Property creators reporting dashboard: joins, gaps, limits

A property creators reporting dashboard should let a reviewer reproduce a result before asking whether it looks good. Build it as a controlled report for a defined England property set, not a collage of whatever each channel exports. Before anything is charted, name its working parts: the BI layer that renders the tiles, the channels that feed it and the property key that lets their rows meet.

What to take away

1. Fix the header

Show the decision owner, reporting period, Europe/London timezone, eligible properties and content IDs, channels, commercial model, data cut-off and refresh time. Add metric-contract version, query release and dashboard build ID. A viewer should see whether a chart covers editorial content, commissioned work, advertising, affiliate links or lead collection.

Name the channels and the tool in the header, so no reader has to guess what is being joined. That means the portals — Rightmove, Zoopla and OnTheMarket — the social and video accounts on Instagram, TikTok and YouTube, Google Analytics 4 for on-site sessions, Meta Ads Manager for paid reach and spend, and the BI layer that renders the tiles, whether that is Looker Studio, Power BI or Tableau.

Put data maturity beside the period. HM Land Registry says its Price Paid Data covers specified England-and-Wales sales lodged for registration, has exclusions and is not necessarily current. It must not appear as a live valuation or an England creator-performance measure.

2. Separate four report pages

The operations page covers release candidates, source refreshes, corrections and accessibility tests. The audience page contains source-labelled channel observations and consented journey events. The commercial page shows observed enquiries, valuation requests, viewings booked, instructions won and portal or paid-media spend without calling them impact. The evaluation page is reserved for causal estimates with their design and uncertainty.

Every tile displays event, population, numerator, denominator, unit, period, timezone, source system, export date, query or filter version, inclusions, exclusions, owner and uncertainty. Blank means unavailable; zero means the defined event was not observed. Never swap one for the other.

The events on these tiles are enquiries, leads, click-through rate, cost per lead, viewings booked and valuations requested, each labelled with the channel that produced it.

3. Make the data route visible

Draw the route from source export to cleaned table, joined record and displayed value. Reconcile row counts at every move. Show late events, duplicates, bots, internal tests, unmatched identifiers, deleted records and suppressed small groups. Store the query and data dictionary with the report version.

A worked join: portal enquiry exports from Rightmove, Zoopla and OnTheMarket carry a property reference and an enquiry timestamp, the CRM carries the lead, the viewing booked and the instruction that followed, Google Analytics 4 carries session and campaign data, and Meta Ads Manager carries spend and platform-reported conversions. Join portal and CRM rows on a stable property identifier such as a UPRN and on the enquiry or lead reference, and join sessions to campaigns on the UTM parameters. That route is what a reviewer reproduces.

The gaps live at those joins and belong on the page by name: enquiries arriving without UTM parameters, portal leads with a missing or duplicated property reference, sessions that never match a lead reference, small groups suppressed in an area report, and Price Paid Data rows that lag the transaction they describe. Show the count behind each one rather than dropping it.

The UK Statistics Authority's quality framework is aimed at statistics serving the public good, but its focus on suitable sources and methods is useful here. It does not confer official-statistics status on a private dashboard.

4. Label interpretation honestly

Prefix channel values with "platform observation" and retain that supplier's event definition, modelling and coverage. Label a rule-based join as "attributed outcome", then publish its window and tie-break rule in full: the window in days and the event it runs from, which source wins when a portal enquiry and a paid click land on the same lead, and whether a consent record is needed before the row counts at all. Use "causal estimate" only where a qualified evaluator has approved a counterfactual or another defensible design. The 2026 Magenta Book distinguishes monitoring from impact evaluation and requires attention to alternative explanations. State the limit plainly: a platform-reported conversion and a CRM-attributed instruction are different events, so the two figures will not agree, and neither is a property value or a completed sale.

Add a separate analyst note for editorial interpretation. Name the writer, list competing explanations and state the bounded decision. Do not turn attributed enquiries into property value, return or evidence of a completed transaction.

5. Protect the report

Map each field to purpose, lawful assessment, device storage or access, recipients and retention. ICO data-minimisation guidance supports collecting only what the purpose needs. Use grouped results where possible; raw enquiry text and home-security details do not belong in routine executive views.

Restrict access, log changes and prevent silent query edits. NCSC security logging guidance recommends choosing logs for defined incident questions and protecting them from inappropriate alteration. Retention still needs a project-specific privacy decision.

6. Add gates and actions

Display property, statistical, privacy and PECR, and security separately; a favourable chart cannot offset a failed gate. Display accessibility, rights, advertising and consumer, and FCA decisions separately.

End each page: owner action, due evidence, pause condition, next refresh trigger. Stop distribution if denominator changes without explanation, consent cannot be established, a join cannot be reproduced or source definition has changed.

Before you act

Common questions

How should a dashboard treat HM Land Registry Price Paid Data?

It must not appear as a live valuation or an England creator-performance measure. The data covers specified England-and-Wales sales lodged for registration, has exclusions and is not necessarily current, so it needs a data maturity label beside the period.

When can a value be called a causal estimate?

Only where a qualified evaluator has approved a counterfactual or another defensible design. The 2026 Magenta Book distinguishes monitoring from impact evaluation and requires attention to alternative explanations, so attributed enquiries stay separate from causal claims.

What should happen if the denominator changes without explanation?

Stop distribution. The article lists that as a pause condition alongside consent that cannot be established, a join that cannot be reproduced and a source definition that has changed. Each page should end with owner action, due evidence, pause condition and next refresh trigger.