Define the visitor task, completion event, review window and stop conditions before changing the website.
A team changes a button label, moves the contact section and publishes a new offer in the same week. Enquiries rise. Which change helped, and what was the team trying to improve in the first place?
Write the observation plan before the redesign. It gives you a stable question to return to when the results are attractive, disappointing or too sparse to interpret. It also exposes measurement gaps while they are still easier to fix.
This article helps plan one narrow website change. It does not forecast traffic, set a universal sample size or promise that a before-and-after comparison will establish cause.
Start with a visitor task
For a fictional adult craft-workshop site, the task is: “Find the preparation instructions before arriving.” Visitors currently reach those instructions through a vague “Resources” link. The proposed change is to label the link “What to bring” and place it beside the workshop details.
The hypothesis is that visitors looking at a workshop page will more readily find the preparation information. Keep the hypothesis about that task. A rise in newsletter sign-ups would not directly answer it.
The GOV.UK Service Manual’s metrics guidance connects service purpose, goals, hypotheses and measurement. Its government-specific requirements do not apply automatically to a private business; the sequence is a useful reference for this original planning worksheet.
Fill in the experiment card
Visitor and task: Adults viewing an individual workshop page who need to know what to bring.
One proposed change: Replace and reposition the preparation-information link. Keep the underlying instructions unchanged during the observation unless an accuracy issue requires a correction.
Primary observation: Eligible recorded sessions that reach the preparation page through the workshop-page link, divided by eligible recorded workshop-page sessions. Count each session once in each part.
What that observation does not prove: A page visit does not establish that someone read, understood or followed the instructions. Add a small consented usability check if understanding is the actual concern.
Guardrails: The link remains usable by keyboard, has a clear name and does not hide the ordinary enquiry route. Incorrect preparation advice is corrected immediately rather than preserved for the sake of the experiment.
Review window: Choose and record actual start and review dates before release, with comparable operating periods where practical. The dates depend on the site’s real activity; this fictional example sets no traffic target.
Owner: Name the person who verifies measurement, the person who can restore the previous layout and the person who decides the next step.
This is a worked brief. Replace the fictional task with your real visitor need rather than treating the example event as a universal measure of success.
READER WORKBENCH / WEEK 45
Write the observation plan first
Define the visitor task, event, eligibility, dates and stop rule before looking for results. This builder neither measures traffic nor infers causality.
Fictional local exercise. No live AI, backend, business verification, analytics, accounts or paid services. Working state is kept in page memory; deliberate copies, prints and browser/device-managed data are outside this tool’s control. This is not a privacy guarantee, release approval, measured outcome or accessibility certification. Enter fictional, non-sensitive examples only.
Review pending · current working preview
Brief incomplete: resolve the named planning gaps
- Planned elapsed window: 14 days, not a sample-size or traffic target.
- One change: Rename Resources to What to bring and place it beside workshop details
- Primary observation: Unique eligible sessions with a successfully loaded preparation page through the workshop link / Unique recorded workshop-page sessions; exclude staff tests and duplicate events in both periods
- Limits: A page visit does not prove reading, understanding, attendance or causation.
- Stop and restore: Broken link, inaccessible control, misleading advice, failed form, or measurement defect; restore owner: unknown
- Sparse evidence: Report inconclusive; consider an observed task attempt rather than wait indefinitely
- No observations were collected. Dates do not establish adequate evidence or causality.
Unresolved warnings
- Unresolved: Trigger verification record
- Unresolved: Fictional measurement verifier
- Unresolved: Fictional decision owner
- Unresolved: Fictional restore owner
- The exact trigger and eligibility definition have not received a simulated verification.
- Precommitment review is pending.
Superseded evidence snapshots retained: 0. These are exported as historical records, never counted as current passes.
WEEK 45 — Write the observation plan first
Fictional local exercise. No live AI, backend, business verification, analytics, accounts or paid services. Working state is kept in page memory; deliberate copies, prints and browser/device-managed data are outside this tool’s control. This is not a privacy guarantee, release approval, measured outcome or accessibility certification.
Review pending: the current input preview has not been reviewed
Brief incomplete: resolve the named planning gaps
Planned elapsed window: 14 days, not a sample-size or traffic target.
One change: Rename Resources to What to bring and place it beside workshop details
Primary observation: Unique eligible sessions with a successfully loaded preparation page through the workshop link / Unique recorded workshop-page sessions; exclude staff tests and duplicate events in both periods
Limits: A page visit does not prove reading, understanding, attendance or causation.
