I just came out of a design thinking workshop with a marketing team. Result: a feature-rich AI solution, well thought out, nicely packaged. One problem: nobody in the room had clearly defined the problem to solve.
And the best part: those same teams refused to work in co-creation with generative AI during the workshop. The very AI they want to sell to their clients. The shoemaker's children found their 2026 equivalent.
It's not a judgment. It's a symptom. And it alone explains a fair share of the 79% of AI projects that don't deliver value (Gartner).
The false problem: "we need to do AI"
AI is so fascinating it short-circuits reasoning. People jump straight to "what can we do with it?" before even answering "what actually hurts the teams?"
→ Doing AI for the sake of AI — seductive.
→ Doing AI to relieve a profession from its real pain points — useful.
The nuance is enormous. The results, even more so.
The real problem: "problem first" became an empty slogan
Everyone says "problem first, solution later". Almost nobody actually does it. Here's the test I systematically apply at kick-off:
If you can't state the problem in one sentence without mentioning a technology, you haven't found the problem yet.
Concrete examples:
→ "We want to deploy a customer chatbot" → that's a solution, not a problem.
→ "We want to do generative AI in marketing" → solution without a problem.
→ "Our salespeople spend 4h/week copy-pasting data between Salesforce and Excel, costing them 12 client visits per month" → real problem. AI may not even be the answer.
The shift: AI-Native design thinking vs 2010 design thinking
Classic design thinking isn't enough anymore. Not because it's wrong. Because it was designed in a world where idea production was expensive. Today, AI produces 50 quality ideas in 2 minutes. Rare becomes the filter, not the idea.
Real case: a scoping workshop usually took 5 weeks to formalize a client need. With AI integrated into structuring, ideation and synthesis: 3 hours.
Not because we replaced humans. Because we redesigned the work gesture:
→ AI produces 30 reformulations of the problem — humans pick the 3 most accurate.
→ AI synthesizes 200 client verbatims — humans identify the 5 recurring patterns.
→ AI generates 50 solution ideas — humans arbitrate based on field constraints.
The human no longer produces the analysis. They arbitrate. Different posture, and you don't learn it watching a slide.
Monday morning: the 1-sentence brief test
Before authorizing an AI design thinking workshop, run these 4 checks:
- Can the problem be stated in 1 sentence without mentioning a tool? If no, it's a solution looking for a problem.
- Is the irritant quantified? Hours per week, euros per month, clients lost. Without a number, no post-project value measurement.
- Were end users observed in their real daily life (not interviewed in a meeting room)? Shadow AI often shows you where the real irritants are — far better than a scoping workshop.
- Does the team running the workshop actually use AI in their daily work? Barefoot shoemakers don't sell shoes well.
The tool doesn't define the need. The need defines the tool. And to guide others through AI transformation, you first need to have started walking the path yourself.
What's the last problem your team tried to solve with AI without being able to state it in 1 clear sentence?