Stewart Kettle | 22 Oct 2025

When a UN team or government asks us to run a short set of interviews, the questions are often simple: What's happening with this programme right now? What barriers are people experiencing? What ideas might improve it? Yet before we speak to a single person, months can pass in inception, ethics and approvals. The intention is right: protect people, avoid harm, take consent seriously. But while evaluation waits for perfection, the programme carries on unchanged, leaving barriers in place and people without access.

This is the quiet paradox of public service and development work. We treat small risks with extreme caution, and large uncertainties with remarkable confidence. Data collection is scrutinised, yet multi-million-pound decisions proceed as if evidence from elsewhere automatically fits.

There's a helpful lens from decision science: in complex systems you either test and learn (explore) or optimise and scale what you think you know (exploit). Both matter. But our systems have drifted toward a culture where appearing certain beats learning quickly. The result is a perfectionism that slows investigation and, perversely, keeps services far from perfect.

Perfectionism shows up in familiar ways: long reports written for accountability rather than learning, risk analyses that make small variations feel hazardous, and a culture where even known problems aren't investigated. None of this is malicious, it's a rational response to incentives. But it pushes organisations to behave as if interventions proven somewhere are proven everywhere.

The lesson isn't that ethics or evidence don't matter. They do. The lesson is that speedy, low-risk exploration: asking timely questions, in context, at a modest scale, should use proportionate approvals, not the same machinery as a clinical evaluation. Design lightweight ways to learn early, and the heavy decisions you make later on become a lot less risky.

How to be problem-solvers in systems that are rightly cautious

• Proportionate ethics. Keep consent and safeguarding strong, but use right-sized approvals so low-risk, time-sensitive investigations can start quickly.
• Design for decision points. Tie quick listening to moments when teams make decisions (intervention design, targeting rules, comms content). Exploration earns its keep when it changes something tangible.
• Make exploration routine. Commit to regular, short investigations - say, monthly - to surface frictions, check on adaptations, and gather information on today's problems (which were unknowable a few months ago).
• Evaluate rigorously when it's time. RCTs and other heavy methods are powerful, once you've checked the basics. "Proven elsewhere" is a hypothesis, not a guarantee.

Perfection has a role. It belongs in choices that carry real risk. But when perfectionism becomes the default operating mode, it delays the small, timely corrections that keep programmes impactful. Don't aim for perfection. Aim for progress.