Stop and restore: Broken link, inaccessible control, misleading advice, failed form, or measurement defect; restore owner: unknown
Sparse evidence: Report inconclusive; consider an observed task attempt rather than wait indefinitely
No observations were collected. Dates do not establish adequate evidence or causality.
UNRESOLVED WARNINGS
Unresolved: Trigger verification record
Unresolved: Fictional measurement verifier
Unresolved: Fictional decision owner
Unresolved: Fictional restore owner
The exact trigger and eligibility definition have not received a simulated verification.
Precommitment review is pending.
EXACT CURRENT INPUTS
{
"visitor": "Adults finding workshop preparation instructions",
"change": "Rename Resources to What to bring and place it beside workshop details",
"hypothesis": "Eligible visitors will more readily reach the preparation page",
"numerator": "Unique eligible sessions with a successfully loaded preparation page through the workshop link",
"denominator": "Unique recorded workshop-page sessions; exclude staff tests and duplicate events in both periods",
"limits": "A page visit does not prove reading, understanding, attendance or causation.",
"measurementOwner": "",
"measurement": "",
"measurementChecked": "no",
"start": "2027-08-09",
"review": "2027-08-23",
"decisionOwner": "",
"rollbackOwner": "",
"guardrails": "Keyboard-operable link, clear name, unchanged enquiry route, accurate preparation instructions",
"stop": "Broken link, inaccessible control, misleading advice, failed form, or measurement defect",
"confounders": "Email campaign, holiday closure and traffic-source changes need recording",
"inconclusive": "Report inconclusive; consider an observed task attempt rather than wait indefinitely",
"decisionRule": "Consider only if guardrails pass; interpret the defined observation with consent and measurement gaps visible",
"checked": "no",
"note": ""
}
HISTORICAL EVIDENCE SNAPSHOTS (not current; preserved failed and superseded attempts)
[]
External verification not performed: live sites, actual services, source refresh, production integration, real browser/mobile, screen readers, physical keyboard, actual clipboard permissions, printing/pagination.Use the text-only exercise
The complete unchanged manuscript remains readable if the controls are unavailable.
Return to Fill in the experiment cardDefine the event before counting it
Ask the implementer what evidence creates the numerator. Is it a link click, a successfully loaded destination page or another observable event? Name it precisely. A clicked link that leads to an error should not be described as a completed information-finding journey.
Check exclusions such as staff tests and duplicate events. Use the same eligibility rule in every period you plan to compare. Record any gaps created by the site’s approved analytics and consent arrangements. Do not add invasive tracking just to make a chart look complete.
For an enquiry experiment, define successful receipt separately from an attempted submission. Google Analytics’ recommended events illustrate how submitted, qualified and converted leads are different events. Verify the actual trigger in your own setup instead of relying on a familiar event name.
Decide what would make you stop
An observation window should not prevent you from responding to a broken page. Specify stop conditions such as an inaccessible control, misleading service information, a failed form or a measurement defect that makes the comparison unusable.
Also record foreseeable confounders: an email campaign, holiday closure, offer change, traffic-source shift or another page redesign. If one occurs, note it at the time. Do not remove inconvenient days after seeing the results without explaining why and preserving the original view.
A small site may reach the review date with little evidence. “Inconclusive” is a valid outcome. Decide whether another method, such as observing a visitor try the task, would answer the question more directly than waiting indefinitely for a large analytics sample.
Keep the decision rule practical
Before release, finish these sentences:
- We will consider the change usable only if these functional and accessibility checks pass: …
- We will describe improvement using this observation and these limits: …
- If the evidence is sparse or inconsistent, our next decision will be based on: …
- We will restore or revise the change immediately if: …
Do not invent a precise percentage threshold because the form has an empty field. If a decision needs statistical analysis or a controlled design, involve someone qualified to plan it with your actual traffic and constraints.
Review against the original question
Keep the dated brief, page versions and observation record together. At review time, compare what happened with the prewritten task and guardrails. Report raw counts as well as rates and explain important differences between periods.
Your finished card should let another person understand what was changed, why, what evidence matters and what would make the team change course. That is a useful starting point for a website improvement even when the final result remains uncertain.
Primary sources checked 4 October 2026
Sources & review
Primary sources checked on . The checklists and planning examples are AI-assisted editorial guidance, not source quotations or reported client results.
This AI-assisted guide uses fictional examples for practice. It does not report client results or establish that a live system will behave the same way.
Originally published: .