The 20-Minute Business Profile Experiment

Minutes 0–4: write the decision question before touching the profile

Choose one question a customer appears unable to resolve. Examples include: Can people recognize the correct entrance? Do they understand that appointments are required? Can they see the principal service environment? Is the website path obvious? Phrase the question from the customer’s side, then write a narrow hypothesis: “A current entrance photo may reduce direction-related confusion,” not “New photos will boost SEO.”

Name one primary observation before making the change. It may be directions requests, website clicks, bookings, calls, or a repeated front-desk question, depending on what Google reports and what the operation can record consistently. Google makes clear that only applicable performance metrics appear. Choosing the measure in advance reduces the temptation to search afterward for whichever number rose.¹

Also write a stop condition. Do not experiment with facts that must simply be correct, such as the legal business name, actual address, safety information, or operating hours. Accuracy is maintenance, not a treatment arm. Test presentation and helpfulness only within Google’s representation and content rules.²

Minutes 4–8: freeze a baseline and its context

Capture the current live state with dated screenshots and record the comparison window you plan to use. In the performance view, Google allows a date range and defines views, searches, and interactions separately. A profile view counts unique visitors under Google’s stated rules, calls count clicks on the call button rather than completed conversations, and directions include Google’s handling of repeated taps, cancellations, and spam. Preserve those definitions in the log.¹

Add a short context line: paid campaign running or paused, special hours, major event, promotion, price change, unusual closure, review surge, website outage, or strong seasonal pattern. Google says performance data can include activity from organic results and Google Ads. A change in the total may therefore reflect media you did not alter.¹

Use a comparable period long enough to contain the normal rhythm of the business. Do not prescribe a universal number of days: a high-volume café and an appointment-only specialist generate evidence at different speeds. Decide the window before launch, include comparable weekdays, and extend it when counts are too sparse to interpret responsibly.

Minutes 8–12: make one small, truthful change

Change one presentation element tied directly to the question: replace an obsolete entrance image, select a more representative cover, add one missing service accurately, clarify a permitted attribute, or correct the website destination. Record the exact old and new states. If the change requires new photography, use this session to specify the shot and schedule it rather than uploading a weak substitute just to finish the timer.

The one-change rule is a practical constraint, not a complete experimental design. NIST’s handbook explains that rigorous design of experiments uses systematic data collection, and it warns that one-factor-at-a-time approaches can miss interactions between variables. A local business usually cannot randomize Google’s interface or hold demand constant. Limiting the change still improves traceability, but it does not remove confounding.^4,5

If several urgent inaccuracies exist, fix all of them immediately and label the period “maintenance,” not “experiment.” Learning is secondary to truthful customer information. Begin a cleaner observation only after the profile is stable enough that one deliberate change can be described honestly.

Minutes 12–15: verify the release state

Check whether the edit is saved, pending, rejected, or live. Google reviews uploaded media and notes that photos and videos may take time to appear. A performance window should begin when the customer-facing change is actually visible, not when the file left the desktop. Save the live-verification timestamp separately from the upload timestamp.³

Inspect from Search and Maps on a phone. For photo tests, note that the preferred cover is not guaranteed to appear first. If the changed image is live but rarely surfaced in the route you are testing, the experiment did not receive the exposure you assumed. Record that as a delivery limitation rather than declaring the creative ineffective.³

Minutes 15–20: schedule the read and define the verdict

Put the read date on the calendar and write four possible verdicts now. Keep means the cue is accurate, useful, and the evidence is directionally favorable or operationally clearer. Revise means the question remains but the asset is weak. Revert means the new state is misleading, less useful, or operationally harmful. Inconclusive means delivery was uncertain, volume was too low, or outside events overwhelmed interpretation.

At the read, compare the chosen metric and the qualitative observation with the baseline and context log. Google’s search-term data updates monthly and may lag at the start of the month, so do not treat a temporarily empty query view as immediate evidence. Downloadable performance data can support a longer series, especially for teams managing multiple locations, but more rows do not cure a poor comparison.¹

Report what changed in plain language: “After the entrance photo went live, direction requests were similar and staff recorded fewer wrong-door calls during the observation window.” Then state the limits. Do not say the photo caused the operational change, improved ranking, or will repeat elsewhere. Google identifies relevance, distance, and prominence as the main local-ranking factors; the twenty-minute routine is for disciplined learning about usefulness, not reverse-engineering the algorithm.²

• Question and hypothesis
• Primary observation and comparison window
• Before state, after state, and live timestamp
• Promotions, ads, closures, events, and other obvious confounders
• Verdict: keep, revise, revert, or inconclusive