Everyday AI at the Desk: Lessons from Google Workspace in Banking and Retail

Adoption of AI in the enterprise has been increasing at a rapid pace over the past few years, transforming how organizations tackle everyday challenges and improve productivity. What does that shift look like in practice?

To hear first-hand about implementations of AI in banking and retail, CCS Insight’s Chief of Enterprise Research, Bola Rotibi, is joined by Oscar Barlow, Head of AI Advocacy at Starling, and Colin Featherstone, Head of Technology at Morrisons. Oscar is focused on making Starling an AI-fluent bank, and Colin is responsible for networks, platforms, cloud and workspace.

We uncover how a major retailer and a digital-first bank are using Google Workspace and Gemini to tackle everyday challenges, such as connecting large, distributed workforces and cutting the time leaders spend on long documents, while keeping privacy, governance and human accountability at the centre.

The discussion explores:

• The business problems that made AI a priority.
• Rules for rolling out AI in different industries.
• Patterns of usage across roles and responsibilities.
• Lessons on how to have a successful pilot programme for AI tools.

CCS Insight has produced a free companion report outlining the major highlights from the discussion and providing practical recommendations for organizations at all stages of the AI roll-out. Access the report here.

Creators and Guests

Host

Bola Rotibi

Chief of Enterprise Research at CCS Insight

What is The CCS Insight Podcast?

Insightful audio from the global tech advisory firm.

Bola Rotibi: 00:04

Hello. My name is Bola Rotibi, and I'm the chief of enterprise research here at CCS Insight. I'm your host for today's discussion on Everyday AI at the Desk: Lessons from Google Workspace in Banking and Retail. Joining me today are Colin Featherstone, head of technology at Morrisons, accountable for networks, platforms, cloud and workspace,

Bola Rotibi: 00:28

And also Oscar Barlow, head of advocacy at Starling Bank, focused on making the bank AI -fluent. Welcome to you both.

Colin Featherstone: 00:37

Hello.

Oscar Barlow: 00:38

Hey, it's great to be here.

Bola Rotibi: 00:40

Excellent. Excellent. Most organizations aren't short of an AI headlines. They're short of workable approaches. Today we're looking at what it takes to use assistive tools inside everyday work with a bank and a retailer who are running this in production.

Bola Rotibi: 00:57

Both use Google Workspace as the primary environment, which reflects a wider shift from experimentation to practical impact at Google. Tools people can use at the desk with governance in mind. Over the next half an hour, we'll stay high level and concrete. We'll cover why each team chose its first use case in Workspace, the non-negotiables they set at the start and the habits that helped pilots stick. We'll also touch on the guardrails that make privacy and sovereignty real. Who can use what? What stays out? And how accountability works when an assistant suggests the next step. Think of Workspace here as the entry point for day to day adoption rather than an architecture discussion. By the end, you should have a shortlist to act on: how to frame the problem, what to greenlight, what to defer and how to check if the tool is generally helping.

Bola Rotibi: 01:54

So I'm going to start with Colin. You've had over 18 years of leading major programmes from Oracle Apps and the original loyalty scheme to ServiceNow Integration and Outsourcing, and recently completed a sustainability masters on the environmental ethical impacts of implementing Google Gemini at Morrison's. I'd like to ask you, what was the first business challenge you chose to fix with Workspace and AI and why that one?

Colin Featherstone: 02:22

So I think with Workspace, bear in mind we've had Workspace in for just over 10 years now. So we went live early 2015. So obviously predating any of the generative AI tooling. But the big thing for us was really how do we connect our workforce. We've got 100,000 employees across multiple business areas.

Colin Featherstone: 02:47

We've got front-line workers in our stores, we've got people working in warehouses, distribution centres, manufacturing sites. And the point of going live with Google Workspace, it was difficult to communicate and collaborate. So over the years, we've really kind of addressed how do we engage with colleagues. So I guess for me, there's been a seismic shift from traditional communication to through email, through various kind of traditional methods, non interactive methods into the world of talking to colleagues to Google Spaces, Google Chat, and then into the world now of how do we kind of assist that with things like Google Gemini where we can actually provide more of a sort of synthetic way of answering queries. Have many, many people in the company that need help.

Colin Featherstone: 03:46

That generates a lot of chat messages and Google Space posts. And that becomes very, very difficult to manage when the expectation is that we can respond quickly on those messages. So incorporating something like an AI reasoning or AI response tool, to be able to even answer simple queries is phenomenally impacting on you know running the business, running the business as much faster.

