Five questions
The five questions are a quick way to think about a piece of work, or a change programme, and consider if it is robust. They are best asked in order.
This page sets out how each question tends to go wrong in practice, and the first sign to look for.
How each question goes wrong
Reports and formulaic governance have replaced knowledge and understanding. Evidence is gathered, summarised, then summarised again. By the time it reaches a decision, nobody can say who checked what or what the real basis for the recommendation is. No one has lied, and each step seemed reasonable. The result is a confident picture with a lack of clarity about what has been confirmed to support it.
Warning sign: your evidence is a summary of a summary, and nobody can say who checked the original.
Usually someone has already raised or defined the problem, in a meeting, a review or a staff survey. It was noted but there was no proactive discussion and feedback, no investigation and to the wider business it looked as if nothing happened or not enough of the right thing happened.
The bigger cost comes after this point, when teams stop being attentive to the issue and raising new issues. This can happen surprisingly quickly with AI, where errors and "workslop" (AI output that looks finished but doesn't move the task on) go unchallenged. Add "work intensification", where AI lets people take on more and work longer hours without anyone deciding they should, and your team's ability to respond, improve or simply keep up suffers.
Warning sign: someone raised this months ago, nothing happened, and they've been quiet in meetings.
Real change is never easy, but now more than ever having a clear and steady rate of change is essential. If the perceived "fix" is mechanical, and it becomes updating paperwork instead of the work, the knowledge is wasted. A policy is issued, a template goes round, an online training session is run. All three are real and easy to count, which is partly why they get chosen. Meanwhile the work is done exactly as it was before.
Warning sign: the fix was a policy, a template or a training session, and the work itself hasn't changed.
The change is held up by attention, and the work itself hasn't been redesigned. It holds while there's a programme name, a steering group and a monthly slide. Those end on a planned date. Nothing replaces them, because keeping the change going was the last bullet on the final slide.
Warning sign: it held while the programme had a name and a steering group, then slipped within six months of the final report.
People measure activity because it's easy to measure: sessions run, controls tested, tickets closed, attendance, completion. All of those numbers go up. None of them would have gone down if the work had achieved nothing, which is why they're comfortable to report.
Effective change comes from outcomes and really focusing on what makes a real difference. The activity measures are important, but they more often than not simply add to the administration burden or, worse, create distraction or conflicts which distract from really progressing the outcomes.
Warning sign: you can show sessions run, controls tested and tickets closed, but not one number that would have moved if the work had failed and not achieved its outcome.
Where the questions come from
None of these questions is new on its own. Each already exists, and it is important to say where they come from. Here's the list, as it's a good guide to how to think about real impact and change, with iteration and incremental improvement needed to avoid errors, reduce risks and deliver real outcomes.
Boyd's OODA loop. Observe, orient, decide, act. It starts with observing, and Boyd saw orientation as the decisive step, the one that shapes the others. What it misses: orientation corrects the observer, but it doesn't ask who checked the observation.
Goldratt's Theory of Constraints. The first question is "what to change?", answered with a current reality tree before any solution is allowed. The tree is then tested with the Categories of Legitimate Reservation, a formal way to challenge its logic, including whether each stated fact exists. What it misses: it asks whether a fact is real, but not who checked it, or how.
Rumelt's kernel. Diagnosis, guiding policy, coherent action. Mistaking goals for strategy is one of four hallmarks of bad strategy. What it misses: it stops at action. It doesn't go on to ask whether the action landed or made a difference.
The Institute of Risk Management, Risk Culture: Resources for Practitioners, October 2012. A ten-point plan that starts by evaluating the current risk culture. What question one adds: it also starts with diagnosis, which is the right place to start, and it asks who checked the evidence about culture before anyone relied on it.
Lewin's action research, 1946. A spiral in which each turn plans, acts and finds out what the action produced. The familiar summary, "diagnosis, planning, action, evaluation", isn't Lewin's wording, and I haven't found a source that says who wrote it. A year later Lewin argued that a planned change should aim for the new level to last, as well as to be reached. What it misses: it assumes the fact-finding can be trusted.
Keeping change going, in four models. Control in DMAIC, Reinforcement in ADKAR, Maintenance in RE-AIM, refreezing in Lewin. Kirkpatrick's level 3, Behavior, covers applying learning on the job, which is close but not the same. What they miss: in all four, keeping the change going is a sub-step inside a larger stage.
What I haven't found published elsewhere
Question one is about checking evidence as well as collecting it. The key step is finding out who checked it.
Question four stands as a question in its own right, instead of a footnote to the last one.
And the whole set is aimed at work that a machine helped humans produce. This new way of working needs careful thought about how we develop our skills, organisation design and careers.