Media buying agencies measurement: a practical guide

This guide covers media buying agencies measurement, starting with the decisions the evidence must support. Media measurement should organise decisions. Before an agency buys anything, the client and agency should agree what will be observed, where each record comes from, how credit will be assigned and which result would change the plan. Reporting then documents delivery and evidence. It does not convert platform attribution into proof that advertising caused a commercial outcome.

This guide is for organisations running campaigns in England. UK regulator guidance is labelled UK. Great Britain audience sources include England, Scotland and Wales, while global platform documentation describes product mechanics. None of those boundaries should be rewritten to make a result appear more local. There is no England-only advertising regulator: the Advertising Standards Authority applies the Committee of Advertising Practice (CAP) Code across the UK, which is why the UK rules govern England campaigns.

Which media buying agencies this guide covers

UK media buying is concentrated in a few large groups, among them GroupM, Publicis Media, Omnicom Media Group, Dentsu and Havas Media, together with independent agencies such as the7stars. Each agency runs its own measurement stack, so the checks below are written to be applied agency by agency rather than to a group or network as a whole. Where an agency plans and buys inside a media-management system such as Mediaocean's Prisma, the plan, buy and invoice records held there form part of the evidence trail.

What to take away

Begin with a measurement charter

Put the commercial decision at the top of a one-page charter. "Increase sales" is too loose. Name the accepted outcome, the group or place concerned, the period and the decision owner. Then separate four questions:

The answers will often come from different systems. That is normal. Trouble starts when a team labels all four as performance and expects one number to settle them.

The charter should also list exclusions. These may include cancelled orders, test events, existing customers, activity outside England or costs outside the agreed media scope. An exclusion is part of the definition, not a footnote to add after results arrive.

Worked example. For a retailer reviewing a paid-search budget line, the charter reads: decision — renew or cut the budget line; accepted outcome — first-time customer orders recorded in the client's order system, net of returns; group or place — new customers ordering to an England delivery address; period — the agreed trading quarter; owner — the client's ecommerce lead; exclusions — staff and test orders, resellers, existing customers and activity outside England; sources — the platform delivery export for spend and delivery, the analytics platform for observed sessions and the client's order system for accepted outcomes; review — monthly, with an extra review whenever the budget decision changes.

Give every metric a complete definition

A metric is ready only when another competent reader can reproduce it. Record the name, business question, numerator, denominator, unit, source field, time zone, attribution window, geography, exclusions and reporting lag. Add an owner and the action attached to the result.

Filled in, a metric card reads: metric — cost per accepted enquiry; business question — what one qualified enquiry costs in media; numerator — media cost authorised in the plan; denominator — enquiries accepted under the client's qualification rule; unit — pounds per accepted enquiry; source fields — the plan cost line and the client's enquiry table; time zone — Europe/London; attribution window — the agreed enquiry window; geography — England; exclusions — duplicate, spam and test enquiries; reporting lag — the client's qualification period; owner — the agency's measurement lead; action — reallocate or pause the line when the figure moves past the threshold agreed in the charter.

Do not use a rate without both components. Cost per accepted enquiry, for example, needs the qualifying enquiry rule and an agreed statement of cost. Media spend alone and a fully loaded campaign cost will produce different answers. Neither is inherently wrong, but the label must show which one was used.

Keep forecasts apart from observed delivery. Google's global Reach Planner documentation describes forecasts for supported video campaign settings. That makes it first-party evidence about the planning product, not a promise of delivery or an independent audience result.

Select evidence by medium and question

Audience currencies are medium-specific because their populations, collection methods and outputs differ. Barb explains its UK television measurement and the roles of panel and census-level data.

Route describes Great Britain out-of-home audience research. Each can inform planning and evaluation within its stated scope. Neither establishes that an individual advertisement caused a purchase.

For other UK media, RAJAR measures radio audiences; PAMCo measures published-media audiences, and Ipsos iris is the UKOM-endorsed online audience measurement service. JICMAIL provides audience measurement for mail. Check each source's scope and methodology before comparing it with platform or client records. Ofcom's media plurality measurement framework names comScore as the web measurement provider used by UKOM, the joint industry committee for online measurement in the UK.

