How marketing agencies measure campaign performance

What to take away

Most measurement programmes fail at the contract stage, not the analytics stage. A client asks for 'more reporting', an agency adds a dashboard, and nobody agrees what a good month looks like. Six months later the review meeting argues about numbers that were never defined.

This guide sets out how to avoid that. It covers baselines, metric choice, attribution, tracking rules, reporting format and the evidence that makes an agency's claims credible. It is written for buyers and for agency-side people who have to defend the numbers. On the agency side, a measurement lead or analyst typically turns the brief into a measurement plan, checks tracking and data quality, and maintains the reporting view; channel specialists interpret platform results, while the account lead explains decisions and actions to the client.

Agree the question before the metric

Every measurement plan should answer one question: did the work change something that matters to the business? That framing forces you to name the business outcome first, then work backwards.

A useful discipline is to write the question down in one sentence before any tool is discussed. For example: 'Did the paid search programme increase qualified enquiries per pound spent compared with the same quarter last year?' That sentence tells you the metric, the comparison and the time frame.

Choose the comparison period deliberately. A month measured against the previous month is distorted by seasonality, holiday periods and stock cycles. A year-on-year comparison removes most of that noise, though it assumes the market has not shifted underneath you.

If you cannot write that sentence, you are not ready to buy media or build a dashboard. Our key metrics guide covers the individual measures in more detail.

Capture a baseline before work starts

A baseline is the state of play before the agency touches anything. It should cover traffic, conversion rate, cost per acquisition, lead quality and any offline measure you can access. Record the date range and the source system for each figure.

Baselines get messy when tracking changes mid-period. If you install new consent management or switch analytics platforms, say so, because a step change in reported conversions is usually a tooling artefact rather than a performance jump. A common web and paid-media stack might include Google Analytics 4 (GA4) for site events and conversions, Google Ads and Meta Ads Manager for platform delivery and spend, Looker Studio for dashboards, and the client's customer relationship management (CRM) system to validate lead quality and sales outcomes. These sources are not interchangeable: record which one supports each figure and reconcile their definitions.

For a worked example, a team spending £8,000 a month with a 1.8 per cent conversion rate might record a baseline of 144 conversions a month. That is an illustrative figure, not a benchmark. The point is that the number is written down, dated and attributable to a named system.

Record what you cannot control as well as what you can. Marketing budgets, pricing changes and stock availability all move conversion rates. A short note in the baseline file explaining the trading conditions gives the later report somewhere to point when a figure needs context.

Choose metrics that survive scrutiny

Good metrics share three traits. They are defined, they are comparable over time, and they are hard to move without doing real work. Vanity measures fail the third test.

Split your reporting into three layers. Business outcomes sit at the top, such as revenue or qualified pipeline. Channel measures sit in the middle, such as cost per lead or return on ad spend. Diagnostics sit at the bottom, such as impression share or click-through rate.

Report the top layer to the board and the bottom layer to the people optimising campaigns. Mixing the layers is how dashboards become unreadable. Our reporting dashboard guide explains how to structure that split visually.

Separate leading from lagging measures. Cost per lead moves within days and tells you about efficiency. Revenue and retention move over months and tell you about effect. A programme judged only on leading measures will be optimised into a corner.

Name the attribution model and its limits

Attribution assigns credit for a conversion to one or more touchpoints. Last-click gives everything to the final interaction. Linear spreads credit evenly. Data-driven models weight touchpoints by observed contribution.

None of these is the truth. Each is a simplification, and each favours the channels that sit closest to conversion. If your agency reports only last-click, brand and upper-funnel activity will always look weak.

State the model in the report header, along with the lookback window and the date the data was pulled. If you change models, restate prior periods so the trend line stays honest. Our measurement mistakes guide lists the errors that follow when this is skipped.

Where the budget allows, test rather than model. A holdout group, a geo test or a paused campaign gives a direct read on incrementality. Those results arrive more slowly and look less tidy than a dashboard, but they settle arguments that attribution reports cannot.

