: but ultimately actions and outcomes are reality,
- 6:07SPEAKER_00: those are the things that happen and in this case it's purely the agency's fault
- 6:13SPEAKER_00: if something were to go wrong so that's why it's so important to cover governance so
- 6:16SPEAKER_00: that people can innovate and experiment and adopt AI in a way
- 6:20SPEAKER_00: that It doesn't put them at any risk and a simple fix for the shadow AI as
- 6:27SPEAKER_00: a problem is to get on a company plan in all AI tools.
- 6:32SPEAKER_00: That's almost the first thing I tell every agency.
- 6:35SPEAKER_00: Abolish all free and personal plans.
- 6:38SPEAKER_00: This is particularly good because we discussed context in one of our episodes in
- 6:43SPEAKER_00: the past and It allows you to have a shared environment,
- 6:46SPEAKER_00: which has some of these shared workflows and context,
- 6:49SPEAKER_00: but in relation to AI governance,
- 6:53SPEAKER_00: you can actually claim and comply with UK GDPR with your clients because A lot
- 7:01SPEAKER_00: of these team and enterprise plans on things like Gemini,
- 7:04SPEAKER_00: ChatGBT,
- 7:04SPEAKER_00: Claude, they explicitly state, yes, for commercial plans, we comply with UK GDPR.
- 7:11SPEAKER_00: You can then update your client MSAs to reflect that.
- 7:14SPEAKER_00: We use this tool on this plan.
- 7:16SPEAKER_00: It has all of these certain data obligations.
- 7:20SPEAKER_00: It's all protected.
- 7:21SPEAKER_00: As long as you consent to that, everyone's happy.
- 7:24SPEAKER_01: Perfect.
- 7:25SPEAKER_01: So use personal plans within a business.
- 7:29SPEAKER_01: Huge risk.
- 7:29SPEAKER_01: You need to absolutely put out in your business a policy and communication to state
- 7:38SPEAKER_01: that that's not acceptable.
- 7:40SPEAKER_01: Do not use personal plans.
- 7:42SPEAKER_01: Only use teams plans, company plans that we pay for.
- 7:46SPEAKER_01: Makes perfect sense.
- 7:47SPEAKER_01: That's one of the risks, isn't it?
- 7:48SPEAKER_01: That's a risk that can be mitigated by doing that.
- 7:52SPEAKER_01: There's a cost to that, of course, isn't there, for the company?
- 7:58SPEAKER_00: That's the issue, of course.
- 8:00SPEAKER_00: It's a case where if people are using AI in an ad hoc way,
- 8:06SPEAKER_00: AI itself isn't directly contributing to certain workflows that an agency has
- 8:14SPEAKER_00: and charges.
- 8:16SPEAKER_00: The clients sort of charge them for,
- 8:18SPEAKER_00: but ultimately it's needed,
- 8:22SPEAKER_00: of course, to be safe before any of this experimentation begins.
- 8:26SPEAKER_00: And rightfully so,
- 8:29SPEAKER_00: from what you said there,
- 8:30SPEAKER_00: the teams and enterprise plans, they require a minimum amount of spend.
- 8:35SPEAKER_00: So consumer plans, most of the time they're around 15 to 30 pounds a month.
- 8:41SPEAKER_00: This is the same for consumer plans for the most part.
- 8:44SPEAKER_00: Some are higher tiers because they allow you slightly more usage.
- 8:47SPEAKER_00: For example, Claude, which we love, which has team and enterprise.
- 8:53SPEAKER_00: But recently,
- 8:54SPEAKER_00: Claude has allowed a minimum of two users,
- 8:56SPEAKER_00: which is great because instead of buying everyone in your agency a seat,
- 9:01SPEAKER_00: you can simply buy two to start off with,
- 9:04SPEAKER_00: start getting used to the tool,
- 9:06SPEAKER_00: maybe creating some workflows,
- 9:07SPEAKER_00: and then upgrade it over time as you start to see more of the value.
- 9:11SPEAKER_01: Sure.
- 9:12SPEAKER_01: Makes sense.
- 9:12SPEAKER_01: And I guess ultimately it's about the cost is false economy, isn't it?
- 9:17SPEAKER_01: To not pay this higher commercial seat plan and hope that nothing goes wrong.
- 9:24SPEAKER_01: Because if something goes wrong, this could be, well, not just the money, is it?
- 9:29SPEAKER_01: The risk here is reputation of the agency.
- 9:31SPEAKER_01: The risk here is the agency's future, really.
- 9:36SPEAKER_01: If there was even a leak which somebody had no intention of If it was sensitive data
- 9:43SPEAKER_01: of any kind and the client found out or something caused an issue,
- 9:46SPEAKER_01: it could be a massive issue for the company,
- 9:48SPEAKER_01: could absolutely decimate or even close the company down.
- 9:52SPEAKER_00: And if that's in the news and it's a big scandal,
- 9:55SPEAKER_00: businesses will be less likely to sign up that marketing agency,
- 10:00SPEAKER_00: so it's very bad for reputation.
- 10:03SPEAKER_01: So, avoiding shadow AI seems to be an absolute no-brainer.
- 10:09SPEAKER_01: That's very important, of course, but it's not the only aspect of AI governance.
- 10:14SPEAKER_01: This is a wider topic that includes
- 10:17SPEAKER_00: more things.
- 10:20SPEAKER_01: Talking to lots of agencies,
- 10:24SPEAKER_01: when you see an agency,
- 10:25SPEAKER_01: which I know you've had a couple of examples that certainly as things stand,
- 10:29SPEAKER_01: that you feel like they've nailed AI governance and are really set up really well.