Beyond the industry currencies, commercial suppliers active in the UK market include Kantar and Nielsen, whose audience and campaign measurement appears in tenders and audits. Read their scope statements on the same terms as a currency: population, method and coverage first.

Digital platform records have another purpose. They help buyers understand activity processed by the account and product. Google's global conversion tracking overview distinguishes website, app, phone and offline conversion routes within Google Ads. If an agency reports one of those totals, retain the conversion action, configuration and extraction date. Call the result Google Ads-attributed when that is what it is.

For LinkedIn activity, use its first-party Campaign Manager report export instructions to record the chosen report type and breakdown. Do not merge its total with another platform merely because both columns say conversions. Reconciliation needs a shared outcome definition and a documented identity rule.

Client systems should govern client-accepted outcomes. The sales ledger, booking system or customer database may hold cancellations, qualification stages and revenue the buying platform does not know.

Where a permitted, controlled route exists, preserve the link between the media record and client outcome. Otherwise report the sources alongside one another and state the unmatched portion.

Design collection before campaign build

Write an event specification for each digital action. It should say what fires the event, on which page or system state, how duplicates are handled and what test proves the setup works. Use test records that cannot be mistaken for genuine customers. Assign someone other than the implementer to check material events before launch.

Campaign naming matters because it joins plans, platform exports and invoices. Use fields for client, objective, market, channel, campaign, version and date only where the team can apply them consistently. A shorter controlled convention is safer than an elaborate code that buyers regularly improvise.

Record all material configuration changes. If an event, lookback period, audience rule or import process changes on Tuesday, the weekly total may contain two measurement regimes. Mark the break rather than drawing one smooth trend across it.

Online measurement also needs a privacy decision. The UK ICO guidance on online advertising covers storage and access technologies used for advertising selection, delivery, tracking and profiling. Relevant rules include the UK General Data Protection Regulation (UK GDPR) and the Privacy and Electronic Communications (EC Directive) Regulations 2003 (PECR). A qualified UK reviewer should apply the current guidance and rules to the actual technologies, parties and consent journey. The availability of a platform feature does not establish that a proposed implementation complies.

Test the trail, not just the event

A launch check should follow representative records through the whole route. Confirm the approved creative and destination, trigger the intended event, inspect the platform record, verify any transfer and find the result in the report. Note time lags so the team does not diagnose a delayed import as zero response.

Use four status values for control results: passed, failed, not applicable and unresolved. Do not average them into a quality percentage. One unresolved event that governs the main commercial decision can matter more than many passed formatting checks.

After launch, reconcile at the level needed to locate faults. Media cost can be checked against the authorised plan, platform export and invoice scope. Response totals can be compared by event, day and campaign identifier. Client outcomes can be tested against acceptance rules. Investigate unexpected gaps before optimising the campaign.

Choose attribution language deliberately

Attribution allocates credit under a method. It does not, by itself, identify what would have happened without the advertising. The report should name the method and use language that matches it.

Platform attribution is useful for decisions inside that platform, subject to its configuration and coverage. Named credit-allocation models include last-click, first-click, linear, position-based and time-decay; data-driven attribution uses a model rather than a fixed allocation rule. Models assign credit to eligible contacts but do not, by themselves, establish incremental impact. For causal evaluation, consider a randomised holdout or geo-holdout incrementality test. Marketing mix modelling (MMM) is another aggregate evaluation approach; named tools include Google's open-source Meridian and Meta's open-source Robyn. Choose a method and tool based on the available data, study design and qualified analytical input.

Write a method card that states the eligible media, outcome, lookback period, identity approach, exclusions and treatment of duplicate claims. Keep the earlier method when a rule changes. If two reasonable approaches lead to different budget decisions, show the sensitivity rather than selecting whichever result looks better.