Bola Rotibi: 04:11

I mean that's really interesting because that kind of points to a kind of efficiency saving and streamlining your workflow. So getting the information faster and making much more kind of like targeted so that you smooth that transition to get sort of to help, to really kind of like collaborate quickly. So that's a fantastic use case. I'd like to draw Oscar in here, because I mean, let's draw on your engineering roles at Accenture and Starling. And you steer technology in your role and risk frameworks and skills behind, you know, Starling's own proprietary AI tools for staff and customers.

Bola Rotibi: 04:52

You know, can you also talk about the first business challenge you chose to fix with Workspace plus AI and why that one in particular?

Oscar Barlow: 05:00

Yeah sure. So I guess just to introduce Starling a little bit. We're a bank that is also a technology company. It's something that's really culturally apparent when you work here. We're very tech forward and this goes all the way up to exec level.

Oscar Barlow: 05:17

We have a history of innovation in AI. We've been using it in what you might call back office operations for seven plus years. So actually when generative AI along the first thing that we did because we are a tech company is say well let's make our own one. Let's make our own chatbot that we can use internally. And there was a lot of demand from senior people in that.

Oscar Barlow: 05:38

Again not to exec level. A lot of demands there partially because they've used it in their own personal lives but also I wanted to make sure that senior leaders would get hands on with this technology. First of all because it is so approachable but second of all because if you know the people who think that it's going to revolutionize the world are right then it's really important for leadership to have that appreciation of what this thing is, what it can do, where it's going, that type of thing. So we did that. We onboarded everyone onto this internal pilot.

Oscar Barlow: 06:12

And to just speak to the business challenges I think something that they really appreciated was being able to get through a lot of text, like a lot of text really quickly. So committee papers for instance right? You get a lot of those as a senior leader in the bank just like I've got a question about this paper that I'm reading. Know I need it answered now please and I don't want to scan 60 pages of it. We then just stopped for a moment and we went, hold on a sec.

Oscar Barlow: 06:41

Do we want to spend time building workspace productivity software or do we want to buy really great workspace productivity software and focus on being the best bank that we can? And the answer to that question was pretty clear that we want to focus on being a really great bank. We want to get really great productivity software from Google. And there were applications all over the place. I mean committee papers, answering emails.

Oscar Barlow: 07:06

I heard from so many people who were like I just cleared my inbox and it took me like a tenth of the time that it normally does. So use cases were spilling of people's mouths as you could see.

Bola Rotibi: 07:19

That is fantastic, actually. I was actually so engrossed in that. I was thinking, oh my gosh, there's so many things that you're but it's do you know what I really love about that is the fact that, you know, on one hand, you've got Colin talking about streamlining and test, bringing all the workflows together and efficiency. On the other side, Oscar, you're talking about really getting into hands of users straight away so that they could sort of really experience it, but also the senior leaders. Because actually what we've often found is the fact that commitment, when you get the commitment from up top, then that also helps with adoption and buy in.

Bola Rotibi: 07:59

But at the same time, it's actually being able to sort of get people making that decision that once you've got people sort of subscribed to this, it's actually focusing, well, where do you think the best course of action lays, whether you be a bank, whether you bring in a sort of really good sort of platform or solution that can help you. And so that is the other thing is that making that decision at the end to sort like, let's focus on, we're good at banking. And we're a technology company. But you know, Google Workspace is going to give us all of those features. And so that's where they're good at.

Bola Rotibi: 08:34

So that's a really, I like that.

Oscar Barlow: 08:36

Exactly.

Bola Rotibi: 08:37

I really like that. And a good thought for many of our listeners as well. But I'm going to ask you, what were the kind of toughest, know, so can you name any tough non negotiables you know, and did you feel you had to make any trade off? Colin, you know, could you kind of come to that?

Colin Featherstone: 08:59

I think for me that the immediate non negotiable is debt protection. Yeah, so one of the reasons why we've pursued Google is, you know, one because we already had it, which is obviously quite an advantage, but also because the system, the ecosystem protects against external data leaks. So previously, you know, we've as the kind of generative AI has taken off in the public domain, probably faster than the business domain or the corporate domain. You know, people are used to logging on to ChatGPT or logging on to some sort of tool and asking questions, and then potentially, you know, importing files and taking data out of the corporate ecosystem, which is an absolute no no. So anything that's, you know, GDPR regulated, company sensitive, anything really is just an absolute no for us.

Colin Featherstone: 09:49

So we've gone into the kind of world of, you know, forget everything else first, just do not leak data into any other system. Use Google Gemini, that's what it's there for. And you know, maintain the guardrails that that protects us against. So there was and to the point where you cannot even get access to Gemini unless you've reread the data protection policy. You've signed up to that, you've understood that, you've passed the test to confirm you've understood that because that was absolute non negotiable for us.