- 10:34SPEAKER_01: What are the sort of things you've seen?
- 10:36SPEAKER_00: Yeah, so I was auditing an agency in Eastern Europe,
- 10:41SPEAKER_00: a team of around 60 people,
- 10:43SPEAKER_00: and they had a great structure and architecture for AI safety.
- 10:48SPEAKER_00: They'd invested a thousand hours,
- 10:50SPEAKER_00: in their own words, on AI,
- 10:52SPEAKER_00: they'd had an enterprise Gemini plan,
- 10:55SPEAKER_00: model training completely turned off,
- 10:56SPEAKER_00: contracts were updated,
- 10:58SPEAKER_00: they even had a approval policy where If team members found a new tool,
- 11:03SPEAKER_00: they'd have to go through a certain process,
- 11:06SPEAKER_00: policies to get these new tools and make sure they're on the right plans
- 11:09SPEAKER_00: and whatnot.
- 11:11SPEAKER_00: All useful tips for other people here.
- 11:13SPEAKER_00: Completely, yes.
- 11:14SPEAKER_00: So great architecture,
- 11:15SPEAKER_00: but that isn't the only,
- 11:17SPEAKER_00: it's one part of a multifaceted kind of system and way of,
- 11:25SPEAKER_00: I think it's compliance ultimately is the next thing.
- 11:30SPEAKER_00: People can have the architecture,
- 11:31SPEAKER_00: but in this agency,
- 11:32SPEAKER_00: people weren't actually using the Gemini plan, they were just still on ChatGBT.
- 11:37SPEAKER_00: And it was found that around two-thirds of the team didn't have any usage on Gemini
- 11:44SPEAKER_00: at all when tracking that through the analytics data.
- 11:48SPEAKER_00: So yes,
- 11:48SPEAKER_00: just because you set up the right system doesn't mean people are going to have
- 11:53SPEAKER_00: that accountability.
- 11:54SPEAKER_00: And so it's very much a behavioral aspect to this.
- 12:00SPEAKER_00: And I think the main kind of lesson here is that People need to be educated
- 12:06SPEAKER_00: on why it's so important to comply with the structure that the business sets up.
- 12:12SPEAKER_00: Of course,
- 12:12SPEAKER_00: the leadership team,
- 12:14SPEAKER_00: upon creating that system and all the policies and the right company plans,
- 12:19SPEAKER_00: know why they're doing it.
- 12:20SPEAKER_00: It's for data protection.
- 12:21SPEAKER_00: And you mentioned shadow AI there.
- 12:24SPEAKER_00: It's only one part of the equation, but ultimately it's data.
- 12:28SPEAKER_00: It's protecting their data, whether that's their own or their clients.
- 12:33SPEAKER_00: People need to be aware of just how bad this can get and I would take
- 12:39SPEAKER_00: the quite blunt and aggressive claim that it could be a fireable offence
- 12:45SPEAKER_00: if people weren't to comply with some of these regulations that are set because
- 12:50SPEAKER_00: if they don't get fired by the leadership team's own will,
- 12:56SPEAKER_00: they'll get fired automatically through the company getting hit with
- 13:00SPEAKER_00: a massive insurance claim that costs the business a load of money.
- 13:03SPEAKER_01: Everybody might lose their jobs.
- 13:04SPEAKER_00: And then they have to let people off anyway.
- 13:07SPEAKER_00: Because that's just how bad it can get if some of this data isn't controlled.
- 13:11SPEAKER_01: I think that's really important.
- 13:12SPEAKER_01: It does sound harsh,
- 13:13SPEAKER_01: but if you need a policy of no use of personal plans on company business,
- 13:21SPEAKER_01: AI personal plans,
- 13:23SPEAKER_01: that should be a A gross misconduct offence if you do use that
- 13:31SPEAKER_01: because data can leak
- 13:34SPEAKER_01: and that can become a massive problem.
- 13:36SPEAKER_01: So I think it does sound harsh but it's really important and I would stand by that.
- 13:41SPEAKER_01: And also,
- 13:41SPEAKER_01: what you've just touched upon there,
- 13:44SPEAKER_01: it's alright having these policies in place,
- 13:45SPEAKER_01: but sometimes communication,
- 13:47SPEAKER_01: the art of communication is giving people context, isn't it?
- 13:50SPEAKER_01: Just as it is prompting AI chatbots.
- 13:53SPEAKER_01: And if you don't explain to people Why that's important,
- 13:57SPEAKER_01: why you're demanding in a way people don't use their own personal plans,
- 14:05SPEAKER_01: what could go wrong, what you are mitigating as a risk here, how big it could get.
- 14:10SPEAKER_01: People might just naturally say, well, it's not such a big deal.
- 14:12SPEAKER_01: I've got my personal plan on my phone.
- 14:15SPEAKER_01: I'm at home.
- 14:16SPEAKER_01: I've just got out of the shower.
- 14:17SPEAKER_01: I've just thought of something about work that I could quickly research and I'll put
- 14:23SPEAKER_01: a document in or whatever it might be.
- 14:24SPEAKER_01: And suddenly you've contravened that policy.
- 14:27SPEAKER_01: So people,
- 14:28SPEAKER_01: if they're not thinking it's just something the leadership team wants to do,
- 14:32SPEAKER_01: they want us to use their own plans.
- 14:34SPEAKER_01: No, it's much bigger.
- 14:36SPEAKER_00: Yeah and I guess an analogy for this,
- 14:38SPEAKER_00: we have a Whippet,
- 14:39SPEAKER_00: he's a one year old and we're trying to teach him all of these different commands
- 14:45SPEAKER_00: and Whippets have a high prey drive,
- 14:47SPEAKER_00: we don't want him to run off and get lost or get hit by a car in the road.
