Reputation metrics with review cohorts and reply clocks defined properly

Reputation management key metrics should answer a specific operational question: what arrived, what received a response, and what happened afterwards? For an England-based business, define the branch population before calculating anything. A national profile cannot supply an England-only result unless the underlying records can be separated on a defensible basis.

The definitions below are proposed reporting conventions. They are not findings about any business, performance targets or claims that software improves customer behaviour. No England benchmark or target is published in this article. Every figure used in a calculation comes from the operator's own export and branch register.

What to take away

Reputation-management software that holds these records

Review records arrive either from a review platform or from reputation-management software that collects several platforms into one register. Google Business Profile, Trustpilot and Reviews.io publish reviews against a single profile, while Birdeye, Podium, Reputation and Yext are multi-location platforms that consolidate reviews, replies and branch details into one account.

In software, a cohort is a saved view. The tool stores a review identifier, a branch, a rating, a first observation time, a publication time and a reply timestamp against each review, and the saved view selects the rows that share those attributes. Build the cohort in the tool before exporting it, because a cohort assembled later in a spreadsheet will not match the cohort the software used for its own counts.

Reply clocks are configured in the same place, usually as a first-response SLA or a resolution SLA against the review record. Note whether the clock starts at first collection or at publication, and whether it pauses outside service hours. A clock that starts at first collection measures a different quantity from one that starts at publication, and the two should not share a heading.

Define the reporting population and period

Write an approved branch register with stable identifiers before collecting records. Decide whether the report covers every England location, a defined subset or one profile. Keep the dates on which a branch enters or leaves the population. Annotate a coverage change because the total can move independently of service performance.

Keep a UK-wide account separate when its records cannot be allocated to England. Do not infer a reviewer's residence from the branch they mention. Branch, customer and platform audience geography differ, and some may stay unknown.

Review registers hold personal data. In the UK, the Information Commissioner's Office enforces the UK GDPR and the Data Protection Act 2018, so an export that carries a reviewer's name, photograph or profile link needs a lawful basis, a retention period and a stated list of people who may read the free-text fields. On authenticity, the Competition and Markets Authority holds powers over fake and misleading reviews under the Digital Markets, Competition and Consumers Act 2024, which is one reason to record an edit, deletion or restored record as an event rather than overwriting the original.

Specify the reporting period and the extraction cut-off. A month of publication events differs from a month of first observations if collection started late. Preserve both timestamps where the export supplies them. If publication time is missing, put that limit in the metric name.

The review export should carry a platform review identifier, branch identifier, first observation time, publication time where available, rating and reply time. Google's performance record names calls, website clicks and direction requests as fields.

Count review events without confusing the total

Create a review register with a platform identifier, branch identifier, first observation time and any available publication time. Retain a separate event type for an edit, deletion or restored record. A revised comment should not automatically become another new review.

Name the cohorts before the first calculation. A workable taxonomy separates new reviews observed in the period from cumulative stock visible at the cut-off, and then splits by branch, by platform, by rating band and by service line. Mark which cohorts are eligible for the response measure and which are excluded, and keep the taxonomy in one list so that two reports cannot invent two definitions of the same cohort.

Report new reviews observed during the period alongside the stock visible at the cut-off. These quantities answer different questions. Reconcile the opening stock, additions, removals and closing stock where the export permits it. Put unexplained differences in a reconciliation field instead of silently forcing the totals to agree.

For ratings, keep the platform's scale and definition. Google's review-score guidance calls its score the average of published ratings. Updates can take up to two weeks.

This is Google's global product definition, not an England market statistic. Store the displayed score with its capture date. Do not swap it for a differently filtered calculation under the same label.

Measure replies against an eligible cohort

Define response coverage as eligible reviews with an observed public reply divided by all eligible reviews in that cohort, expressed as a percentage. Choose eligibility before inspecting branch results. Record legitimate exclusions and their counts so a manager can reproduce the denominator.