Bola Rotibi: 10:21

That's a good point actually. And to be honest with you in our research that we do, we do a couple of foundational studies that we do every year both for the employee technology workplace transformation and senior leaders at IT investment. And both talk around sort of some of the challenges that people talk sort of like are cognizant of or aware of or wary of. And privacy and security is one of that. So hearing what you guys do there and almost ensuring that people have kind of accepted level of accountability themselves, you know, and that this is not what they're going to leak.

Bola Rotibi: 10:59

So that's really, I think that's really important. So that's good to hear. Oscar, any non negotiables? I mean, I can imagine that privacy would be a really big one for a bank.

Oscar Barlow: 11:08

Yeah, absolutely. It sounds fairly similar in some ways. So we've taken the view that if we're going to use AI we're going to use it for the good customers and that involves respecting customers and their privacy. They trust us with their data, financial data, it's fairly sensitive data. So you've got to be sure that you're respecting that in the way that you're using In terms of and we have similar structures in relation to sort of comparing access to Gemini internally.

Oscar Barlow: 11:41

If your role is very heavily implicated in customer data then I'm afraid we won't give you access to Gemini just because we don't want to run that risk. Equally you must take a course that we call AI fluency before you're able to get access to Gemini just again so that you can understand what are these things are, right? What am I dealing with here when I see an AI generated output? So you're informed about what it is that you're going to be in contact with.

Bola Rotibi: 12:10

I think it's really nice is that the both of you have kind of sort of like you know sort of put in controls there to sort of like well are you ready to do this? Which I think is really important as opposed to just go ahead and do this, you know, because at the end of the day, you've got to sort of ensure that people can feel that their data is safe and secure. So the fact that you actually both actually have some sort of rules in place and the process and even an application that sort of checks whether people are ready, I think it's actually an incredible, you know, I think that's a really important point.

Oscar Barlow: 12:45

Yeah, think it's the case also that staff are fairly conscientious at Starling anyway. They actively seek out guidance about what okay what can I put through this and part of that course there addresses that question? They're also and maybe we can talk about this in relation to your research and study Colin. Very curious about the environmental implications of AI is something that we wanted to provide some information about that in the course as well. But yeah okay we wanted to talk about trade offs before.

Oscar Barlow: 13:16

I think I would say we didn't really feel like we had to make trade offs. There were kind of okay we've got the non negotiable of privacy and now there are two questions that I'm answering really. The first question is, is the model good? Right? And it's Gemini.

Oscar Barlow: 13:32

It's one of the frontier models produced by one of the major AI labs on the planet. Yeah it's good. Okay next question, can the model get access to what it needs? And I think this is the thing that like really sold it for me is that in Google Workspace, the Gemini Google Workspace, what you need is right there. It's right there where you're already doing your work, right?

Oscar Barlow: 13:54

You're already working in Docs. You're already working in Slides. You're already working in Sheets and you're already having meetings. Gemini is right there. It's close to hand and actually that was one of the pieces of feedback from this early pilot, one of the things that made us think that we don't want to go in the direction of making our own thing.

Oscar Barlow: 14:10

But people were saying like this is cool but it's not at reach, it's not easily at reach, right? Gemini is to hand when you need it.

Bola Rotibi: 14:19

That's a really good point actually because you want it to be part of your workflow. Working in the workflow that your company and your workforce is working in. So as opposed to something that is not, that is an addition. I think that's a really good point. I kind of want to take this on to the next level, actually, and kind of take a little bit of an under the hood, sort of at a very high level, obviously, because we don't want to get into too technical a discussion.

Bola Rotibi: 14:51

But I'd be interested in hearing from both of you as to say, if another bank or retailer called you tomorrow, what's the one implementation habit or rule of thumb you'd tell them to copy and why that? Mean I you've already given me some great examples but is there anything else that you'd actually say?

Colin Featherstone: 15:11

I think my biggest piece of advice and this is kind of generally from a workspace point of view is I'm a massive advocate of Google Workspace but one of the things that is really really terrible at is allowing administrative notes in the system. So when you're configuring things within Workspace it's there's a couple of areas but and I say only a couple and that allows you to put in the reason why you've configured it in that way. So what you've got to do is you've got to document everything and what I would definitely say is that for every setting that you can configure anything where you can put a name of a function or a text in and put something like a change record number in there so you can refer back to why you've done it. Put a review indicator in there. So if you take something like I don't know compliance rules of manipulating emails as they come in or tagging them you know why are they in there, what are they doing, if you've got something like a forwarder or something where you've done some activity, is that a temporary rule, is it a permanent rule, which is a change I could that's associated to so you know why it's done, when should it be reviewed, is that you know monthly or a yearly review.