- 14:51SPEAKER_00: So we do our best to say these commands, just here, come back, you know, we've got a treat.
- 14:56SPEAKER_00: It could be even another dog that's getting aggressive or the owner isn't friendly.
- 15:01SPEAKER_00: So many things can go wrong if Billy wasn't to come back to us.
- 15:05SPEAKER_00: Unfortunately, we can't teach him why.
- 15:07SPEAKER_00: He wouldn't understand why.
- 15:09SPEAKER_00: But your team members can understand why.
- 15:11SPEAKER_00: And as long as, instead of...
- 15:13SPEAKER_00: Well he knows that he should come back but when he's playing with other dogs
- 15:17SPEAKER_00: and he's sort of tunnel vision he just doesn't do it because he doesn't understand
- 15:20SPEAKER_00: but your team members could understand I think that framing we said before about
- 15:24SPEAKER_00: the fact that The company could lose a hell of a lot of money and then have
- 15:29SPEAKER_00: to let people off.
- 15:30SPEAKER_00: That is a crazy but realistic outcome.
- 15:34SPEAKER_00: And so if people understand just how severe it is, it would help them comply.
- 15:38SPEAKER_00: And I think you can actually track this.
- 15:41SPEAKER_00: Well, I don't think, I know that you can track this objectively in some of these AI plans.
- 15:45SPEAKER_00: If it's an enterprise plan,
- 15:47SPEAKER_00: you can look at each individual user on a single sign-on
- 15:51SPEAKER_00: and see how much usage they have.
- 15:53SPEAKER_00: Sometimes this doesn't tell you the whole story
- 15:55SPEAKER_00: because you might simply have no workflows in AI yet,
- 15:59SPEAKER_00: but I think that's almost the next step.
- 16:02SPEAKER_00: If you make it clear that in this certain process everyone should be using AI,
- 16:07SPEAKER_00: if it isn't coming up that they aren't using AI for that process,
- 16:11SPEAKER_00: perhaps it's a package system like a skill or a GPT we've talked about
- 16:15SPEAKER_00: in previous episodes.
- 16:17SPEAKER_00: They're either not using the company plan,
- 16:19SPEAKER_00: which is a red flag,
- 16:21SPEAKER_00: or they're not using AI at all,
- 16:23SPEAKER_00: which is a shame as well because these systems are created for efficiency.
- 16:27SPEAKER_01: And we want people to be adopting this stuff where they can and where is necessary, don't we?
- 16:32SPEAKER_01: Yeah, that's a good point.
- 16:34SPEAKER_01: So don't treat people like whippets.
- 16:37SPEAKER_01: Exactly.
- 16:37SPEAKER_01: Give them context is the message
- 16:40SPEAKER_00: from this podcast.
- 16:42SPEAKER_00: Completely.
- 16:43SPEAKER_00: Yeah, make rules but make people understand why the rules exist,
- 16:47SPEAKER_00: which a Whippet couldn't understand.
- 16:49SPEAKER_01: Which is good.
- 16:50SPEAKER_01: So having the right setup is important,
- 16:53SPEAKER_01: having the right rules and policies is important,
- 16:56SPEAKER_01: the structure, instilling the right behaviours in people.
- 17:01SPEAKER_01: What else in terms of AI deployment for agencies,
- 17:05SPEAKER_01: as we say, what else do people running agencies need to be wary of?
- 17:08SPEAKER_00: Well, you'd think that there would be nothing left.
- 17:11SPEAKER_00: If it's all about these,
- 17:13SPEAKER_00: if the system's correct and it's well configured and the
- 17:15SPEAKER_01: behavior's there,
- 17:16SPEAKER_01: what else is there?
- 17:17SPEAKER_00: Unfortunately,
- 17:19SPEAKER_00: there are millions of different softwares and to get the most value out of AI,
- 17:26SPEAKER_00: you want all of the data and all of these different softwares to interact
- 17:29SPEAKER_00: with each other and share that data so AI has more context.
- 17:35SPEAKER_00: If you want to know the importance of context,
- 17:37SPEAKER_00: check out one of our episodes in the past, episode four, I believe.
- 17:43SPEAKER_00: Just because the AI itself has a governed structure doesn't mean that
- 17:49SPEAKER_00: the connector is governed.
- 17:52SPEAKER_00: So let's take Office 365 as an example.
- 17:56SPEAKER_00: I did this this week with an agency and we were setting up the connector
- 18:03SPEAKER_00: as we needed to access certain spreadsheets to analyse it for workflows
- 18:07SPEAKER_00: that we were building.
- 18:09SPEAKER_01: This is using an AI LLM
- 18:13SPEAKER_00: and connecting it to their Office.
- 18:15SPEAKER_00: Of course.
- 18:15SPEAKER_00: Yeah, so a connector is,
- 18:18SPEAKER_00: we talked about all these different softwares,
- 18:21SPEAKER_00: turning a connector on would allow Claude,
- 18:23SPEAKER_00: in this case, to communicate with Office 365.
- 18:27SPEAKER_00: It can then view all of the files in there, and that is part of the problem.
- 18:32SPEAKER_00: So we wanted certain Excel spreadsheets as part of this workflow,
- 18:37SPEAKER_00: but The connectors as they sit are almost an all or nothing situation.
- 18:44SPEAKER_00: You turn it on,
- 18:45SPEAKER_00: you log in,
- 18:47SPEAKER_00: and it's got access to everything,
- 18:48SPEAKER_00: which we don't want for many different reasons, as we'll get into.
