Verify. The stage most often skipped, because it can only produce bad news.

• Observe
• Interpret
• Decide
• Adapt
• Verify
• Learn

How do you know a change worked, rather than that something else moved at the same time?

Verification checks whether the adaptation produced the effect the decision predicted. It is the least satisfying stage in the cycle. It creates no new work, it shows nothing to be proud of, and roughly half the time its output is that the change did not do what was expected.

That last property is why it gets skipped, and why a business should be suspicious of any arrangement where it never happens.

The evidence is named before the change, not after

This is the whole discipline in one sentence. A decision that says what would count as success, in advance, can be verified. A decision that does not can only be assessed afterwards using whatever number happened to move, and there is always a number that moved.

Naming it in advance also forces honesty about timing. Some changes should show an effect in two weeks. Some cannot show anything for a quarter. Writing that down at the point of decision prevents both of the usual distortions: declaring victory early, and quietly extending the deadline until the result improves.

What counts as evidence

• The measure named in the decision, on the date named in the decision. Not a substitute measure introduced later.
• A comparison the business could have predicted. If the only way to see the improvement is to construct a comparison after the fact, that is not evidence, it is presentation.
• An explicit unknown where the answer is unknown. Unmeasured and measured zero are different findings, and reporting the first as the second is the most common quiet dishonesty in this industry.

Verification includes verifying the interpretation

This is the part that most reporting misses. The question is not only whether the number moved. It is whether the reason the number was expected to move turned out to be the real one. A change can produce the predicted effect for a reason nobody anticipated, and a business that records only the outcome will apply the wrong lesson next time with full confidence.

When the outcome matches the prediction for the wrong reason, that is a more useful finding than a clean success, and it belongs in the record as one.

Where this stage fails

Two limits, and both are structural rather than technical.

The first is attribution. Markets do not hold still while a change is tested. A competitor moves, a platform changes its behaviour, a season arrives. In most real situations the honest verification is a degree of confidence rather than a proof, and any supplier presenting clean causation for a change made in a live market is overstating what the evidence supports.

The second is the obvious conflict: the party that made the change is usually the party verifying it. Digilu is not exempt from this. What reduces it is that the prediction is written before the work, the record is append only, and a re-evaluation is added as a new entry rather than replacing the old one. A history that cannot be quietly edited is a weaker guarantee than independence and a considerably stronger one than good intentions.

Verification is the fifth of six stages, and the one that makes the other five worth running. The full cycle is set out on the Adaptive Brand Management page.