Colin Featherstone: 16:18

So I would say that everything in the system and unfortunately you at the moment you'd have to do it offline because you can't do within the admin console is document why it's there, what it's doing and because that will be gold to you in you know three, four, five, six years time when you're going back to that or you're troubleshooting it and trying to figure out, you know, where is an issue, what's that rule actually doing. I think within the kind of world of Gemini, think there'll be more and more of that. Think it's quite light from an administrative point of view, but there'll be more and more things that you can configure, you can set, you can train, any opportunity you've got to document that and to reference it, do it because you will be thankful of that in years to come.

Bola Rotibi: 16:57

Actually that's a good point because I mean this is the thing is having both explainability and documentation and showing that when others come to it that they can easily and recognisably see what changes were made, what particular path was chosen. And that's something that is well, I should say well identified in this sort tech world is actually having the right documentation and really kind of being able to visibly see that making it accessible. That's a really good point. Oscar, anything you'd like to add to that from a banking perspective?

Oscar Barlow: 17:34

I've got so many thoughts about this one. Yes. Okay. Sure. Yeah.

Oscar Barlow: 17:39

Okay, my first thought on this and I think I would tell anyone AI I think of it now is kind of like a Rorschach test for the whole organization. The way you're going to roll out AI will be the same way that you'd roll everything else out. It's going to reflect your culture. It will do. There's no way around that.

Oscar Barlow: 17:56

In the case of Stalin that meant that we went big on empowerment. You know you at home listening, your organization made it different. For us it was big. We went big on empowerment which reflects Stalin culture generally I think. We needed to make it really easy for staff to know what they couldn't put in and then we needed to build structures to encourage them to experiment right.

Oscar Barlow: 18:20

So they could know the parameters and then they could go and find the uses in their lines of work themselves. In the case of what they can't put in that's really simple like and you've got to make it simple. It's customer data. Don't put customer data in there. And that shouldn't be in Google Workspace to begin with from our perspective.

Oscar Barlow: 18:41

But then it's quite easy right? Like if it's in your Google Workspace because Gemini is going to respect the security structure sort of this security structure and patterns that you already have set up in Google Workspace then you're kind of fine basically. And that allows your staff to go on and do that experimentation. I think I'd add another thought to this which is that there's a hard thing about teaching people to use AI which is generality right. So let's compare this with some other kind of pattern.

Oscar Barlow: 19:15

Imagine you spin up a new experiences system. There may be some factors about using the interface that are a little bit difficult but it's doing one thing and you can very easily hook that right in people's minds. They understand what the expenses system it does one thing. You did expenses like this and now you do it like that. Okay.

Oscar Barlow: 19:35

But generative AI is so general right? It can help you write an email to your manager and research competition law and draft a recipe for scones if you want. So how do you hook that onto your employee's concepts that they're already using? You need to figure out what those hooks are and again that's probably going to be different from organization to organization. That's my second point.

Oscar Barlow: 20:01

And then my third point is I've already spoken about this a little bit but you've got to get your senior leaders hands on. It's a bit of a lift but ten plus hours is a good idea to really get the feel for what you can do with this stuff. And that's going to set you up for success too.

Bola Rotibi: 20:17

I mean those are all really good points. I mean I really like the fact that it is about sort of making sure you kind of record and document and then sort of getting everybody on board. But also I like the empowerment point because I think ultimately if you empower people to give them the space and the environment with all the kind of the guardrails to kind of really test it out, then they will find the best. They'll find their own path, the best path and that's it. And then so long as they can document that and make sure that

Colin Featherstone: 20:51

it isn't a one off. And create communities.

Bola Rotibi: 20:54

Exactly, create communities.

Colin Featherstone: 20:55

Yeah, because I don't know how you've got on us, but we've used some of the Google you know space type technology to start to build those community of use cases and this yeah there's a real buzz there because you get some fantastic ideas and then people building those ideas and suddenly you've got these use cases you thought wow you know there's some really really creative people in our business.

Oscar Barlow: 21:15

It's great and it's leveraging the fact I think that we're so in some respects we're so lucky that our staff are going out and taking the trouble to educate themselves about this big new thing that's happening right. Those communities are a key way to to find to get those people together and amplify them internally and speed up the change that you know is bringing. Yeah I mean I have a couple of weekly calls about this where people just tell me about all the cool things that they're doing with AI and I'm like this is awesome!

