AI-implementatie

How to automate your processes with AI

Marc Cornelius19 October 2026 · 4 min read

You have a Zapier here, a macro there, three spreadsheets someone once made that are now sacred. Each piece saves a little time. Together it is a house of cards nobody dares to touch anymore. Scattered automations are fun until they pile up. At some point you no longer want hacks, you want a system.

Automating a task is not the same as building a system

Automating a single task is tying two tools together. A new sign-up comes in, lands in a sheet automatically. Fine, that is what Zapier is for. Quick, no custom work needed. Building a system is something else. It means turning all those little steps into one working whole, with one place where the truth lives. The tipping point is here: the moment you start gluing several hacks together to make them cooperate, you are already building a system. Just a wobbly one.

Clean up first, automate second

The biggest mistake is automating a messy process. You get mess, only faster and at larger scale. Automate a step that three people each do their own way and you freeze that chaos into code. So map your entire process first. Without thinking about a tool. Start at step one in a simple sheet: what goes in, what is the next step, and the one after. Only once the process is right do you decide what AI handles, what custom software handles and what a simple automation tool handles.

“Automate a messy process and you get mess, only faster.”

Which processes lend themselves to automation

Not everything needs automating. You are looking for processes that tick three boxes: lots of manual work, lots of time spent, and nobody enjoys doing it. Repetitive retyping done by highly paid staff is nearly always the best candidate. The most expensive half hour of the day usually does not sit in the expertise, but in retyping dimensions into Excel, formatting in Word, printing, scanning, emailing. That is exactly the work you hand back to the machine. The human work, the judgement and the contact, stays with people.

From hundreds of spreadsheets to one system: TOADS

TOADS tracked hundreds of locations, across five brands and two countries, in scattered spreadsheets. Adding new participants, linking data sources to a location, checking who still had no GA4 or business profile. All error-prone manual work in sheets that kept drifting apart. We built them a custom CRM. Nearly 700 locations, seven types of data source per location, all neatly connected. And controllable from Claude via a dedicated MCP server, a kind of universal plug between the AI and the system. Now the whole network sits in one system instead of scattered sheets nobody trusts anymore.

Where the human stays in the loop

A running system is not a system on autopilot. AI is sometimes too eager and has no taste. We once built a packaging supplier a quote tool that generates quotes automatically from emails and WhatsApp messages. On one request the AI combined a box type with a layout that looked terrible. The AI approved it instantly. The human followed up to check they were sure. Turned out to be a mistake. That is humans in the loop: the machine does the boring work, the human makes the call where taste and judgement matter.

What it delivers, and over what timeframe

The gain is not only in time, but in errors that stop happening. A schedule that used to be made in Excel, with all the sloppy Excel mistakes that come with it, one client replaced with a custom planning tool. No more errors, no more versions drifting apart. And the retyping that disappears adds up hard when you total it per week. Half an hour a day per person looks like nothing, until you work it out over a year and a whole team. Do not expect a miracle in week one. Expect a system that, from month two, removes the same hassle every single week. Not a sexy promise, but one that holds up.

Automation is not a one-off project

A running system is not finished. Your business changes, so your system changes with it. That is why we come back after every phase for an adoption check, roughly every month, to see whether it is genuinely used and where it chafes. Because an automated process nobody uses is more expensive than the old manual work. Building the tool is one step among many. Making sure the team works with it and that it keeps fitting, that is the work that holds on to the gain. So together with the client we appoint an internal AI Lead who ensures enough support inside the company.

Do it in the right order

Start with the process. Clean it up. Prioritise by pain. And only then decide what goes to AI, what to custom software and what to a simple integration. Do not tinker with a broken process. Want to look together at which process nags you most and lends itself best to being tackled first? Book an intro and we will walk through it with you.