The conversion workflow takes a surface you want to improve and produces ranked, evidence-grounded test ideas. It does the convergent work of reading the funnel and generating candidates, and stops at the call that is yours: which hypothesis to test, or whether the surface is ready to test at all.
Point it at a goal and a surface. It reads the funnel from your connected analytics.
Goal: [the conversion you want to improve, e.g. more signups]
Surface: [the page or step to focus on, e.g. the pricing page]
Analytics: [your connected source]
A baseline read of the funnel, then a ranked set of test candidates. Each candidate is a specific hypothesis built from a behavioral lever (friction, motivation, clarity, social proof, urgency, cognitive load, or trust), with its reach, a modeled impact, and a confidence band. It scores and ranks them, but it does not pick the one to run.
It does not invent effort. Each candidate's effort stays blank for you to size.
When the data is thin, it caps its own confidence to match, rather than presenting a shaky idea as a strong one.
And it does not pick the hypothesis to run. It ranks the candidates and hands you the choice, because which one to back is a judgment, not a calculation.
Input:
Goal: More newsletter signups
Surface: The article footer signup form
Analytics: connected
What it read (excerpt):
1,140 visitors reached the footer form over 30 days. 6 signups. The conversions are well below the level a reliable test needs.
What comes back (excerpt):
Candidate 1 (clarity lever): Replace the generic "Subscribe" label with a one-line statement of what the reader gets and how often. Reach 1,140. Impact: medium, modeled. Confidence: low, capped, because conversions are below the testing floor.
Candidate 2 (friction lever): Drop the name field and ask only for the email. Reach 1,140. Impact: medium, modeled. Confidence: low, capped.
Recommendation: do not test yet. The candidates are real, but the form does not have enough conversions to run a test that could produce a trustworthy result. The honest move is to wait until conversions clear the floor, or drive more traffic to the form first. Which candidate to eventually run, and whether to test sooner anyway, is your call.
This is the workflow's signature behavior. It did the full convergent job, read the funnel and generated grounded candidates, and then declined to recommend a test the data could not support. The decline is the result.
With a surface that clears the floor, the same workflow returns the ranked candidates and recommends the top one as a test candidate, framed as a recommendation rather than a decision. The choice of what to run stays yours either way.