- 18:54SPEAKER_00: Instead of the connector being malleable, which...
- 19:00SPEAKER_00: Unfortunately, it's very rigid.
- 19:01SPEAKER_00: Like I said, it's all or nothing.
- 19:03SPEAKER_00: It's all about the login that you use.
- 19:06SPEAKER_00: So the idea and what we did and what has worked in the past for me is that
- 19:11SPEAKER_00: the IT provider for your agency, we would invite them to the meeting.
- 19:16SPEAKER_00: And it's all about configuring a new user.
- 19:20SPEAKER_00: And this is almost the Claude user.
- 19:21SPEAKER_00: And the rule of thumb is whatever that user can see, Claude has access to.
- 19:26SPEAKER_00: So I think it's always good practice to start bottom up where you give it access
- 19:32SPEAKER_00: to nothing but the things that you need and then
- 19:35SPEAKER_00: over time you can gradually introduce more files as it's proven that it's all safe
- 19:40SPEAKER_00: and scoped correctly.
- 19:42SPEAKER_00: And so this isn't just a case of Let's say client data that you have in Office 365.
- 19:52SPEAKER_00: I give Office 365 as an example,
- 19:54SPEAKER_00: Google Workspace works as well because this is where all
- 19:57SPEAKER_00: of your sensitive data is held and combined with data that's very useful.
- 20:01SPEAKER_00: So it's all about making that right scope.
- 20:05SPEAKER_00: So of course, client data, we need that protected.
- 20:08SPEAKER_00: And unless they have given you confirmation of the fact
- 20:11SPEAKER_00: that they allow third party processing,
- 20:14SPEAKER_00: regardless of UK GDPR on some of these enterprise plans that we talked about,
- 20:18SPEAKER_00: if your clients say they don't want it, then you can't do it.
- 20:20SPEAKER_00: So it's worth checking all of those.
- 20:22SPEAKER_00: So in general, we keep that off limits until the client complies with the agency.
- 20:28SPEAKER_00: But it's also about the internal data that you have,
- 20:31SPEAKER_00: because Someone,
- 20:33SPEAKER_00: rightfully so,
- 20:34SPEAKER_00: looks at a connector,
- 20:35SPEAKER_00: figures out what one is and goes,
- 20:36SPEAKER_00: yeah, we want Claw to access certain files for workflows that we need.
- 20:41SPEAKER_00: If it's the founder of an agency who would have the master login for one of these,
- 20:48SPEAKER_00: if it's Office 365 or Google Workspace,
- 20:50SPEAKER_00: if they log in using their login,
- 20:53SPEAKER_00: everyone on that Claw Teams plan or any AI enterprise plan would see everything that
- 21:00SPEAKER_00: the founder could.
- 21:02SPEAKER_01: Which is financial data for the company, HR, employee data.
- 21:09SPEAKER_00: Perhaps even confidential transcripts of meetings that are held between leadership.
- 21:14SPEAKER_00: And so we don't want to give the keys to everyone in that respect.
- 21:19SPEAKER_00: It's much safer and better to scope the connector for certain things,
- 21:26SPEAKER_00: which is why configuring a separate system is so valuable.
- 21:30SPEAKER_01: So setting up connectors and having the correct login priority levels.
- 21:39SPEAKER_01: There's a better term for that, priority levels, isn't there, in software?
- 21:43SPEAKER_01: Of course.
- 21:45SPEAKER_01: So the MD can see everything,
- 21:47SPEAKER_01: people further down the chain might not be able to see everything to varying levels
- 21:51SPEAKER_01: of degrees.
- 21:53SPEAKER_01: That's really important, isn't it?
- 21:55SPEAKER_00: Yeah, definitely.
- 21:56SPEAKER_01: You mentioned earlier, actually, there's client data, internal data.
- 22:02SPEAKER_01: There's also potentially,
- 22:04SPEAKER_01: many agencies have some degree of client conflict,
- 22:08SPEAKER_01: confidentiality, working with clients in a similar space.
- 22:10SPEAKER_01: So again,
- 22:11SPEAKER_01: you might need to specify that clients often are not very happy with that
- 22:16SPEAKER_01: and some are completely unhappy with it and won't work with you
- 22:19SPEAKER_01: if there's any conflict of interest.
- 22:21SPEAKER_01: I've worked in agencies for over 30 years
- 22:23SPEAKER_01: and I don't know anybody that's ever had any time to sit down and compare notes
- 22:29SPEAKER_01: on different clients.
- 22:30SPEAKER_01: It's more procedural, isn't it?
- 22:31SPEAKER_01: But an AI doesn't need that time.
- 22:34SPEAKER_01: Good point.
- 22:36SPEAKER_01: If an AI can access Client A and Client B and there's a degree of conflict
- 22:40SPEAKER_01: of interest amongst them or that they shouldn't be talking to each other,
- 22:43SPEAKER_01: then an AI won't discriminate or won't morally take a step back and not do that,
- 22:48SPEAKER_01: will it?
- 22:49SPEAKER_01: It'll just source the information.
- 22:50SPEAKER_01: So logins for people working on client A, people working on client B are hugely important.
- 22:55SPEAKER_01: The internal side of it, we both know an instance of it wasn't somebody we were working with.
- 23:06SPEAKER_01: We know an instance of an agency that failed to lock down their Google Workspace in
- 23:16SPEAKER_01: as much as they should have done,
- 23:17SPEAKER_01: opened it up to AI and one of their employees who they had an issue with,
- 23:22SPEAKER_01: there was a The individual who was the employee going to the tribunal for
- 23:28SPEAKER_01: a certain reason
- 23:44SPEAKER_01: Did a search of the Google Workspace for information about their work,
- 23:50SPEAKER_01: connected to it,
- 23:51SPEAKER_01: but they found some very incriminating text that the management of
- 23:57SPEAKER_01: that company had said about them.
