SuperAPI AI Directory

improve-conversion

General & other

View repository

Updated 10-16 08:47

About this skill

Improve conversion performance

Return a diagnosis, revised work, and one discriminating test—not a decorative scorecard.

Establish the intended path

Identify:

  • intended audience and costly moment;
  • promise, mechanism, proof, and objection;
  • acquisition source and message expectation;
  • primary action and downstream value;
  • available exposure, behavior, conversion, revenue, retention, and qualitative evidence.

Separate missing data from zero. Treat vanity engagement as diagnostic, not business performance.

Locate the first meaningful break

  1. Reach: the right people did not encounter it.
  2. Attention: they encountered it but did not stop.
  3. Comprehension: they stopped but did not understand.
  4. Belief: they understood but did not trust the promise.
  5. Motivation: they believed it but did not care enough now.
  6. Friction: they wanted it but the next step was costly or broken.
  7. Value: they acted but the product or offer did not deliver.

Inspect audience, channel, device, geography, and customer mix before aggregating. Treat co-timed movement as correlation until evidence distinguishes causes. State the mechanism connecting each proposed cause to the observed behavior.

Prioritize a discriminating test

Name plausible alternative explanations. Choose the smallest change that produces different predictions for the leading explanations.

Define:

  • hypothesis and causal mechanism;
  • one intentional change;
  • target segment and surface;
  • primary outcome and guardrails;
  • observation window;
  • keep, revise, or kill rule.

Accept “inconclusive” when evidence cannot discriminate.

Deliver

Lead with:

  • Keep: what evidence supports retaining.
  • Drop: what is weak, harmful, or unnecessary.
  • Change: the revised message, structure, or experience.
  • Test: the highest-value uncertainty and controlled experiment.

Include the ready-to-use revision whenever the medium permits. Tie every recommendation to observed evidence, an explicit inference, or a labeled assumption. Do not claim causality or conversion lift without evidence.

Keep deployments, experiment activation, spend, publishing, and other external writes behind explicit approval.