Introduction: From Google to “AI Freedom”

I've worked for the last six years at Google, and when I say that, a lot of people assume two things. Number one, I'm a highly technical person. No.

Number two, I had access to the most amazing and advanced AI tools in the planet. Well, Google, like any company, is also a company that

wants to protect its data and wants to prevent their employees from experimenting too much with with AI in an unfiltered way. And so that does mean that I had one thing missing, what I call the AI freedom.

A New Role at Eugeo

So in October, I moved from Ireland to Switzerland. And in the process, I decided to change company. And I joined a much smaller company named Eugeo as their CMO, so Chief Marketing Officer.

So Eugeo is a marketplace where you can find service providers for at -home services. So if you're looking for a handyman, a cleaning lady, community, someone to help you move, someone to look after your kids, pets, et cetera, you can find a provider on the platform. So very simple business, let's say.

The Marketplace Business in Brief

Early Challenges: Team Fit and Data Access

But when I joined the company, I faced two challenges, main challenges. Number one, everybody in the company working in the marketing team was put there because there was a gap. That didn't mean that they had the skills nor the will to be in that specific team.

And the second challenge I faced is when I joined the company, I moved from a company, so Google has access to a lot of data, and I moved to a team in marketing where no access to data whatsoever.

And I'm like, how do we make decisions around here? I'm, like, lost.

Working With a Skeptical CTO

But I was working with our CTO, and our CTO is the kind of person that says, obviously, no before saying yes, so everything is impossible until you prove him otherwise, which made things very difficult for me at the start.

Metabase, SQL, and Undocumented Data

So when I joined we had no access to data but we had one thing that we were using that was Metabase. Metabase for those of you who don't know it is a database analysis and like a kind of a Power BI thing where you can query your data through SQL.

I'm not a technical person I'm like great how do I do that and so I kept bothering my CTO to help me with SQLs because, like any small companies, the data lives and the tables of data lives in the heads of people and is not documented. It was very hard for me at the beginning.

What do I do? I don't know how to code. I don't know how to write SQLs. I don't know how the data is structured. What do I do? And so that was kind of the challenges

and the environment in which I was fast forward a couple of months later I decided to fire a couple of people in my team which is a good idea but then you know you're left with a smaller team and how do you deliver and do things with a

The Spark: Automating a Personal Pain Point

smaller team and at the same time in my personal life I have a property in France and I know and I need to do my taxes with the French states and last Last year, I paid an accountant 700 euros to do my accountancy for this. And this year, I'm thinking, can I save 700 euros?

So as my partner, we decided to build a skill on Cloud and later an agent to do our taxes for us in France. Works great. And I'm like, well, can I do that for tasks in my team?

I didn't know at the time what was an MCP. I didn't know what was an agent. I didn't know what was a skill.

so I said I'm a non -technical person so I decided okay let me try to experiment and there is one

First Marketing Use Case: Translation at Scale

thing that we were spending a lot of time on that I thought we could definitely automate automate yeah yeah automate thank you is the translation of marketing materials from one country to another we operate across 10 countries and that does mean six different languages that that we need to cover with marketing materials.

So I'm like, how do we do that? And also, if you're not familiar with marketing, when you receive an email from a company, the company is using a CRM to send that email. And that email has a lot of different things in it that needs to be localized every time we change languages. And so that does mean a lot of different things, alt text and UTMs, and da, da, da, to track and to change every time.

So, I ended up spending a bit of my weekend, don't do that, but stay with me on the idea, on how do I build kind of like an agent doing that. And I started using Codex from OpenAI to do that. I clicked on a few buttons that honestly I shouldn't have clicked on, but here we are, learning, failing fast and learning. And I happened to build that translation agent that does 80 % of the job and bring it back to my team. And I'm like, what? The amount of time we're winning? So I'm excited about that.

And later what happened is that I'm like, okay, let me share with my leadership what's happening there and see if we could make it bigger. And so I shared this little win, what I call a little win with them. And what happened was a couple of next steps that I will tell you in a second.

