Last week I visited a small soft drinks company with a quality team of five. One of them is Willem

A year ago, Willem never brought up AI. When I was there, he showed me how he now runs his label checks with an AI assistant. He set it up himself, without an IT project or a consultant.

Willem doesn’t have a technical job. What he has is curiosity, and that turned out to be enough to get started.

I see versions of Willem in a lot of quality teams right now. They’ve tried AI, it worked for something, and now the question is how to make it part of the everyday work. That’s where most of what we’ve learned this year sits, so that’s what this piece is about.

Most quality teams are past the question of whether to use AI

A year ago, a lot of my conversations with quality managers started with whether AI was even relevant for them. That conversation has mostly gone. Doing nothing with AI doesn’t seem like a real option in this market anymore.

What I hear now is how. Do you give your team Copilot or ChatGPT and let them use the chat? Do you expect people to build their own assistants, and keep them running as part of their job? Do you wait for your QMS or supplier management system to add AI features? Ask an agency to build something? Or choose a specialised tool? Most teams are trying a few of these at the same time.

A question I still get sometimes is what’s actually new here. Plenty of quality teams have been linking systems for years, with Power BI or flows between SharePoint and other tools. My colleague Niels, who worked in quality himself before joining us, explains the difference well. Linking systems turns an existing process into a longer automated one, and that only works if the process follows the same steps every time. AI can work with context. Give it what your team has done before, and when an exception comes up, it can often deal with it based on that context and how earlier exceptions were handled.

The teams most careful about AI are often the best prepared

Niels made another point I keep thinking about. Quality people find AI a bit scary, and he understands why. When he worked in quality, he knew that in the worst case a mistake could put an unsafe product on the market. Handing part of that work over feels like losing control.

At the same time, look at what quality teams have built up over the years. Specifications, certificates, supplier data, procedures, audit reports, complaint logs. The folders might not always be tidy, but the content itself is usually in better shape in quality than in most other parts of the business.

That is exactly what AI needs to be useful. So the teams most careful about AI are often the ones best prepared for it. Not every team sees it that way yet.

Start with quality tasks where you can check the AI’s work

The obvious place to start is the biggest time sink. For a lot of teams that’s customer questionnaires. They’re also complex, full of exceptions, and one of the hardest places to build trust.

What we see is that adoption moves much faster with smaller, standard tasks, like a label check, a CoA review or checking whether a certificate is still valid. These go right almost every time, and people can check the result quickly. Once you’ve checked it a few times, you start to trust it. And once you trust it, you use it again the next day.

One customer, a meat processor, decided to take one workflow at a time. They started with label checks and moved on to product specifications after that.

A prospect recently put the bar well. AI has to be demonstrably faster than doing the work by hand, without any loss of quality. Small, checkable tasks are where you can actually show that.

And in quality, being able to check the work matters a lot. The professional needs to see what the AI did and decide whether it’s right. That decision stays with them.

Getting from a first try to everyday work takes a plan

Most companies have experimented with AI by now. Someone filled in a questionnaire with it, someone compared two specs. That’s a good start, and on its own it rarely changes how the team works.

The teams that get real results start with something concrete. They want every new label checked within a day. Or customer questionnaires answered in days instead of weeks. Or never to miss a certificate expiry again. A goal like that tells you what to try first and how you’ll know it worked.

Our own thinking changed here too. Six months ago, when a team asked whether AI needed to be connected to their systems, we said they could just start by uploading documents. For a first task, that’s fine. But once something works, people want to use it every day, and the uploading becomes the problem. The latest spec is in SharePoint, the supplier data is in the ERP, the certificate is in a customer portal. Connecting to the places where those documents live is what makes it work every day, for the whole team.

A few more things we learned along the way. It works better to get one task running and trusted before starting the next. The setups that work best are shaped by someone who has done the job. At Passionfruit, that’s people like Niels, who knows what a good label check looks like before any AI touches it.

And there’s work in keeping it running. Documents change, exceptions show up, templates get updated. Our hunch is that many teams, after building something themselves, decide they don’t want to keep maintaining it on top of their day job. It keeps coming up in our conversations.

From daily tasks to spotting risks and opportunities

When the daily tasks run smoothly, people start asking different questions. They shift from “can you check this label?” to “why do we keep getting the same complaint?”

One quality manager put three years of complaints side by side with AI. What her team had handled as separate incidents turned out to be one recurring problem with one cause. It was hard to see before, because the answer was spread across the complaint log, SharePoint, line data and emails.

Every quality professional has a feeling for where the risks are. This is a way to back that feeling with data, and to convince the rest of the business.

It also raises a question worth asking now. If you want those insights in two years, what should you start recording today, and how?

Where to start

If you’re in quality and wondering where to begin, this is what I’d suggest:

• Find your Willem. Someone on your team is probably already experimenting. Give them room, and ask what they’ve learned.

• Pick one small task that’s repetitive, well defined and easy to check, like a label check, a CoA review or a spec comparison.

• Write down what success looks like before you start. Time per task, number of checks, fewer errors.

• Think about day 100 as well as day one. Where does the data come from, who keeps it up to date, and what happens when a template changes?

• And decide which complaint and deviation data you’ll want to learn from later, so you can start capturing it consistently now.

AI in quality starts with curious people like Willem. Making it part of everyday work takes a bit more, and that’s where most teams are right now.

I’ll go deeper into what we’ve seen work, and what hasn’t, at the VMT Food Safety Event on November 19 at the Midden Nederland Hallen in Barneveld. If you’re there, come and say hi. 

Get tickets here!