- 24:00SPEAKER_01: So it became a very unnecessary and very embarrassing and legal issue because
- 24:06SPEAKER_01: the employee had found something else that they used That was exhibit A
- 24:14SPEAKER_01: of what we said about Scope Connectors there.
- 24:17SPEAKER_00: This company were on a free Gemini plan.
- 24:21SPEAKER_00: This certain employee typed their name in and all the data came up about their name
- 24:26SPEAKER_00: because Gemini transcript and track meetings,
- 24:29SPEAKER_00: emails, which is a complete nightmare for any of the agency founders listening.
- 24:34SPEAKER_01: They were looking at that stage for their work or something to do with their work.
- 24:37SPEAKER_01: But found some transcripts of meetings that they weren't in and the content
- 24:42SPEAKER_01: of those meetings wasn't very favourable to them.
- 24:45SPEAKER_01: So obviously they copied and pasted it and saved some of that information
- 24:48SPEAKER_01: and used it as ammunition in that process.
- 24:51SPEAKER_01: So hopefully we've explained the danger.
- 24:53SPEAKER_01: I'm just using that to give context for explaining the dangers.
- 24:56SPEAKER_01: Yeah, completely.
- 24:56SPEAKER_01: Without divulging any sense of information on the part of that business,
- 25:00SPEAKER_01: of course,
- 25:00SPEAKER_01: and that individual,
- 25:01SPEAKER_01: but that was a real example that happened that we know of,
- 25:04SPEAKER_01: and that's another danger, isn't it?
- 25:07SPEAKER_00: Yeah, and the sheer abundance of connectors,
- 25:10SPEAKER_00: if you look on any of these AI platforms and you go in the connectors tab,
- 25:15SPEAKER_00: you will see that there are connectors for almost anything.
- 25:18SPEAKER_00: I look and there seems to be more and more each time.
- 25:22SPEAKER_00: It grows.
- 25:22SPEAKER_00: And so it's not just one connector that you need to make sure it's safe.
- 25:26SPEAKER_00: Every time you configure a new one, you have to make sure that it's scoped correctly.
- 25:30SPEAKER_00: That should be a policy.
- 25:32SPEAKER_00: I think that's a core lesson.
- 25:34SPEAKER_00: Every time a team member wants to use a connector,
- 25:37SPEAKER_00: because it's very easy to do,
- 25:39SPEAKER_00: even on these team plans,
- 25:41SPEAKER_00: someone could easily just click yes and then log in,
- 25:45SPEAKER_00: even if it's not necessarily a founder,
- 25:46SPEAKER_00: someone who has access to a lot of files that maybe everyone shouldn't be having
- 25:51SPEAKER_00: or be able to see, then it would be a nightmare.
- 25:54SPEAKER_00: So it's one of those that connectors The thing to bring to leadership or the AI lead
- 26:02SPEAKER_00: and the tech lead to make sure that it's all okay and governed.
- 26:07SPEAKER_00: There are certain things that cannot be controlled unfortunately and that is
- 26:11SPEAKER_00: the ability for someone to export a file.
- 26:16SPEAKER_00: If someone has access to data that the scope connectors don't have access to,
- 26:20SPEAKER_00: they could simply export the file,
- 26:23SPEAKER_00: drag it into the chat,
- 26:24SPEAKER_00: and now Claude or ChatGPT has access to that data, the connector scope.
- 26:30SPEAKER_01: All of that goes out the window.
- 26:31SPEAKER_01: So it's possible, but I guess we come back therefore to policy, don't we?
- 26:36SPEAKER_01: To making sure that it is in contractually and in employment contracts
- 26:44SPEAKER_01: and very well communicated in the agency.
- 26:48SPEAKER_00: Don't do that.
- 26:50SPEAKER_01: If our AI doesn't have access to it in the drive, don't copy and paste or export stuff in.
- 26:58SPEAKER_00: There are technical workarounds here,
- 27:01SPEAKER_00: there are things where you can get your IT provider to categorize files by type
- 27:06SPEAKER_00: and let the AI and its configuration notice these certain types.
- 27:10SPEAKER_00: I think this will be a standard in companies in the future,
- 27:14SPEAKER_00: but I think simply it should just be a policy that We set up your AI environment.
- 27:20SPEAKER_00: Everything you could ever need data-wise is here.
- 27:22SPEAKER_00: We've set it all up nicely.
- 27:24SPEAKER_00: You don't need to drag in any external files.
- 27:26SPEAKER_00: And that is a simple rule that does the job.
- 27:29SPEAKER_01: Good advice.
- 27:29SPEAKER_01: Great.
- 27:30SPEAKER_01: So, Callum,
- 27:31SPEAKER_01: I feel like just before we move on,
- 27:33SPEAKER_01: there's quite a lot of negativity around what we're talking about here potentially,
- 27:36SPEAKER_01: isn't it?
- 27:38SPEAKER_01: I guess I was going to say we're not trying to scare anybody.
- 27:41SPEAKER_01: I suppose in a way we are because this is about safety, isn't it?
- 27:45SPEAKER_01: This is about mitigating risk, how to keep your agency safe, There is no wonder
- 28:07SPEAKER_00: The amount of media and fiction that has been made about artificial intelligence in
- 28:11SPEAKER_00: a foreboding and scary,
- 28:14SPEAKER_00: ominous way,
- 28:14SPEAKER_00: Terminator for example,
- 28:16SPEAKER_00: things like this,
- 28:17SPEAKER_00: because there are severe risks and I think It's almost good to be slightly scared.
