Measurement
Continue, investigate, stop: a landing page dashboard rule
Write the continue, investigate and stop rule for a landing page dashboard tile, with one worked example and what each funnel stage must also state.
A landing page dashboard tile earns its colour only after the rule for that colour exists. Green or red before then is decoration, and it invites a decision the evidence cannot support. The rules below cover each stage of the lead-capture funnel and decide whether the report can be released at all.
What to take away
- No tile turns green or red until its threshold rule exists.
- Each of the three thresholds needs a metric version, a minimum evidence condition and an action.
- Each funnel stage, from form submission to revenue, needs its own rule.
- The buyer supplies every figure, because this design holds no benchmark thresholds.
- A missing metric card, correction route or uncertainty note means a hold notice, not a colour.
Fix the decision before the tiles
At the top, state the decision, decision owner, evidence period, Europe/London timezone and earliest action date. These five facts tell a reader what the page is for, and when anyone may act on it, before a single number appears. Then add three blank fields:
Decision strip before any colour
Do thresholds exist?
tiles may show colour
leave tiles blank
- continue threshold
________;- investigate threshold
________;- stop or correct threshold
________.
A filled strip might read: Decision: review paid search spend; Owner: marketing lead; Evidence period: 1 to 30 June 2025; Timezone: Europe/London; Earliest action date: 5 July 2025. The three thresholds stay blank until someone with authority supplies them.
What the rule for each funnel stage must state
A landing page report mixes stages that count different things. Give each stage its own tile and its own rule, and never let one stage's colour stand in for another. The table names what each tile counts and what its rule must also state.
| Funnel stage | What the tile counts | The rule must also state |
|---|---|---|
| Form submission | Accepted submissions out of sessions on the landing page | Which events count as accepted, with consent-blocked sessions shown as unmatched |
| Qualified lead | Leads that meet the buyer's qualification | The qualification rule and its version |
| Recorded transaction | Transactions in the customer record | The join to the submission, with an unattributed row where it failed |
| Finance-approved revenue | Revenue approved in the finance ledger | The approver and the approval date |
| Platform-attributed conversions | Credit assigned by an advertising platform | Which model produced it, never summed with observed transactions |
A worked rule for one tile: accepted form submissions
The first row of the table shows the shape, and the same three states apply to every stage. The figures stay blank because they belong to the buyer, and each one needs the metric version and evidence condition beside it.
- Continue:the card is the current version, the evidence condition is met and the reading is at or above [buyer figure]. Action: record the reading and carry on.
- Investigate:the evidence condition is met but the reading falls between [buyer figure] and [buyer figure], or data health shows missing counts, unmatched counts or delays. Action: a named owner finds the cause before the earliest action date.
- Stop or correct:the reading is below [buyer figure], or the definition changed without a new version. Action: hold the spend decision and correct the report first.
Where no rule has been agreed, leave the tile neutral and say so on the tile. A rule copied from a published benchmark brings a comparability problem of its own, and the benchmark comparability test shows how to check it first.
Who supplies the numbers
The design is blank on purpose. It holds no results and no benchmark thresholds, so the buyer must set each one. A named statistician, a measurement owner, a privacy and PECR reviewer and a finance professional must approve the completed report.
Approval is a gate, not a courtesy. If any of the four roles is unnamed, the report stays in draft and every tile keeps its neutral colour. Naming the roles before the first draft avoids a late search for a signer.
Tie every threshold to a definition
The visible label of each tile links to a card holding the exact event and eligible population, numerator, denominator or unit, period, timezone, source, exclusions, deduplication, missing-data treatment, uncertainty and owner. A change to the definition creates a new version.
A threshold written against version one does not carry over to version two. Treat the old and new series as separate, and say so beside the tile.
Corrections that move a threshold result
The correction panel records the affected metric and period, old value, corrected value, reason, approver, timestamp and decisions already made. Preserve the original report. If a correction changes a threshold result, name the person responsible for reconsidering the action.
A correction that changes nothing a reader acts on still goes in the panel. Silent overwrites are one of the measurement mistakes that leave earlier decisions standing on a value that no longer exists.
Put an uncertainty note beside the decision metric
Add a plain-language note beside every decision metric. It should cover sampling where relevant, missing records, classification error, delayed events, model dependence and external changes. Government Analysis Function guidance explains how to communicate quality, uncertainty and change for different types of statistics.
Where a tile depends on platform credit, say which model produced it, using the attribution methods compared. Do not sum platform credit and observed transactions in one tile.
Hold the report instead of colouring it
Release is a short sequence, and any failed step leads to a hold:
- Run source totals against the extracted data.
- Reproduce the report from the locked inputs.
- Peer-review the query and test access controls.
- Record the output checksum and the approval.
- Confirm every primary metric has its card, correction route and uncertainty note.
If any primary metric lacks one of those, publish a hold notice rather than a decision colour. The Government Analysis Function's Reproducible Analytical Pipelines strategy says good peer review and audit rely on reproducibility, which is why step two comes before the review.
Before you act
- Fill the decision strip before any tile.
- Write one rule for each funnel stage you report.
- Leave every threshold figure blank until the buyer supplies it.
- Link every tile to a versioned definition.
- Add a correction route and an uncertainty note for each decision metric.
- Publish a hold notice if any card, route or note is missing.
Common questions
When can a tile be coloured green or red?
Only after its threshold rule exists: a metric version, a minimum evidence condition and an action. Until then the tile stays neutral.
Who sets the thresholds?
The buyer. The dashboard design holds no results or benchmark thresholds, and a named statistician, measurement owner, privacy and PECR reviewer and finance professional approve the finished report.
What happens if a correction changes a threshold result?
The panel records the old and corrected values, the reason and the approver, and a named person reconsiders the action. The original report is preserved.
