Your company is already a system, even if nobody designed it

I start from a fairly simple idea: a company is a system, and that system can be made programmable. By programmable I mean you can see how it works, understand why it fails and improve it without relying on one person's memory, not that it is full of code.

That system takes in much more than the software: the website, the people, the conversations, the spreadsheets and the habits. AI is one more piece, so it works as well or as badly as the rest of the system you put it in.

Think about the whole journey, which starts when a customer lands on your website and writes in. Someone gets the notification and passes it to sales or support, where it gets discussed in an internal chat, decided in a meeting and logged in a tool. And with luck, someone eventually gets back to the customer.

Something is lost at every hop, whether it's context, time or the whole piece of information. And it's hard to see from the inside, because each person only sees their own stretch, but the customer does see the whole journey and notices.

When the team works remotely, the company is, to a large extent, what people write. A decision is made in a chat thread, refined in a video call and lost in an email. And a month later it's hard to tell what was decided, where or why.

Your team does what it can with what it has, and what's missing is design. That means deciding where each thing lives, who owns each piece of information and how to find it later.

Put an AI assistant on top of messy information and it will sound just as sure when it's wrong as when it's right. And if you automate a process that isn't well understood, all you get is a process that fails faster.

That's why, instead of starting with the tool, I start by drawing the flow: what comes in, who touches it, where it's kept and what goes out. And with that map in front of you, many problems jump out on their own.

Often what's missing comes before AI. It might be one place to keep the information, someone who owns each step or a rule about what gets written where. It isn't flashy, to be honest, but it works.

Other times it is time to build. For example, an agent that reads what comes in and routes it, an assistant that answers from the company's own documents or an alert when something gets stuck. The difference is that you build on a flow that already makes sense, so every decision is on record and people stop hunting for information across tools.

I've worked in agencies and large companies, across marketing, sales and product, for brands like Apple and Unilever. In London I was Global Lead, Automation & Tech at Dentsu International (Dentsu X and iProspect), one of the large media agency networks. And in 2021 I co-founded Chat Ergo Bot, a startup working on conversational AI, which I led.

I'm Head of AI GTM Engineering at Sesame HR, a title that sounds like sales but comes with a mandate over internal AI projects across the whole company. I also teach at EDEM, a business school, where I'm a visiting lecturer. And all of that together lets me walk into a company, understand how it works and propose the solution that fits, flashy or not.