Bola Rotibi: 21:45

Sounds like a good, I mean, in a kind of way you do want that kind of that feeling of people coming up with ideas, innovation, it makes everything very much more dynamic and that they're part of that solution and they're coming up with new ways. And we see that in our own studies that when we think about the workforce experience, both in training but also being having access to these tools and being able to have a good safe environment, that is really positive because it drives both engagement, it drives the experience, but it also drives the kind of like, I guess the innovation and connectivity with the company as well, which I think is what everybody wants. They want to feel that this isn't going to be taking anything away from them, but it's actually going to be adding to the way that they work. And that is very powerful. It's very positive in that way.

Bola Rotibi: 22:38

So it's a positive experience. Before we move on to the next one, I just kind of like want to say, was there anything you retired as a result of this? As a result of your kind of implementations? I mean, anything that you particularly retired? Because I mean obviously automated a lot of things that goes without saying.

Bola Rotibi: 22:57

But did it move other things out?

Oscar Barlow: 22:59

You know the thing I'm thinking of in response to this is not so much retiring but maybe you could say it was retiring several hours of unfulfilling work. So minute taking for committees. And you've got someone there who's just paying attention to everything that is happening and trying to make as faithful a record as possible as they can in the minutes. But we have Gemini for Google Meet stuck on in the committee and that person's job suddenly gets a lot easier. Right?

Oscar Barlow: 23:27

They were there, they know the narrative of the session, but they've also got the transcript and the notes to help them. So it's taken a job that used to be like quite hard, monogamous, take quite a long time and just collapse the amount of effort and time that's required to be taken for it.

Bola Rotibi: 23:43

Okay that's I mean that's really good and that kind of leads into the next section. I mean it seems to me that you've all sort of already given me some great certainly given our audience some great examples of things that have stuck. But I'd be interested to say what surprised you both that you were expecting to be hard that wasn't to the kind of sand in the gears that you didn't see coming. What was like, oh, I didn't expect that. That's great.

Bola Rotibi: 24:19

But oh, I didn't see that coming. Colin, any thoughts?

Colin Featherstone: 24:23

I think the thing for me because I'm a techie, I'm kind of an IT enthusiast, sometimes you've got to remind yourself that you're working in a large organization, and you need to look on the outside. So we've had a number of different sessions. And with our chief executive looking at things like customer behavior, for things like a local store, you might use Gemini to say for this store what are the 10 positive things that customers are saying and the 10 not so good things that customers are saying and then start to give some indicators to the store manager to help them out on you know the kind of themes. I think you know, from a Gemini point of view, we're very much focused on the internal use cases. But I think there's more and more kind of external use of that you know, within the kind of as well as internet based data rather than internal data.

Colin Featherstone: 25:11

That was actually when you look at it you look at it it's like that's actually phenomenal information you know to get kind of a customer perception or a top five things as a store manager what should I be looking at? And even if it's not 100% accurate it's certainly you know from a theme point of view and gives some great indication. And you can do that as you know if you're visiting store and we encourage all of our colleagues to visit multiple stores. It might be that I'm traveling on holiday in Scotland and say actually for this store tell me 10 things I can discuss with the store manager about. And then even equally internally, it's you know, what are the five or six things that might be a technology issue in this store.

Colin Featherstone: 25:48

So you can start to use you know, we've started to use it a lot more around just prompting better conversations internally, and guided conversations internally. And that to me, I think because I was looking at Gemini very much more technical perspective, having that kind of business value and customer value, you know, that for me personally was a great eye opener.

Bola Rotibi: 26:06

That's fantastic actually. That's really, I mean, in a kind of way that shows actually the sort of the importance from a business perspective, not sort of overruling the technical side, but really kind of marrying it together. Sometimes you can lose sight of one or the other. It can be a challenge. Seeing that, that's a really nice example of the business value really sort of sticking to it.

Bola Rotibi: 26:33

But it does make me laugh in the fact that you're going on holiday. And so what can I talk about this store? It's a bit of a pessimist holiday really.

Colin Featherstone: 26:44

The thing is if you've got on holiday, you've got to visit a Marcin store that's just in our DNA. Absolutely.

Bola Rotibi: 26:50

I love it. Oscar any any surprises that came any anything that surprised you in terms of oh I didn't really think that or anything that was a kind of like sand in the gears you know from your perspective?

