Everyone is asking what AI can do.
The better question is: what would your organization do with the answer?
Capability is no longer the bottleneck. The tools can summarize, draft, analyse, classify and code, faster every month. What most organizations lack isn’t access to AI. It’s the ability to act on what AI tells them.
Faster at what?
Adopting AI blindly is just a faster way to execute nothing.
If your strategy leaks between the boardroom and the team, AI will help it leak faster. If your meetings don’t end in decisions, AI will write beautiful minutes of meetings that didn’t decide anything. If nobody knows which number matters, AI will generate more dashboards nobody reads.
Automation amplifies whatever system it’s pointed at. A healthy process gets faster. A broken process gets broken faster, and at scale, with more confidence.
So before you point AI at a process, make sure the process actually works.
Know what you’re measuring before you automate it
The most common mistake isn’t choosing the wrong tool. It’s automating something without a baseline.
If you don’t know how long idea validation takes today, you can’t tell whether AI made it faster. If you don’t know your current error rate, you can’t tell whether AI made it better. If you can’t tell, you’re running on anecdotes and vendor slides. After a while, you’ll either keep a tool that does nothing or kill one that works.
Three questions before any AI initiative:
- What decision will this improve? Not what task it automates. What decision it makes faster or better. If the answer is “none”, it’s a productivity toy. That can be fine, but call it that.
- What’s the number today? A baseline, measured before you start. Cycle time, error rate, hours spent, whatever matters for that decision.
- Who acts on the output? A named person with the authority to change something based on what the AI produces. Without them, the output is just more reading.
These are the same questions a good Product Owner asks about any feature. AI doesn’t get an exemption.
Hypotheses, not evidence
The second mistake is subtler: treating AI output as evidence.
I use AI heavily in my own work. That includes a synthetic customer panel, where AI personas react to product concepts and workflows before real customers ever see them. It’s genuinely useful. It surfaces objections, confusing steps and blind spots in hours instead of weeks.
But it runs on one rule, written at the top of the protocol: the system generates hypotheses about artifacts. It does not generate evidence about people.
A synthetic persona saying “I wouldn’t trust this” is a prompt to go and ask real people. It isn’t a finding about your customers. The moment an organization forgets that difference, AI stops being a research accelerator. It becomes a confidence machine that confirms whatever you already believed.
The discipline is simple to state and hard to keep. Use AI to generate better questions faster. Then answer them with real data.
Where AI actually pays off
When the foundations are in place, AI is transformative. The foundations are clear decisions, measured baselines, owned outcomes and meetings that end in action. With those in place:
- Validation gets faster. Ideas are stress-tested in days with synthetic research, then confirmed with real users, instead of spending months in committees.
- Documentation shrinks. AI drafts what’s needed and humans decide what’s kept. Nobody writes a 40-page document nobody reads.
- Signals get through. In regulated industries like medtech, AI can sift thousands of regulatory and industry updates and surface the handful that change a decision.
Notice that none of these is “AI replaces the process”. Each one is “AI accelerates a process that already works”.
The order matters
Strategy that can be measured. Execution that doesn’t leak. Then AI, done deliberately.
Reverse the order and you get an expensive tool on top of an unclear strategy, executed through meetings that don’t decide and measured by numbers nobody trusts.
Get the order right and AI becomes what everyone hopes it will be: a multiplier on an organization that already knows what it’s trying to do.
If you want to find out where AI would actually pay off in your organization, and what needs to be true first, let’s talk.