Specifying a media buying dashboard that finishes on actions

A media buying dashboard for a UK advertiser ends in decisions, not decoration. Specify it so every panel says what happened, how the figure was produced and which action is due. No figure belongs there without a definition, a source date and an owner.

What to take away

Fix the header firstperiod, cut-off, scope, geography, currency, freshness by source and plan version.
Build four panels (delivery, audience, response, commercial) and give each one decision, one owner and one date.
Document every measure: formula, unit, source field, attribution window, filters, exclusions, owner and version.
Label data provisional, complete, corrected or unavailable, and list open discrepancies with a response date.
Close with a decision log holding the observation, decision, owner, due date and the evidence needed to close it.

Build the header before the charts

Put the reporting period, data cut-off, campaign scope, currency and geography at the top. Show freshness per source, because one feed can be live while another waits on an offline import. Name the approved plan version, so a spend variance cannot compare live delivery with a superseded budget.

Header essentials before charts

Reporting period and data cut-off
Campaign scope and currency
GeographyEngland, UK or other
Freshness shown by source
Approved plan version reference

Deciding which decisions the dashboard must support starts with the problems the agency is contracted to solve, which where agencies solve real advertiser problems sets out.

Four panels, four decisions

Required content

Delivery: Spend, booked or served activity, pacing, scope changes
Audience: Medium-specific exposure measure, method, geography
Response: Named conversion action, attributed total, data lag
Commercial: Client-accepted outcomes, agreed cost scope

Decision it ends in

Delivery: Continue, investigate or cap delivery
Audience: Change reach or placement
Response: Inspect the journey or change activation
Commercial: Review value against the objective

Keep platform totals and client-accepted totals in separate fields even when the labels look alike. Google's overview of conversion tracking lists website, app, phone and offline actions, so name the action each panel displays and never relabel a platform-attributed total as a verified sale.

Record the export type, chosen columns and extraction time behind any LinkedIn figure, as the Campaign Manager export guidance describes. Keep LinkedIn and Google Ads totals in separate rows unless a documented method reconciles them.

Example: one filled decision log entry

Observation (illustrative): spend reached 118 per cent of the June plan by 14 June, while cost per accepted lead rose 22 per cent week on week. Decision: hold the daily cap and move 15 per cent of remaining budget to the two cheapest audiences. Owner: agency media manager. Due: 18 June. Evidence to close: the revised pacing line.

Document each measure in a data dictionary

Every displayed measure needs an entry: formula, unit, source field, attribution window, filters, exclusions, owner and version. Version the entry when an element changes, and do not rewrite past values under a new rule without flagging the break. Matching a window to the data is the subject of our guide to attribution methods.

Data dictionary entry fields

Formula and unit
Source field
Attribution window
Filters and exclusions
Owner and version

Document each measure

Measure: Accepted lead, client-held
Formula: Count of deduplicated enquiries with a confirmed sales appointment
Source field: CRM lead_status
Unit: Count
Window: 30 days from first click
Filters: England, paid search and paid social
Exclusions: Internal test leads
Owner: Client insight analyst
Version: 2025-06-01

A reader who selects England must know whether the filter uses campaign targeting, delivery location, billing address or CRM address, because a UK-wide record must not be turned into an England result.

Show data quality and reconciliation

Label each panel provisional, complete, corrected or unavailable. Keep a reconciliation block for unresolved discrepancies, listing the source pair, the size of the gap, the owner and the expected response date. Zero means the system observed none under the definition, never that a feed failed.

Comparing your cost per accepted lead with outside figures needs benchmarks that state their sources, which our benchmark research notes cover.

For online measurement, ICO guidance on storage and access technologies explains consent duties for tools used in ad selection, delivery, tracking and profiling. Record the consent basis each tag depends on, so a thin segment can be explained rather than estimated.

Before you release the dashboard

Header carries period, cut-off, scope, geography, currency, freshness and plan version.
Every panel names its decision, owner and due date.
Every measure has a dictionary entry with formula, source field, window and owner.
Someone outside the reporting team has traced one chart to its source export.

Common questions

What must the header of a media buying dashboard contain?

The reporting period, data cut-off, campaign scope, currency and geography, plus freshness by source and the approved plan version. Without it, a variance compares live delivery with a superseded budget.

How should platform results and client results be kept apart?

Hold them in separate fields and separate source rows, even when the labels sound alike. Name the conversion action each panel displays, so no platform-attributed total passes as a verified sale.

What should a decision log contain?

One entry per decision: the observation, the decision taken, the owner, the due date and the evidence needed to close it. Keep the earlier snapshot when a correction changes a material conclusion.

How do I show that a number is unreliable?

Use the labels provisional, complete, corrected and unavailable on the panel itself. Add the gap to the reconciliation block with an owner and a response date.