Your team uses ChatGPT and nothing changes. Start with the process.

Your staff make pictures with ChatGPT and have it write the odd email. The business runs exactly as it did last year. That is not the model's fault. AI only pays off once someone redesigns the process first. What that looks like at SiteOptima, with the numbers. Plus the question to ask your supplier because of it.

Animation: a team dashboard with a few ChatGPT subscriptions and no time saved, the business runs as it did last year. Then Baak the lighthouse with his lamp lit. End frame: start with the process, not with AI.

ChatGPT in the business, and still nothing changes

You took out a few subscriptions. One employee uses it to make a picture for social media, another has it rewrite a tricky email. All fine. And yet: the quote still takes three days, the same colleague still retypes data from the email into the system, and at the end of the quarter you still spend an evening on the VAT.

That is not the model’s fault. You put a clever tool on top of a way of working that has not changed. Then one employee saves a quarter of an hour and the business saves nothing.

Where the money is

In the US, large companies now pay top salaries to so-called forward deployed engineers: people who go in at the client, first work out how the work really runs, and then rebuild it with AI. In a masterclass by Greg Isenberg his guest put it like this: AI is not a coat of paint you put over a business, it is redesigning the process.

So the valuable part is not the model. Anyone can buy the same model. The valuable part is the implementation: mapping where the data comes from, where the work waits on a handoff and which exceptions there are. And only then deciding where AI takes over.

What it looks like at my company

SiteOptima has four routines I used to do myself that now run by themselves. For none of the four did I start by asking which AI tool to pick. Each time I started by asking how the work runs.

  • The VAT return takes me two minutes. The bank statements come in, every booking gets a category, and the dashboard calculates every box. I copy the numbers into the tax office form and that is it.
  • Follow-up emails go out by themselves. Since 20 April, 775 personal follow-up emails went to 166 leads without me clicking send. Personal here means: an email only goes out once there is an analysis of that lead’s site, so it contains something about that lead. The email shows a screenshot of the recipient’s own site. Not everything went well. In one week in May I hit my mail provider’s daily sending limit, and 109 emails got stuck because of it. I test the first email in two versions against each other, so I adjust on numbers rather than gut feeling. So far there is no winner: of the 85 emails in version A, 34 were opened, of the 80 in version B, 35, and both brought in three clicks. That is a result too.
  • The analysis leads to a change. A team of AI agents looks at ads, organic traffic, the funnel and where visitors drop off. Since May it has run about thirty times. What comes out is not a report to read but a proposal to change a campaign, a blog post or a landing page.
  • The inbox sorts itself. Since mid-October 30 emails came in. Ten landed in the file of the client or lead by themselves. For a client, a summary and a draft reply come with it. Twenty, spam and newsletters, never reached me. A supplier’s invoice is recognised and processed.

That last one has only run for a few weeks, so I am not turning it into a yearly figure. The point is the order: first the process, then the data in one place, and only then the part AI does.

Why you cannot simply copy this

Now the honest part. I had it easy. I started from scratch, with no existing way of working and no old systems. Every part of the business has lived in one database from day one, and my agents can reach all of it.

You do not have that. Your process lives in people’s heads, in a shared mailbox, in an accounting package and in three spreadsheets only their maker understands. Then you cannot put AI on top, because there is nothing to put it on. The first step is mapping the process and bringing the data together.

And that is hard to do yourself. Not because you cannot, but because you do not see your own way of working as something that could be different. What has gone this way for ten years does not feel like a choice. That is what you need someone from outside for, someone who looks at it without the habit.

The question to ask your supplier

When someone comes by wanting to sell you AI, do not ask what AI can do for your business. The answer to that is always yes and always vague.

Ask this: which process do you redesign first, and how do you measure that it got faster?

A good supplier then names one concrete routine, wants to see how it runs today, and agrees up front on what will be measured. Sometimes they also say that process needs no AI at all. That is not a weak answer. It is the proof that they looked at your process first.

Frequently asked questions

We already pay for ChatGPT or Copilot. Is that money wasted?

Not necessarily. It helps an employee with a single task. But as long as the way of working stays the same, the business gains little: the quote still waits on the same person and the data still sits in five places. The gain only comes when someone tackles the process itself.

Which process do you start with?

A routine that comes back often, runs the same way every time, and where someone now retypes data by hand or waits on someone else. For me that was the VAT return, following up leads and sorting mail. Which one it is for you only shows once you map it.

Does the new process always need AI?

No. Sometimes mapping shows that a simpler way of working, or one shared database, already solves the problem. A good supplier says so. Someone who comes back with AI for every process has not looked at the process first.

Can't I just do this myself?

The building, perhaps, if you have the time. The hard part is seeing your own way of working as something that could be different. What has gone this way for years does not feel like a choice. That is why it helps when someone from outside maps the process.

Ready to be found by AI?

SiteOptima helps SMBs in and around Eindhoven become visible in Google and in ChatGPT, Gemini, and Claude. Start with a free audit.

Start free audit →