- 28:22SPEAKER_00: I've worked with clients in the past where they're very concerned
- 28:26SPEAKER_00: about their own data, rightfully so, because they have all these client contracts.
- 28:29SPEAKER_00: I always say, great,
- 28:31SPEAKER_00: this is what we want,
- 28:32SPEAKER_00: ultimately, because it means that you will protect your data, which is what we want.
- 28:37SPEAKER_00: And protect your business.
- 28:40SPEAKER_01: A couple of those examples,
- 28:41SPEAKER_01: or one in particular I know you want to touch upon,
- 28:43SPEAKER_01: is some agencies,
- 28:46SPEAKER_01: all agencies are at risk to some degree of this,
- 28:48SPEAKER_01: but some agencies are more at risk than others because some agencies work
- 28:52SPEAKER_01: in more highly regulated sectors or with clients in more highly regulated sectors
- 28:58SPEAKER_01: than others, don't they?
- 28:59SPEAKER_01: Do you want to talk a little bit about that?
- 29:00SPEAKER_00: Yeah.
- 29:00SPEAKER_00: Yeah, completely.
- 29:01SPEAKER_00: So I have agencies on either side of the spectrum and in this case I work with
- 29:07SPEAKER_00: a health comms agency and we talked a lot about connectors there.
- 29:12SPEAKER_00: We have said to begin with that we're just not going to bother because there's
- 29:15SPEAKER_00: so much regulated client patient data, it's simply not worth the risk early on.
- 29:23SPEAKER_01: Of using AI?
- 29:24SPEAKER_01: With client data.
- 29:26SPEAKER_00: Yes, completely.
- 29:29SPEAKER_00: Which involves, of course, turning the connectors on as much as they'd be helpful, that data.
- 29:34SPEAKER_00: We want to make sure we're starting bottom up in a way that we only introduce data
- 29:38SPEAKER_00: when we need it.
- 29:39SPEAKER_00: And I think this is great in the format of,
- 29:43SPEAKER_00: or through the lens of AI adoption,
- 29:46SPEAKER_00: a lot of teams I work with and when I audit agencies,
- 29:48SPEAKER_00: the staff are paralyzed with fear.
- 29:51SPEAKER_00: They feel that they don't exactly know the do's and don'ts of AI safety,
- 29:57SPEAKER_00: which means that they don't use it at all.
- 29:59SPEAKER_00: And that's not what we want.
- 30:01SPEAKER_00: We want people to feel safe and be able to use these tools, experiment, innovate.
- 30:06SPEAKER_00: That's where adoption grows, where people can see the value.
- 30:09SPEAKER_00: And so there are different ways of embedding context into systems,
- 30:14SPEAKER_00: and we touched upon this in episode four.
- 30:17SPEAKER_00: But plugins,
- 30:18SPEAKER_00: and as the name would suggest,
- 30:19SPEAKER_00: it's a way of packaging certain data,
- 30:22SPEAKER_00: whether that be workflows or context,
- 30:24SPEAKER_00: and giving it to the AI tool in a nice packaged way.
- 30:28SPEAKER_00: And that's what we have done with this agency that I'm working with,
- 30:32SPEAKER_00: and it's going very well to start off with.
- 30:33SPEAKER_00: And you can get a lot of efficiency with your own internal processes with AI, which is great.
- 30:39SPEAKER_00: Most plugins can be controlled, can't
- 30:41SPEAKER_01: they, at the center?
- 30:42SPEAKER_01: Yes.
- 30:43SPEAKER_01: Of course,
- 30:43SPEAKER_01: but that's one of the many benefits,
- 30:46SPEAKER_01: but you can create the plugin,
- 30:48SPEAKER_01: you can amend it,
- 30:49SPEAKER_01: you can update it,
- 30:49SPEAKER_01: you can improve it,
- 30:50SPEAKER_01: but from the center so that everybody then should use the latest version,
- 30:56SPEAKER_01: but they can't amend it, change it, develop it themselves.
- 31:00SPEAKER_00: Exactly.
- 31:01SPEAKER_00: Yeah.
- 31:01SPEAKER_00: And this is good in the case of keeping it as a source of truth.
- 31:05SPEAKER_00: Context is very important with AI, but it's not the volume of context, it's more the quality.
- 31:11SPEAKER_00: And so if people are just dumping in loads of different files and workflows,
- 31:15SPEAKER_00: and we have a workflow here that's been upgraded five different times,
- 31:19SPEAKER_00: and one person has v2, one person has v6, it can all get out of hand very quickly.
- 31:24SPEAKER_00: And so the plugin can act as a gate where There is an approval period and we go,
- 31:29SPEAKER_00: this is our source of truth, the holy grail of our agency.
- 31:32SPEAKER_00: We want to keep this nice and organized and pristine.
- 31:36SPEAKER_00: I also think that it's worth saying,
- 31:39SPEAKER_00: I do this with every single agency I work with, create a plugin.
- 31:42SPEAKER_00: I think it's one of the most valuable things that these agencies can do.
- 31:48SPEAKER_00: It's simply a way of creating your own employee that has context of who you are,
- 31:54SPEAKER_00: who you serve and how you work.
- 31:57SPEAKER_00: The biggest advantage is future-proofing.
- 32:01SPEAKER_00: There's a lot of worry,
- 32:03SPEAKER_00: rightfully so,
- 32:04SPEAKER_00: about the trajectory of AI development in businesses and who knows,
- 32:11SPEAKER_00: once these AI companies get their claws in and get integrated very much
- 32:16SPEAKER_00: in everyone's day-to-day processes, they might then 10x the price.