Getting Buy-In for AI Automation

But the one thing that I wanted to share is how do you get buy -in? Because in bigger companies, when it's not just you on your own, sometimes you have to convince people. Sometimes most of the time you have to convince people to do things that they don't necessarily want to do in the first place.

Step 1: Find an Internal Influencer

So the first step was to find, for me, was to find my internal influencer. I'm not a technical person. person but if i want to build agents and integrate open cloud for instance etc i need someone with the technical credibility to convince the leadership and for me that was strangely enough

my cto because i kept asking him about those sqls and at some point he got so fed up that he was like okay let me find something where you don't ask me the question again uh the second thing was

Step 2: Prove Value With a Micro-Win

and is to build a micro win first so you don't need a big case that changed the world and then You need like a micro -in just to show them what is possible.

And people are like, okay, but what do we do with GDPR and what do we do with the data, et cetera? Because we talk a lot about agents, we talk a lot about skills, et cetera, but concretely, what do you do with them? What can they do for the business?

Step 3: Speak ROI, Not Jargon

And the third thing is to really speak return on investments, not speak MCPs, integration, agentique, blah, blah, blah. You lost them, especially if your leadership is more in their 50s. Sorry, guys, but another generation, and sometimes they're a bit lost with all of that.

So what they care about is, does this save money? Does this free up time? How can we use it? Is it simple to use?

Those kind of questions that you can share with your micro -in.

What They Built: Agents in the Marketing Team

And so we ended up integrating OpenCloud and building a couple of things and a couple of agents that help us now in the marketing team. I can't share all of the agents we built and what they give in terms of outputs because of confidentiality, but I will share like what are they doing in general and maybe a couple of use cases.

So the first thing, so at the moment we have four agents existing in the team. It feels like I have four additional team members. It's just the beginning, so we'll have a bit more coming up.

Agent 1: Translation and Localization

But we have the translation agents, so basically the upgrade from my garage thing, let's say, that is translating into the languages we want, any newsletter, any content, any piece of marketing.

For some of them, it's reintegrated into our CRM or reintegrated into other acquisition channels, et cetera. For some of them, it's just the output in a specific format because the API of the tool we're using doesn't support the reinjection. So in that case, we're like, okay, it's going to do 80 % of the work and then it's fine already.

Agent 2: Self-Serve Data Analysis in Metabase

The second one is a data analysis so one that operates on Metabase where we can ask whatever dashboard we want, whatever analysis of our back -end data we want.

For instance today I was asking it in the flow of reservation what are the things where we're losing points in terms of booking rates and to see how can we improve that versus competitors.

Agent 3: Google Ads Monitoring

And the third one we have is one that we're going to do on different acquisition channel but one that we have

because our most important acquisition channel is Google Ads we have a twice a week crunch where it goes through all our 60 maybe campaigns and see does the cost is a cost changing in a way that is weird is a conversion changing in a way that is weird is something off that I should point out to the user so that you

can double check which enable us to pick up things before a human could pick up a trend or something happening in the campaigns.

Agent 4: Cross-Channel Business Analyst

And the last thing we have is one that I call a business analyst.

It's basically connected to all of the acquisition channel, all of the data that we have. So connected to Metabase, Analytics, Meta, Google, TikTok, and is able to analyze our

data across all of those acquisition channels, which in marketing is a freaking pain to do otherwise. So it's one that has access to all of the data.

And if it doesn't know if this triggered this or this explain this it makes assumptions or potential causality element which is really useful for us when we look at the data but we also do

have we also do have skills we all mainly operate on anthropic and so we have clothes and we build a couple of skills that are quite useful for the marketing team so at the moment we have yeah five operating I would say and then

and a sixth one on the go, yeah, four operating. I don't know how to count anymore. And a sixth one on the go.

Skills Layer: Standardizing Brand and Content Production

Skill: Brand Tone of Voice

The first one that we built was a brand tone of voice because now we have new team members coming in

and that obviously words we don't want to use as a brand, a tone of voice that we want to use in different marketing materials.

