How I Use AI in FMCG Innovation Workshops — Without Losing the Human Truth
I started integrating AI into my innovation work before it had a name that everyone recognised. Not because I was early to a trend, but because I had a specific problem: the parts of the workshop process that consumed the most time were not the parts that required the most human judgment.
Researching category trends. Generating initial concept directions. Summarising consumer verbatims. These tasks are important — but spending three hours on them the night before a session meant less energy for the things that actually required a person in the room: reading the dynamics, challenging assumptions, knowing when to slow down.
What I describe in this article is how I actually use AI in FMCG innovation workshops — what it does well, where it fails, and the line I’ve learned not to cross.
The problem AI solves in innovation workshops
Most FMCG innovation workshops suffer from the same structural problem: teams arrive with too little prepared insight and too much reliance on what’s already in the room. The session becomes a sophisticated brainstorm among people who share the same category assumptions.
AI doesn’t fix that by being smarter than the team. It fixes it by being faster at the preparation work — so that the team arrives with richer raw material to react to.
Concretely: I now use AI to generate an initial landscape scan before any session. Consumer trend signals, adjacent category moves, relevant behavioural shifts. Not to accept this as truth — but to use it as a provocation. Something for the team to agree with, argue against, or build on. The quality of the discussion improves when there’s something concrete to push back against.
Where AI accelerates the workshop itself
During a session, I use AI in two specific moments.
Rapid concept articulation. Once the team has identified a promising direction, AI can generate 5–8 concept variants in minutes — different framings of the same underlying idea. This is not about replacing the team’s creativity. It’s about giving them more versions to evaluate quickly, rather than spending 30 minutes writing a single statement in a group. The best concepts almost always emerge from reacting to imperfect first drafts.
Consumer language calibration. I feed AI a draft concept statement and ask it to rewrite it in the language a real consumer might use — not a marketer. This simple step reveals when a concept is being held together by internal jargon rather than a genuine consumer benefit. If the AI cannot translate it, the team usually can’t either.
Where AI fails — and where I keep it out
AI has no access to the things that matter most in an FMCG workshop: the moment a participant goes quiet, the comment that gets laughed off but contains a real insight, the tension between R&D and Marketing about what’s actually feasible.
I deliberately keep AI out of three moments:
- Consumer interviews and observation. Sentiment analysis can summarise. It cannot observe. The insight that led to the Herta packaging redesign came from watching someone struggle with a dough tube in their own kitchen — not from any dataset.
- The 3P decision. Pivot, Persevere or Stop is a judgment call that requires reading the full human context of the project — the history, the team dynamics, the organisational appetite for risk. I have never seen AI contribute usefully to this moment.
- Team alignment. Getting a cross-functional group to genuinely agree on a direction — not just say they agree — is a human process. It requires reading the room, managing power dynamics, and sometimes having a conversation that doesn’t go into any report.
From my experience: The limitation I hit at Nestlé
Early in my AI integration, I tried using automated sentiment analysis on consumer interview transcripts to speed up the insight synthesis process. The output was technically accurate — it identified the main themes correctly. But it missed the pattern that mattered most: that the consumers who were most enthusiastic about the concept in their words were the most hesitant in their behaviour. That tension only became visible when I read the transcripts myself and cross-referenced them with the observation notes. AI gave me speed. The insight came from slowing down.
A practical note on time saved
I’m sometimes asked to quantify the impact. The honest answer is: I save between one and two hours of preparation time per session, and the quality of the starting material is consistently richer. Over a quarter of active innovation work, that adds up.
But the more meaningful impact is less measurable. Teams that arrive at a session with sharper raw material have better conversations. They make decisions faster because they’re reacting to evidence rather than generating it in real time. And they leave with clearer next steps — which means fewer follow-up sessions.
AI didn’t make me a better innovation coach. It made the parts of my job that don’t require judgment faster — so I could spend more time on the parts that do.
“AI is the most useful tool I’ve added to my practice. But it has never replaced the moment where someone in the room says something that changes the direction of the whole project. That moment is still entirely human.“
Amandine Devergies
If this resonates
I’ve written more on how Lean Startup methods apply specifically to FMCG contexts — including the moment after the workshop, when the harder decisions begin. You’ll find several concrete examples in my projects section. Or connect with me on LinkedIn if you want to discuss a specific challenge.