- 32:21SPEAKER_00: And what happens then?
- 32:22SPEAKER_00: You simply will have to pay it because you rely on them.
- 32:25SPEAKER_00: And that's quite a scary, foreboding topic.
- 32:28SPEAKER_00: But if you can privately create your own plugin,
- 32:32SPEAKER_00: and this is external from any AI,
- 32:36SPEAKER_00: I host all of mine on GitHub,
- 32:38SPEAKER_00: that's the most popular place to store these repositories where it's basically code
- 32:42SPEAKER_00: and data information, nice and packaged.
- 32:46SPEAKER_00: If the AI provider that you work with decides to 10x their prices,
- 32:50SPEAKER_00: then you will be safe if you have a plugin because you can simply plug it
- 32:53SPEAKER_00: into something else.
- 32:54SPEAKER_01: You can move it somewhere
- 32:55SPEAKER_00: else.
- 32:57SPEAKER_01: Furniture in a house, you can buy all and create all your own furniture.
- 33:01SPEAKER_01: If you need to go and move somewhere else,
- 33:02SPEAKER_01: you can up sticks and go and live in another house,
- 33:05SPEAKER_01: but still live in the same environment.
- 33:07SPEAKER_01: The context is the value.
- 33:09SPEAKER_00: We talked a lot about this in previous episodes.
- 33:12SPEAKER_00: That's what will future-proof you if you keep that catered to,
- 33:16SPEAKER_00: organized, updated, and actually external from one specific tool.
- 33:22SPEAKER_00: I think the goal is to be tool agnostic.
- 33:25SPEAKER_00: Tools are just the processor of your data.
- 33:28SPEAKER_00: That data is the value.
- 33:29SPEAKER_01: That's good advice.
- 33:30SPEAKER_01: Just before we move on,
- 33:32SPEAKER_01: you were talking earlier about the use of starting from the bottom up and using AI,
- 33:38SPEAKER_01: particularly if you work in highly regulated sectors with clients,
- 33:41SPEAKER_01: using it internally.
- 33:43SPEAKER_01: There are a myriad of examples of how you could use that,
- 33:49SPEAKER_01: but one of them is,
- 33:51SPEAKER_01: and it's purely one of them,
- 33:52SPEAKER_01: Because it's a live example, using it to help you analyse timesheet data, isn't it?
- 33:58SPEAKER_01: So not using it with clients,
- 34:00SPEAKER_01: not processing client data,
- 34:01SPEAKER_01: not using it to answer client problems or create work for clients
- 34:07SPEAKER_01: because it's highly regulated,
- 34:08SPEAKER_01: leaving that for now,
- 34:09SPEAKER_01: but focusing internally for internal efficiency and using it,
- 34:12SPEAKER_01: timesheets, processing or
- 34:17SPEAKER_01: Exactly,
- 34:23SPEAKER_00: and with automations,
- 34:26SPEAKER_00: which I think are some of the most valuable initiatives,
- 34:29SPEAKER_00: you can possibly create workflows that you don't need to be there for and you trust,
- 34:33SPEAKER_00: you configure well and it can just do it the same every time.
- 34:36SPEAKER_00: Who doesn't want that?
- 34:38SPEAKER_00: This was the same agency as the HealthCom's agency and
- 34:41SPEAKER_00: of course they have data that's highly regulated and we didn't even turn
- 34:46SPEAKER_00: the connector on but for this certain automation all we needed is timesheet data and
- 34:51SPEAKER_00: so me working with the IT provider that they have made sure to basically put a lock
- 34:57SPEAKER_00: on everything and only give the automation a key to that one folder.
- 35:03SPEAKER_00: The team members all know everything that's put in that folder,
- 35:06SPEAKER_00: which is the timesheet exports in this case,
- 35:09SPEAKER_00: in various formats, can be seen by the automation.
- 35:12SPEAKER_00: Of course,
- 35:13SPEAKER_00: everyone should be on company plans with UK GDPR,
- 35:18SPEAKER_00: ours are too when we're doing builds,
- 35:20SPEAKER_00: but it's contracts ultimately,
- 35:22SPEAKER_00: and we don't want AI to see data,
- 35:24SPEAKER_00: it shouldn't,
- 35:25SPEAKER_00: and it hasn't been approved, and so bottom up is always the best way, in my opinion.
- 35:30SPEAKER_01: Great, Callum, there's some really useful information there I think that you've given people.
- 35:34SPEAKER_01: One of the things that strikes me I guess people might be thinking in all of this,
- 35:41SPEAKER_01: that this governance piece is hugely important,
- 35:44SPEAKER_01: but it sounds and feels quite slow,
- 35:48SPEAKER_01: quite bureaucratic, scary, lots of boxes to take and things to think about.
- 35:57SPEAKER_00: People want to get on with AI,
- 35:59SPEAKER_01: as with any innovation, any technology.
- 36:01SPEAKER_01: Get to the fun part.
- 36:02SPEAKER_01: Get to the fun part, get to the part that's really helping the business.
- 36:07SPEAKER_01: How would you respond to that question?
- 36:13SPEAKER_00: Yes, and it is completely the right opinion on it and every agency I work with says
- 36:18SPEAKER_00: the same thing.
- 36:21SPEAKER_00: I do an audit to start off with my clients and that audit of course is all about ROI
- 36:28SPEAKER_00: and claiming efficiency.
- 36:28SPEAKER_00: It's like we'll find the value that AI can bring to your business and we've got all
- 36:33SPEAKER_00: of these exciting opportunities that I structure into a roadmap.
