There is a lot of pressure on Product Managers to “put AI on the roadmap”. I think that is backwards.
The starting point should not be: What can we build with AI? It should be: What problem are we trying to solve, and is AI actually the best way to solve it?
In one recent engagement, I redesigned a lead-generation and CRM workflow using in-house product tools, Microsoft Copilot, Claude and automation. The interesting part was not adding an LLM. We looked at the whole workflow: where time was being spent, where judgement was required, what information was needed, what could be automated safely, and where a person still added value.
AI could research companies, assess prospects, structure information and draft communications. But we deliberately retained human judgement around qualification, prioritisation and final messaging.
That is Product Management. The job is not to find somewhere to insert AI. It is to decide where technology creates enough value to justify its complexity, cost and risk.
I increasingly use four questions: Problem: what are we actually improving? Value: can we measure the benefit? Trust: what happens when the AI is wrong? Human judgement: where should a person remain in control?
The best AI products do not necessarily feel like AI products. They simply make a difficult workflow considerably better. Do not start with AI. Start with the problem.
The four questions are a good filter, especially “Trust: what happens when the AI is wrong?” That one tends to decide the design more than the other three. In the lead-generation workflow you describe, keeping human judgement on qualification and final messaging makes sense, because that’s where a confident but wrong output costs the most. It helps to write down the failure case before building anything: who notices, how fast, and what it costs. If nobody can answer that, the workflow isn’t ready to automate yet. Agree that the best AI products rarely feel like AI products.