Oscar Barlow: 27:02

Yeah yeah okay yeah so I find that there's this kind of shift that people undergo in AI usage when they get the idea that you can use the AI to help you learn things. Cool. Okay. Great. And now also one of those things is I can use the AI to help me get better at using AI.

Oscar Barlow: 27:28

Right so this is AI use sort of curving in on itself. Meta prompting is the is one of the techniques that gets that is associated with this. Asking an llm to help you craft a prompt for an llm which might be the same llm right? But like I've got a use case but I'm going work through the prompt with the LLM and then I'm going to try and do it with the prompt that we created. And that's you know the way I've described it there it's a bit of a mindset switch and I thought that it was going to be quite hard for people to achieve that.

Oscar Barlow: 27:59

It isn't, It wasn't. People achieved it fairly spontaneously in several parts of the organization possibly due to the learning that they're doing outside of work. And then because we have these communities of practice like we were just discussing before suddenly it was everywhere. So I thought that was going to be a real lift and we were going have to hold people's hands a little bit to achieve that level of usage but we didn't at all. I guess yeah the thing that was harder than I thought it was going to be was people got their hands on it.

Oscar Barlow: 28:35

I mean this is not okay this doesn't really answer your question but the thing that was a bit harder than I thought it was going to be was people get their hands on it and then straight away they were like this is great can I hook it up to this? Can I hook it up back to back? Can I hook it up to the other thing? And at that point I had to be like I'm sorry like we can't do that. Maybe we can't do that yet.

Oscar Barlow: 28:55

It's cool that you want to but I'm sorry I can't hook you up to Confluence, I can't hook you up to The Economist, I can't hook you up to your legal specialty database. I just can't do that right now but I love where your mind is going with that.

Bola Rotibi: 29:08

That is brilliant, I love that actually because in many respects it is about I've always felt with certainly for anyone who's kind of used Assistant and Workspace, it's very much about it's almost like a peer to peer working XOEO. So you're experimenting. So I'm not surprised that people were starting to sort of like get to helping sort of build out the prompt even with the LLM that they're using. So that does make sense because you kind of start. I've often found this, people who start experimenting as soon as they get the art, they assume they can immediately see the results.

Bola Rotibi: 29:46

Then they can't just start tweaking things. So actually that doesn't really surprise me. But yeah, it is a case when it's almost like you sort of open the cap. Yeah, it's almost like open the tap and then it suddenly starts gutting out. Everyone basically wants to do.

Bola Rotibi: 30:02

I want to hook up this. I really want to do this. I think that in many respects is actually it's a nice problem to have, even though it's still a challenge.

Oscar Barlow: 30:09

100%. Yeah. Can Can I tell you about my favorite thing?

Bola Rotibi: 30:12

Don't go on then.

Oscar Barlow: 30:13

I'd love to tell you about my favorite thing. Okay so this is a bit on the AI and learning side. So you may have seen that Starling did a big rebrand recently and one of our motion designers was working on this and he wanted to do some experimentation. So he woke up one day and was like, do know what today I'm myself a full featured motion graphic software package. He did that using Jamfac.

Oscar Barlow: 30:43

Not only did he just make the motion graphic software package using Gemini, What he was trying to do was we have this like nice concept to do with the flight of birds and the narration of starlings it's called. And he wanted to model that with motion graphics software. So he used the AI to build the software and teach him about motion graphics and teach him about the flight of birds and like aerodynamics and flock behaviors so that he can realistically model the flight of birds in this AI in this graphic software. There's just so many levels of augmentation that are going on and fundamentally it saved the business something like a 5 figure agency fee in this early stage of ideation. I think that just really shows the power of using these things to learn and using them to learn about themselves.

Bola Rotibi: 31:33

That's really positive. Well this has been a very positive and very exciting discussion and it's really kind of opened up the possibilities and certainly from you both companies whether that's Starland Bank or Morrison's retailers. You've got plenty of examples of how it's really kind of engaged the workforce, brought people in and broadened the adoption. So I now want to sort of put a little bit of a level set. So bring it down a little bit as they say.

Bola Rotibi: 32:05

So I'd love for you to both kind of think about certainly from the legal point of view and also from everyday controls. How did you ensure that the rollout was safe? I mean, you've already talked a little bit about this especially around privacy. But anything else that you can speak to in terms of sort of data use and sovereignty? But also what kind of guardrails did you ensure that was there?

Colin Featherstone: 32:36

Yeah, yeah, I mean, we're still relatively early days. So we've created an online training package you mentioned earlier on, got to your sign certain things like reread the data protection policy, but things like copyright law, image generation, things like PII data. So you know, if you're going to use someone's likeness, they've got access from a subject access request point of view to that. So be very careful about using someone's likeness. So we've put a lot of kind of, I suppose we've done a number of do's and don'ts which are kind of the hard and fast you cannot do this and you must not do that.

