Interpret. Most observations mean nothing, and acting on them is expensive.

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

How do you tell a real change from ordinary variation?

Interpretation asks one question about an observation: does this represent a change in the environment, or is it variation inside a system that has not changed?

Getting this wrong is not a small error, because both failure modes are costly and they point in opposite directions. Treating noise as signal produces a business that rewrites its position every few months and accumulates nothing. Treating signal as noise produces a business that is confident right up to the point where it is no longer chosen.

The three tests

Every observation is tested along three axes before anything is decided:

• Isolated or repeated. One occurrence is an anecdote. The same movement across several independent surfaces, or the same movement sustained over time, is a pattern. This is the test that filters most of the volume out.
• Temporary or structural. A dip that recovers on its own was weather. A dip that holds is a change in the ground. The only way to distinguish them is to wait, which is why observation has to be continuous and why the answer to some signals is deliberately nothing yet.
• Local or broad. A change affecting one page, one channel, or one platform is a maintenance item. A change appearing across several at once is usually the same underlying cause showing up in several places, and fixing the instances individually treats the symptom repeatedly.

Why this stage is the one usually missing

Because it produces nothing visible. Observation produces charts. Adaptation produces work. Interpretation produces a judgement, and a judgement is hard to invoice and easy to skip.

So in most arrangements it does not happen. The dashboard shows a movement, somebody asks what we are doing about it, and the question goes straight to the adaptation stage. Work gets done, the number sometimes recovers, and nobody establishes whether the work caused the recovery or whether the movement was going to reverse anyway. A business can operate like this for years and end up with a long record of activity and no accumulated understanding of its own market.

What a good interpretation contains

A statement of what appears to have changed, what the evidence for it is, what would have to be true for the opposite conclusion, and a confidence level that is allowed to be low. The last part matters more than it sounds. An interpretation offered without a confidence level gets treated as fact, and a wrong fact is more damaging than an honest uncertainty, because it closes the question.

Unknown is a legitimate output of this stage. It is not a failure to have looked at a signal and concluded that there is not yet enough evidence to act. It is a failure to record that as certainty in either direction.

Where this stage fails

The honest limit is that interpretation is where bias enters the cycle, and it enters through the party with the most incentive. A supplier interpreting signals about work it performed is being asked to grade itself, and the pressure runs one way: toward finding that the work is succeeding and that the movement is noise.

There is no arrangement that removes this entirely. What reduces it is writing the interpretation down before the outcome is known, stating what would count as being wrong, and keeping the record where the client can read it. A prediction made in advance and kept on file is much harder to reinterpret after the fact than one produced at the end of a quarter.

Interpretation is the second of six stages. How it connects to the decision that follows it is set out on the Adaptive Brand Management page.