When to use it
- Triage put part of your challenge in the complex domain
- You want to try a new approach (skills-based hiring, a mentoring model, an internal gig market) without betting the function on it
- Leaders want a plan and you know the honest answer is 'we will learn'
Why it matters
In complex situations, a single big program is a large bet on an untested theory. A portfolio of small experiments, run in parallel, lets the organization find out what works before scaling it. Each experiment needs to be safe to fail: if it goes wrong, the harm is contained and people are protected.
McKinsey's 2026 work on early-career talent offers a good example. In what it calls an answer-key model, juniors attempt a piece of work first and then compare it with the AI's output alongside a manager. It works only when it is deliberately designed and dosed, which is exactly what an experiment lets you test.
How it works
Describe a direction rather than a target. Design three different probes. For each, decide in advance what you will amplify and what you will dampen, set fairness and legal boundaries, and fix a review date.
Key ideas
Direction, not target'More juniors building real judgment' rather than 'cut time to proficiency by 30%'
Parallel probesThree different bets at once, so you learn faster than with one pilot
Amplify / dampenDecide in advance what signals mean 'do more' and 'stop'
Safe to failContain the harm, protect individuals, check fairness
The steps
- DirectionWhat direction do you want to move in? Describe it as a direction, not a numeric target.
- Three probesDescribe three small, different experiments you could run at the same time.
- Amplify signalsFor each probe, what would tell you to do more of it?
- Dampen signalsFor each probe, what would tell you to stop or change it?
- Safe-to-fail boundariesHow will you contain harm? Consider who is affected, fairness, selection, and legal or works-council issues.
- Time box and learningWhen will you review? What data and what stories will you collect?
- Who needs to knowWho needs to know, and how will you describe this so it isn't mistaken for a finished program?
Earlier in your careerDesign one experiment within your own remit and bring the results to your team.
Senior leaders and CHROsSponsor a portfolio of experiments and protect them from being judged as finished programs too early.