- 36:37SPEAKER_00: Yeah, we can get to this point.
- 36:39SPEAKER_00: Always month one is governance or even week one.
- 36:44SPEAKER_00: It's just the first step.
- 36:47SPEAKER_00: And my opinion is that the foundations need to be solid before anything is built on top of it.
- 36:52SPEAKER_00: Simply, if you were to build your house on sand,
- 36:56SPEAKER_00: It would all fall apart no matter how beautiful it is,
- 36:58SPEAKER_00: no matter how expensive or how much time was put in,
- 37:01SPEAKER_00: how much it was decorated, it would fall apart and be a ruin.
- 37:05SPEAKER_00: And so this is the same with AI adoption.
- 37:07SPEAKER_00: Simply you need all of the safety protocols to be fulfilled and all the risk mitigated.
- 37:16SPEAKER_00: And this then allows people to freely experiment and innovate with AI knowing
- 37:22SPEAKER_00: that Nothing that they do can end up putting the company in despair
- 37:28SPEAKER_00: or even their own job at risk.
- 37:30SPEAKER_00: I think as well as everything that leads to governance of what we've said
- 37:38SPEAKER_00: in this episode,
- 37:40SPEAKER_00: It has nice side effects because an agency plan for example,
- 37:44SPEAKER_00: plugins,
- 37:44SPEAKER_00: connectors,
- 37:46SPEAKER_00: these are all scoping data that is shared but safe data and so you're creating
- 37:52SPEAKER_00: an environment here where you have all the necessary,
- 37:56SPEAKER_00: we use the house analogy there,
- 37:57SPEAKER_00: you've got all the bricks together and so you've got all the materials to build
- 38:01SPEAKER_00: the house and then Doing the governance at the start allows you to then crack on
- 38:09SPEAKER_00: and you can claim this efficiency very quickly without actually having
- 38:14SPEAKER_00: to rip it all apart after months of experimentation and time.
- 38:18SPEAKER_00: You figure out one of the governance steps wasn't done,
- 38:21SPEAKER_00: your client didn't even allow you to use AI with their data in a certain workflow,
- 38:26SPEAKER_00: and you've just wasted multiple thousands of pounds in time and money.
- 38:31SPEAKER_00: So completely...
- 38:33SPEAKER_00: Safety and governance is always first.
- 38:35SPEAKER_01: Makes sense.
- 38:36SPEAKER_01: Maybe a slightly controversial comment from me,
- 38:39SPEAKER_01: just to finish off,
- 38:39SPEAKER_01: I think the agency world, in my opinion, is not one of the most...
- 38:44SPEAKER_01: In terms of how it operates, agencies aren't...
- 38:47SPEAKER_01: Not usually often the most structured environments,
- 38:49SPEAKER_01: not the most process driven environments,
- 38:52SPEAKER_01: not the most businesses that have structured governance in place in any aspect
- 38:58SPEAKER_01: of this,
- 38:59SPEAKER_01: regardless of AI, lots of stuff in people's heads, a fast moving business often.
- 39:05SPEAKER_01: So I see benefits in doing this,
- 39:08SPEAKER_01: not just for AI and with AI,
- 39:11SPEAKER_01: But it's also a process of starting to put structure in the business and process
- 39:16SPEAKER_01: and more understanding and running the business more efficiently and running
- 39:19SPEAKER_01: the business out of the founders heads and into more of an infrastructure and a way
- 39:25SPEAKER_01: of working So IT,
- 39:27SPEAKER_01: we've talked about the benefits of looking at it in that way as well as just AI.
- 39:32SPEAKER_00: It's a new mindset.
- 39:33SPEAKER_01: This is putting a new mindset,
- 39:34SPEAKER_01: a new process,
- 39:35SPEAKER_01: a new structure and a more structure into a business that potentially as
- 39:39SPEAKER_01: an agency isn't the most structured company in the world in the past.
- 39:45SPEAKER_01: So that's a huge benefit itself.
- 39:46SPEAKER_00: Yeah, and you've spoke to me before about the fact of agencies not being
- 39:49SPEAKER_00: the most risk averse type of company and for the reason that a lot of
- 39:55SPEAKER_00: the times no one's going to die if something goes wrong so they can almost part
- 40:00SPEAKER_00: the brakes on some of this important stuff but AI and its development
- 40:05SPEAKER_00: and integration into agencies I think will help and give them that mindset
- 40:10SPEAKER_00: and behaviour and culture of safety and structure
- 40:13SPEAKER_00: which It can greatly benefit pretty much every area of their agency other than AI,
- 40:19SPEAKER_00: which is great.
- 40:21SPEAKER_01: Callum, another episode down.
- 40:23SPEAKER_01: Always a pleasure.
- 40:24SPEAKER_01: Always a pleasure.
- 40:24SPEAKER_01: That was a good one.
- 40:25SPEAKER_01: I enjoyed it.
- 40:26SPEAKER_00: Yeah, episode six.
- 40:27SPEAKER_00: We'll see everybody in episode seven.
- 40:29SPEAKER_00: Yes.
- 40:30SPEAKER_00: Take care.
- 40:30SPEAKER_00: Goodbye.
- 40:33SPEAKER_00: You've been listening to Gareth Healy and Callum Healy.
- 40:36SPEAKER_00: Together, we are Agents of Change, which is also our URL, weareagentsofchange.com.
- 40:42SPEAKER_01: If you want support with growing your agency,
- 40:44SPEAKER_01: including how AI can help you future-proof your business,
- 40:47SPEAKER_01: visit us online to learn more about what we do and find out how to connect with us.