Build a report around exceptions and actions

The main reporting page needs a clear header: campaign scope, reporting period, data cut-off, currency, geography and plan version. Beneath it, keep delivery, audience, response and commercial evidence in separate sections. Each visual should answer one question and carry its source.

Add a compact data-quality panel. It can show delayed sources, failed transfers, unreconciled costs, duplicate-event investigations and definition changes. Use dates and owners. "Data issue" is not enough for someone to decide whether a result can be used.

Finish with the decision record. The minimum entry contains the observation, interpretation, decision, owner and due date. Where uncertainty affects the choice, describe it in ordinary language. A precise-looking decimal does not repair a weak sample or incomplete tracking.

Worked example. A single report entry might read: observation — delivered impressions ran ahead of the agreed profile while accepted enquiries stayed flat; interpretation — the extra delivery reached contacts outside the qualification rule; decision — hold the budget and recheck the audience settings; owner — the agency's trading lead; due date — the next weekly review. The entry names an action and an owner, so the reader can tell whether the point was closed.

The reporting pack should retain source exports, approved definitions, calculation files and correction notes under the organisation's access and retention rules. A client should not depend on a screenshot when the underlying extract can be provided safely.

Divide responsibility without losing accountability

Small teams may give one person several roles. The record should still show which responsibility they exercised. The agency should not approve its own unresolved discrepancy when that discrepancy affects how the client judges the service.

Contract and audit rights deserve attention before data is needed. The UK advertiser guidance in the ISBA agency contract compliance audit guidelines discusses contract compliance, remuneration and stewardship checks. It is industry guidance rather than legal advice. Media audit and consultancy firms such as Ebiquity advise advertisers on agency contract compliance and remuneration, and the audit right written into the agreement is what lets the client's chosen adviser in. Obtain qualified advice for the actual agreement, including access to accounts, exports, invoices and subcontractor records.

Run reporting at more than one speed

Launch monitoring looks for broken settings, destinations and events. Trading reviews consider delivery and controlled changes. Periodic measurement reviews examine definitions, data quality, attribution choices and commercial reconciliation. Contract audits address whether records and charges match agreed terms.

Set cadence according to decision risk and reporting lag. A feed that updates frequently does not require a decision every time it moves. Conversely, a slow commercial outcome should not stop the team checking delivery faults early. Write the trigger for an extra review, such as a material scope change, unexplained variance or definition break.

Recheck every platform feature, regulator page and contract term at the point of use. Continue to label sources, show changes and keep claims within the evidence.

Questions for an agency review

Ask the agency to demonstrate the reporting trail with a fictional or suitably controlled record. The exercise should show who defines an event, who checks it, how an export reaches the report and how a correction is approved.

Then settle these points in writing:

The answers are service and governance commitments, not performance benchmarks. Do not infer quality from a confident demonstration or a polished dashboard.

Put the guide into operation

Take one live or planned campaign and create its charter, metric cards, source schedule and decision log. Trace one delivery figure, one attributed response and one client outcome. Record every mismatch without forcing agreement.

The setup is ready when another authorised reader can reproduce the scope, locate the evidence and explain why the next decision follows. Have qualified reviewers check relevant UK legal or privacy points, and reopen time-sensitive sources before use.

Before you act

Common questions

What should a measurement charter include?

A one-page charter should name the accepted outcome, the group or place concerned, the period and the decision owner. It should also list exclusions such as cancelled orders, test events, existing customers or activity outside England. The charter separates four questions: did the media run, what exposure occurred, which responses were observed or attributed, and what commercial outcome was accepted.

How should platform attribution be described in reports?

Platform attribution suits decisions inside that platform, within its configuration and coverage. Name the method and match the language, such as Google Ads-attributed when that is what it is.

Do not merge totals from different platforms just because both columns say conversions. For causal claims, plan an experiment or other evaluation with qualified analytical input.

What does a launch check involve?

A launch check should follow representative records through the whole route. Confirm the approved creative and destination, trigger the intended event, inspect the platform record, verify any transfer and find the result in the report. Note time lags so a delayed import is not diagnosed as zero response. Use status values passed, failed, not applicable and unresolved.