Colin Featherstone: 33:13

But then we've also built a load of guidance to say this is how we'll guide you. You're not breaching anything if you kind of get it wrong. So you're safe to have a little bit of a play. But as long as you're not doing the you know, you're compliant to the do's and don'ts then that's fine. And I think what we've really done is we've used our pilot group.

Colin Featherstone: 33:30

So we've got about four fifty people on the pilot and we'll ultimately be rolled out to about 20,000. But really kind of building that and revisiting that and adding to it to add those guardrails to it. So from a guardrail point of view, we've gone through a number of legal and ethical processes, just to really make sure that we're protecting our colleagues and protecting our customers. We're at the point now where we've got, we're kind of happy enough to roll out to the next cohort of colleagues. So probably about another 3,000 over the next couple of weeks.

Colin Featherstone: 34:04

And then we'll continually revisit that and make sure that we've got the right areas in place. Again, you're looking at the wider AI, not just generative AI, we're putting a lot in place around agentic, around all the other kind of AI aspects and building that into our ways of working. And so that will kind of, from a governance forum point of view, we've got you know, an AI governance forum now not just around Gemini, but around everything else. And so we're building that kind of out and making sure that we've got those guardials in across all of our AI applications especially when it comes to things like customer facing, customer data.

Bola Rotibi: 34:39

Okay that makes sense actually. So if I understand this right, it's really about you start off with a pilot group of people, training them, looking at the guardrails, putting them into place and ensuring that everybody's up to speed and that what they can do with what they can do and what they can't do. And then making sure that everybody's not only kind of cognizant of that but also ensuring that there is consistency with the tools both in Gemini and Workspace and the other AI solutions right through to before you then roll out to the next cohort of your users And that you want to

Colin Featherstone: 35:14

the big thing for us is we're really spending a lot of time on you know, don't forget your critical thinking, you know, check your sources, question it, be inquisitive. You know, I know Google Gemini in the kind of in the interface does remind you that you can't believe everything you read. But it's quite easy sometimes if you get the right answer for the first couple of times, then you're going to believe the third answer. And we're really, really challenging our colleagues to say challenge everything, you know be inquisitive, slightly mistrusting on the answers because you've got to be sure you're accountable. Gemini's not accountable if the answer is you're accountable.

Colin Featherstone: 35:46

And that's a really really clear message that's going out.

Oscar Barlow: 35:48

Yeah we have a similar thing with accountability. And we also advise colleagues to if you're gonna the first time you use this stuff make sure you do it on something that you're very knowledgeable about because you know I can ask Gemini about competition law and it'll provide me a research report and I'll go like that looks great. That looks like legal research to me. But my lawyer colleagues who look at that and go like yep that's good. That is not quite right though.

Oscar Barlow: 36:16

Right? And then I'm just getting the feeling for the limitations as well as the capabilities is really really important.

Bola Rotibi: 36:23

Okay that's a good point. Oscar is I'm just curious are there any hard controls that stops people doing things? Know do you have those in place?

Oscar Barlow: 36:31

I suppose the hardest control is limiting access altogether to you if you have access to customer data. At the moment that's a bright line inside Starling where I'm afraid that

Colin Featherstone: 36:44

people

Oscar Barlow: 36:44

are just not going to

Colin Featherstone: 36:45

be able to deploy it

Oscar Barlow: 36:45

to you if you've got your hands on custom data every

Bola Rotibi: 36:49

day. Which makes sense. So no, that's a hard fast rule of not you know, and that sometimes it can be the only option. But I like the fact that you both talk about you know sort of check the results, because that's really important. I know sometimes we have a bit of a dichotomy in some of the things that we find in our study where people kind of like privacy and security and one thing, but then you ask how often do they check the results?

Bola Rotibi: 37:12

It's not as often as you would like them to, which kind of things that makes it kind of there's a vulnerability point. But I think it's much about training people, and then see what your guardrails, sort of physical control guardrails that you can put in place. Right, we're coming to the end of this what has been an absolutely amazing conversation. I've thoroughly thoroughly enjoyed it guys. I think our audience will have had lots and lots of insights to follow from you guys.

Bola Rotibi: 37:40

But I want to come to the kind of, you know, really sort of finish off really about trusts and think about ethics and outcomes that you guys are both willing to stand by. First of all, and if you can do this in sort of very short way, we've already talked about the kind of the sticking points in terms of privacy and does ethics have a is ethics a major sticking point? Is there anything else that you believe? And what's the one thing you would do differently in the first thirty days? Even if you think from ethics, if you can talk about you know what would you do differently in the first thirty days as well?