There are also things we absolutely want to put forward because that's our main USP, et cetera, et cetera.

And we package that in a skill include cloud so that whenever we create content with cloud it knows oh okay we're tapping into that and here is oh no you can't say that or here is the content or your idea looks like this etc etc

Skills: Copywriting for Automations and Newsletters

the two others we built are mainly related to copywriting and they're kind of like sister skills because they depend on the brand tone of voice so you have a copywriting for everything

that has to do with automations so when we build let's say if you're not working in marketing building, let's say you book a moving service with us, you're going to receive at some point something to tell you, oh, we've seen you booked a moving service with us, but we do all of those other services you might need, et cetera, et cetera.

And oh, you've done this activity on the side, and this is like an automation that is going to follow, that is going to email you, WhatsApp you, push notification, et cetera, et cetera. So, we built incorporating skills to help us with the automation part of it and help us build those automations.

And then one that is pretty similar, but just for our newsletters, we are firing twice a week to all our database.

Blog Content and SEO: Skill vs. Agent

And then I've built one also on blog content and SEO, and I'm currently trying to decide between a skill and an agent.

I think I will go with an agent for the SEO as well, having an agent that is a master orchestrator that is scrapping content that could be useful to have article of blogs that would help us be referenced on llms and all of that lovely stuff and then have 10 i'm thinking or six because

we have six languages so far a seventh on the way different eight sub -agents within that agent to help us with localized content that respect seo friendly guidelines so this is what we are thinking thinking of.

How It Looks in Practice: Sharing Skills and Designing Flows

In practice, it looks like this. You know where to find, I'm guessing, I hope, the skills.

So it looks like, oh, and I have a push notification one. So we have all of the skills that are here. Some of them are not finished yet.

So we're going to have them soon. And because we share the same account with my team, everybody has access to those skills but you could definitely also share them across accounts that's totally feasible

the other thing we have and we also start experimenting with i think i was

Using Claude Design to Prototype Product Improvements

doing that earlier today is cloud design just because it's easier for us to materialize something with our it teams and easier for them to materialize something so for instance instance, I was comparing our flow on cleaning with one of our main competitors because they're doing way better on this specific category than us.

So I put screenshots of all of their flows and what I found interesting, et cetera. And I did the same for our flow. And I put on also some data from our agent on Metabase, just where we're losing edge, basically, and and losing points in terms of booking rates. And so, so far, this is like the first output.

It asked me a couple of questions of what's important to you, blah, blah, blah, blah, blah. And that's the first output.

The flow that we have today looks like this. Let's say I want a cleaning service. It's going to ask me for how many hours I want, whether I want it like once a month twice a month when do I want to start where is the time so it doesn't matter and now you know where I leave you can add pictures and then you can type like blah blah blah whatever you want and then it's gonna go into like a job post its face but it's gonna say your request is online and then a couple of providers will also position themselves on this and you can book the ones that you do want so this is how the the the flow works um but here

it made like another flow um i think i don't know if i'm back at the beginning yeah another flow based on what i told it um this is the first output tomorrow i'm planning to fine -tune that and be like i don't like that i do like that i want more of this and obviously in my screenshot i didn't take the top of the page so here we are but yeah but it looks like this uh when where do you want the cleaning when are we starting etc etc um that is and you can even interact with it

which i think is like really nice i want three hours it happens apparently in toulouse but you know never mind what do i have let's say i have a maison um i want my kitchen to be clean like Like, it's like a completely new flow based on what I just ask and put in terms of data. And then it stays out.

So I'm planning to refine it tomorrow, but it's also a really good way to simplify the exchanges we can have with our IT team about, like, what we want things to look like, because now it looks much more real and much more...

When I say top right, this is what I actually mean. or when I say I want this to look like this, this is what I actually mean, which has been also really helpful for us in terms of developments and product development. And then for the IT team, they can also pick up the code on this and go much faster.

Conclusion

So this is what I wanted to share with you. Hope that's helpful.