Facebook tracking pixel Skip to main content

How we know it worked

What counts as a result.

A number going up is not the same as a change working. Before we call anything a result, it has to pass the same four checks, whichever way the number moved.

Operating process board for What counts as a result

The four checks

What every result has to show.

The checks run on every change, good news or bad. A change that fails one is still recorded, but it is never reported as a result.

Check one

A baseline from before.

The starting reading is taken before the change goes live, from the same report we read afterwards. A baseline pieced together later does not count.

  • Read before the change
  • Same report, before and after
  • Never rebuilt afterwards

Check two

One change on the path.

Only one change can be live on a path at a time. When two things change together, neither can take the credit, so the second one waits.

  • One live change per path
  • The next change waits its turn
  • Enforced by the database

Check three

Dates and a source.

Both readings carry their date range and where they came from, such as the store platform or the menu report. A number without dates is left out.

  • Exact date range
  • Named source for both readings
  • No vague periods

Check four

What else was going on.

Every row lists anything else that could have moved the number that week: a promotion, a holiday, a change in where visitors came from. You judge the result with all of it in view.

  • Promotions and holidays
  • Shifts in where visitors came from
  • Written down, never dropped

When it does not count

Most changes are small.

Some changes make no difference, and a few make things worse. Each outcome has a set path, so a disappointing week is recorded rather than quietly dropped.

No movement

Recorded as no change.

If the number holds steady through the reading window, the row says so and the next item on the list goes live.

  • Row kept
  • Next item goes live
  • Nothing claimed

Worse

Reversed and kept on file.

A change that hurts the number comes out, and its row stays in the ledger with the drop written down.

  • Change removed
  • Drop recorded
  • Reported like a gain would be

Unclear

Given more time.

When something else clearly moved the number that week, the window is extended or the change is tried again later.

  • Window extended
  • Or retried later
  • Reason noted on the row

Direct answers

Questions about the checks.

Short answers to what people ask most about how a result gets judged.

Answer

Is this the same as A/B testing?

Not quite. A split test shows two versions to separate groups at the same time, which needs a lot of traffic. Most stores and business sites do not have enough, so we change one thing, compare it with the baseline, and record what else happened in the window.

  • Baseline
  • One change
  • Written record

Answer

Who decides how long a reading takes?

We set the reading window before the change goes live, based on how many orders or inquiries the path sees. Choosing it afterwards would let anyone stop the clock on a good week.

  • Baseline
  • One change
  • Written record

Answer

What if the number gets worse?

The change comes out, the row stays in the ledger, and the drop is reported the same way a gain would be. A record that only keeps wins is not measuring anything.

  • Baseline
  • One change
  • Written record

Next step

Measure before you change anything.

The free plan names what to fix first, and every AI System Build starts with the baseline written down.

Build my AI system