Conversion Hypothesis Backlog (Delivery)

Conversion Hypothesis Backlog (Delivery)

1. 01

   Freeze the measurement baseline before any page change ships

   Capture 14 to 30 days of pre-change conversion rate, sample size, and traffic mix per template. Without a frozen baseline, a 2 point lift on a 400-visit page reads as noise and the client will say so in the next review.

2. 02

   Pull friction evidence from at least two independent signal types

   Pair a scroll or click heatmap with session replays on the same URL so a rage-click cluster is corroborated by what users actually did next. Mouseflow records 100% of sessions, which makes replay sampling on low-traffic client pages viable where partial-capture tools leave gaps.

3. 03

   Convert each friction point into a falsifiable hypothesis with a named metric

   Write it as: because we observed X on page Y, changing Z should move metric M by N percent within T weeks. A hypothesis without a metric and a timeframe becomes an opinion the client can out-argue.

4. 04

   Score every hypothesis on expected uplift, traffic available, and build cost

   A checkout form fix on 40,000 monthly sessions outranks a hero copy test on 900. Rank by expected value per engineering hour, not by how interesting the UX problem is.

5. 05

   Run the top hypothesis as a controlled experiment, not a permanent edit

   VWO and Optimizely both support server-side and multi-page tests, which matters when the change spans a cart flow rather than a single landing page. Keep the control live for the full duration.

6. 06

   Set the stopping rule before launch and honor it

   Declare the required sample size and minimum runtime up front. Peeking daily and stopping at the first green number is the most common way agencies manufacture a win that fails to replicate.

7. 07

   Log every test outcome, including losses, into a client-visible backlog

   A documented loss prevents the same idea being re-proposed in month four. It also gives the retainer a paper trail showing the account is running a program rather than reacting to whoever emailed last.

8. 08

   Roll winners into the template library and re-baseline

   Once a variant wins, promote it to the default and restart measurement. Stacked unmeasured changes are how a site accumulates conversion debt that nobody can attribute.