Conversion rate optimization (CRO) is work aimed at improving the share of a defined population that completes a useful action. It involves investigating a problem, proposing a change and evaluating the evidence for keeping it. An improvement is the aim, not a promised result.

A form redesign starts with a practical question: what is preventing a valid registration? This guide follows a fictional workshop proposal from that question to a documented decision. For the underlying action-and-denominator calculation, see the Conversion Rate definition.

Start with the action and the problem

Choose the action you want to understand before proposing a redesign. For a workshop organizer, a completed registration might be the useful outcome; clicking the form’s submit button is a different observation. Write down how the record will establish that registration actually succeeded.

Separate a confirmed fault from a suspected improvement. If a form fails to submit, repair the function and verify that it works. If the form works but you suspect its wording confuses people, record that uncertainty before changing the copy.

Make the proposed change reviewable

Imagine a fictional workshop organizer receiving questions about which session a registration form refers to. The proposed change is to display the selected workshop name and date beside the form.

First, the organizer needs to confirm the session details and the connection between the form and the booking record. A clearly displayed date would still be misleading if it belonged to a different session. Do not infer availability from the presence of a registration form.

Wingify’s CRO guide frames a hypothesis around the change, the desired effect and the reasoning behind it. For this example, keep the unknown beside the proposal so a reviewer can challenge the explanation.

Fictional workshop hypothesis record
Question in the recordFictional workshop proposal
What needs checking?Whether the session is unclear to people using this form.
What would change?Show the confirmed workshop name and date beside the registration fields.
What would we investigate?Whether people identify the intended session and complete a valid registration.
What must remain correct?The displayed session must match the booking record.

Choose the comparison before judging the result

Write an evaluation plan with the person responsible for the analysis. Identify who is eligible, how any comparison groups are assigned, what counts as the outcome and what would justify a decision. Ask that reviewer to explain the sample and duration needed for the question. A fixed visitor threshold would not explain those choices.

If you only compare registrations before and after the edit, describe the result as an observation. Keep a record of other changes, such as a different workshop schedule or promotion. That comparison alone does not isolate the effect of the new wording.

Check the outcome, the records and possible harm

Microsoft’s guidance on trustworthy experiments separates overall outcomes, diagnostic measures, data quality and guardrails against unwanted effects. An improvement in one measure can accompany a deterioration elsewhere. Define what each measure tells you before deciding that a change succeeded.

For the fictional organizer, ask separately whether registrations are valid, whether records are missing and whether the experience has become harder to use. A reported error could reflect a real failed registration or faulty recording; investigate which problem you have.

Microsoft also distinguishes monitoring during an experiment from ordinary end-of-test analysis: repeated checks require statistical methods that account for them. Plan how to detect serious problems and who can stop a harmful change. A favorable dashboard reading is not, by itself, the decision rule.

Record what the evidence supports

The decision record can say that the evidence supports keeping the change, that a problem requires reversal, or that the comparison remains inconclusive. Record measurement defects separately. These findings can coexist: a higher reported rate does not resolve missing records or an operational problem.

Close the record with the action, owner and unresolved question. For the workshop example, that might mean correcting the session mapping before evaluating the copy. Keeping an explanation of why the decision was made gives the next reviewer more to work with than a “winner” label.

Wingify CRO guide: definition and hypothesis

Microsoft Research: experiment monitoring (2021)