Measure elapsed time in hours between the selected starting event and the first observed reply. If the start is first collection, label the result as collection-to-reply time. It cannot establish the delay from original publication when that timestamp is absent.

Show the number still unanswered and their age at extraction next to the median elapsed time for answered records. Otherwise the oldest waiting cases disappear from the timing measure. For a service-hours calculation, publish the calendar and distinguish it from elapsed clock hours. Neither convention supplies an industry target by itself.

Reply clocks also need names. A first-response SLA measures the time to the first public reply; a resolution SLA measures the time to a recorded outcome such as a resolved complaint. Publish the start event, the service calendar and the target for each, and never compare a first-response figure with a resolution figure under the same heading.

Worked example with assumed inputs, not England data. A sample England branch holds 40 reviews eligible for the period. Thirty carry an observed public reply. Response coverage is 30 divided by 40, or 75 per cent.

If the middle pair of 30 ordered reply times read 6 and 10 hours, the median reply time is 8 hours. The 10 reviews without a reply stay unanswered with their age at extraction.

Keep interaction and outcome measures apart

Google's Business Profile performance record defines calls as call-button clicks and website clicks as clicks on the profile's website link. Use those labels literally. A call-button event does not document an answered conversation, and a website click does not document a sale.

Choose an outcome from a separate operational record, such as a completed booking or an explicitly recorded complaint resolution. Specify whether cancellations, reopened cases and repeat events count. Only join that record to a review when the matching basis is documented and appropriate; keep unmatched cases visible.

Keep an event register with source, record identifier, event type, time and branch. Publication, amendment, public response and recorded complaint resolution stay distinct. An event can link to another event without becoming the same thing.

Text categories need their own controls. Write a coding guide for topics and sentiment, retain an uncertain category, and check a sample against that guide. A rating and a text judgement can disagree. Report the disagreement rather than overwriting one with the other.

Survey metrics belong beside the review measures, not inside them. Net Promoter Score and customer satisfaction (CSAT) come from invited respondents, while response coverage and reply time come from published reviews; the populations differ and the two should not be averaged together. Sentiment score is a text measure over the reviews themselves, and share of voice is a volume measure: a branch's or brand's share of the reviews observed in a defined market cohort. Label each measure and keep it separate.

Make the metric dictionary usable

Each dictionary entry needs its purpose, source field, unit, population, period, calculation and missing-data treatment. Add an owner who can explain changes to the extraction. Preserve earlier definitions when a field changes so a trend does not imply continuous comparability.

State whether the question asks for attribution credit or causal impact before analysis. Attribution credit assigns an observed booking to a review or reply under a stated matching rule. Causal impact needs an evaluation design.

HM Treasury's Magenta Book guidance, updated in May 2026, recommends incorporating evaluation into intervention design. Apply that principle when an operational question becomes a question about impact: specify the intended outcome before commissioning a causal study.

At the next reporting review, select one disputed metric and trace it back to its records. Resolve the denominator, clock and exclusions first. A clearly defined measure that remains incomplete is more usable than an apparently precise total assembled from incompatible events.

Before you act

• Freeze event definitions and observation windows before comparing periods.

Common questions

How should new reviews be counted without confusing them with the total?

Create a review register with platform and branch identifiers, first observation time and any publication time. Retain a separate event type for edits, deletions or restored records. Report new reviews observed during the period alongside the stock visible at the cut-off, and reconcile opening stock, additions, removals and closing stock where the export permits.

What does a call-button click actually document?

Google's Business Profile performance record defines calls as call-button clicks and website clicks as clicks on the profile's website link. Use those labels literally. A call-button event does not document an answered conversation, and a website click does not document a sale, so choose outcomes from a separate operational record.

Why preserve earlier metric definitions when a field changes?

Each dictionary entry needs its purpose, source field, unit, population, period, calculation and missing-data treatment. Preserve earlier definitions when a field changes so a trend does not imply continuous comparability. A clearly defined measure that remains incomplete is more usable than an apparently precise total assembled from incompatible events.