Colin Featherstone: 38:19

I mean I think from ethical point of view, big thing for me is the human agency. So as we go down, certainly as we get more into agentic and we potentially give more of the decision making and the reasoning to AI. My fear is that at the moment, there's a lot of human intervention because it's new. As people, you know, and internal corporate or public get more trust in, then there may be less and less human intervention and maintain that human agency from a medical point of view, I think is absolutely critical. So my kind of not fear is positive, too strong a word, but my concern is the further and further we go, and the level of trust increases on AI, you know, there may be decision making that will just not happen with them.

Colin Featherstone: 39:05

We've got to certainly from a leadership point of view, is just make sure that we've got those break points in those human agency points that actually says we're checking things and we're happy with it.

Bola Rotibi: 39:14

That makes a good point. Is there anything you would do differently in the first yeah thirty days?

Colin Featherstone: 39:19

I think for me it's probably not differently but I suppose advice I would give is trying especially if you go into a pilot phase is really try and get the right people onto the pilot. So your creative ideas and people that you get a lot of people who want to be on a pilot but will they give it the right time, will they give it the right use cases, it's got to be spread across the different business units. If you get everybody just doing meeting note transcripts that's not really proven a lot. If you get people that you know have just got those great ideas that really kind of lift up the energy levels they're the people you want on the pilot.

Bola Rotibi: 39:55

Oh that's a really good point and I like that as a certainly as a closing remark. Oscar can I come back to you?

Oscar Barlow: 40:02

Yeah sure yeah so I think in terms of ethics, yes accountability is really important. I think also we wanted to make sure that we would look after our staff. So we know that LLMs do occasionally produce offensive output and we wanted to make sure that we had a mechanism for reporting that and taking action on it if it should occur internally. Like it would be unreasonable and and wrong I think not to allow that not to create that mechanism. So we did we've seen a couple of things come through that and we've taken appropriate action as a result.

Oscar Barlow: 40:43

I think the further thing is the environment. Keep a fairly close eye on this but look there is a climate crisis and these things do use a lot of energy. So how do we provide adequate guidance to staff internally about their usage of AI? Well I think one of those things is just don't be frivolous with it for example right. That's clearly a waste of energy.

Oscar Barlow: 41:10

Some experimentation, some learning is reasonable but you're probably going to know, at which you're just being silly with it. I think one of the things in specific relation to Google that I was really pleased with was that they've instrumented one of their data centers and published the results of the energy consumption of their AI recently. I love that transparency. The results are also encouraging. I'd love to see that work extended back into the training phase and I'd love other AI providers to do that as well so there could be a meaningful comparison between providers in that respect.

Bola Rotibi: 41:42

That's fantastic. No that's really good. Could you sort of is there anything you do differently in the first thirty days that you want to leave with the audience or?

Oscar Barlow: 41:50

Yeah yeah it's hard to communicate too much about it I'd say. Do more communication than you think you need to. I think you know we were talking just before about the different kinds of profiles that get involved. There are going to be some people who bring that energy and they're really important to get in and then there are going to be some people who are maybe a little bit more hesitant and need a little bit more guidance and encouragement to get the most out of these tools. But if you can open them up then some really impactful use cases can come out of those people.

Oscar Barlow: 42:22

Quite often it's people who have a lot of experience and domain expertise so they can give you really, really great feedback about what's good and what's working for them. And you do that through communicating often and a lot.

Bola Rotibi: 42:32

That is fantastic. Guys, this has been brilliant. I mean, I love the fact that it is about kind of communicating, it's about getting the right people on board to give, I would probably say no business, sort of a clear business outcome and a workflow outcome or workforce, a functional outcome to it as opposed to just, oh, let's try doing something sort of general, but actually coming out with something that is really purposeful impactful. And then, you know, sort of and ensuring that you're having the right people. So it's kind of once again going back to the people side.

Bola Rotibi: 43:10

Guys, Oscar, Colin, this has been a fantastic conversation. I've thoroughly enjoyed it. And as I said earlier, this has been given some really great insights and great advice to many of our listeners of this podcast. So I'd like to say thank you to both of you and looking forward to and for our audience. I want to make sure you tune into our next CCS Insight podcast.

Bola Rotibi: 43:34

And until then, goodbye.

Oscar Barlow: 43:37

Thanks very much. Cheers. Thank you very much. Bye.