De Meetlijn — the record

Chapter three · method

Proof, not analytics.

A raw before and after can lie. Every Whyback measurement keeps its control group, its horizon and its caveat in the same record — including the runs where the fix made things worse.

              Horizon 90 daysControl heldRollback in one click
            

§01 — §05

§01

Read the notes your customers already wrote.

Every return carries a reason in the buyer's own language. We cluster those notes per product rather than per store, because a sizing problem on one shirt is not a sizing problem on your catalog.

§02

Separate cause from noise.

Page-caused returns sit above the threshold; change-of-mind sits below it. Only the first can be repaired by a product page, so only the first is ever reported as actionable.

§03

Hold a control, always.

Your unfixed products carry the same season, traffic and discounting as the fixed ones. Whatever moves in both is the market. What is left over is the fix.

§04

Print the record.

Before, after, control, horizon and caveat on one Return Receipt — the same document whether the number went down or up.

            Fig. 1 — Return Receipt · Linen Shirt 04Horizon 90 / 90 · illustrative
          

Before15.0%
After9.8%
Control drift−0.4 pts
Adjusted ∆−4.8 pts
CaveatSingle product

Caption — solid line: fixed product. Dashed line: control. No seal, no badge, no recovery claim. Figures shown for illustration only.

            Fig. 2 — the control, weighedClick either pan to add a week of data
          

FIXED PRODUCTS: 3 WEEKS. CONTROL GROUP: 5 WEEKS.

Caption — the beam settles only when fixed and control carry the same season, traffic and discounting. A result read off an unbalanced beam is not a result.

§05

Failure stays in the record.

When a fix performs worse than the control, Whyback says so in the same document and offers the rollback. The failed branch is never deleted.

+2.4

Points worse. Rolled back the same afternoon.

              Recorded in RotterdamNo seal · no badgeRollback preserved
            