Track consent before you track users

The Information Commissioner's Office publishes guidance on online tracking that shapes what agencies may collect and store. Its online tracking guidance affecting agency analytics and ad tech sets out the consent expectations for cookies, pixels and similar technologies. UK campaigns must also account for the UK General Data Protection Regulation (UK GDPR) and the Data Protection Act 2018; the Privacy and Electronic Communications Regulations (PECR) are relevant to storing or accessing information on users' devices through cookies and similar technologies.

In practice, this means measurement design and privacy design are the same project. If consent rates fall, reported conversions fall too, even when underlying demand is flat. Model that gap rather than pretending it does not exist.

Server-side tagging and modelled conversions can help, but they do not remove the legal requirement. Document what you collect, why, and how long you keep it. That record is also what a client needs when its own legal team asks questions.

Set retention limits at the point of collection and stick to them. Analytics properties accumulate personal data quietly, and an agency that has not mapped what it stores cannot answer a client's data protection questions with confidence.

Judge web performance on mobile first

Most consumer traffic now arrives on a phone, so mobile experience is a measurement input as well as a design concern. Google's mobile optimisation guidance for agency web performance work covers page speed, layout and landing page relevance for paid traffic.

A slow landing page depresses conversion rate and raises cost per acquisition, which then looks like a media problem. Before blaming targeting, check the page. Test the actual journey on a mid-range handset, not a desktop browser resized to a narrow window.

Record page speed and mobile conversion rate in the baseline. If the agency improves the page, the gain should show up in the same report that shows the media spend, otherwise the credit is invisible.

Treat page speed as a metric with a named owner. Where nobody owns the landing page, gains made in one quarter tend to disappear quietly by the next.

Set the reporting rhythm and format

Reporting should match the decision cycle. A weekly one-page summary works for an optimisation team. A monthly review suits budget holders. A quarterly deep dive suits the board.

Keep one source of truth. If the client's finance system and the agency dashboard disagree, agree which one governs the reported figure and reconcile the other to it. Finance usually wins, and that is fine as long as everyone knows.

Reports should always include the same sections: objective, activity, results against baseline, what changed, and what happens next. Consistency matters more than length. A three-page report with stable structure beats a twenty-page deck that reinvents itself each month.

Every report should explain variance, not just display it. If cost per acquisition rose 20 per cent, name the factor that caused it and what changes next period. A number without a cause invites the reader to invent one, usually at the agency's expense.

Ask for evidence, not anecdotes

The strongest support for an agency claim is independent sector evidence. The Institute of Practitioners in Advertising maintains an advertising effectiveness evidence base supporting agency value claims drawn from award entries and case data.

For media planning specifically, ISBA and Ebiquity published evidence on Total TV effectiveness that is relevant when an agency argues about reach and channel mix. Cited evidence of this kind is harder to dismiss than a single campaign story. For UK audience measurement, relevant sources depend on the channel: BARB measures television audiences, RAJAR radio audiences, JICMAIL direct mail, UKOM with Ipsos iris online audiences, and ABC audits circulation and distribution data.

Published studies also give you a way to challenge a plan without turning the meeting personal. You are questioning the assumption, not the person who wrote the slide.

Ask any shortlisted agency which published studies inform its planning. An agency that can name them is usually one that measures its own work properly. Our agency comparison guide sets out how to compare those answers across suppliers.

Benchmark against something external

Internal month-on-month comparison tells you whether things improved. It does not tell you whether the result is good. For that you need an external reference point, such as sector conversion rates, cost per lead ranges or published industry studies.

Benchmarks are context, not targets. A sector average includes businesses with different margins, price points and sales models. Use it to frame a conversation, not to set a bonus.

Where no credible benchmark exists, commission one. A short research project that establishes typical performance in your category pays for itself the first time a budget conversation turns difficult. Our benchmark research guide explains how to scope that work.

Write the contract so measurement is possible

Measurement rights belong in the contract, not in a side conversation. Specify access to ad accounts, analytics properties and tag management. Confirm who owns the data and what happens to it at the end of the engagement.

Agree the reporting cadence, the named metrics and the definition of each one. If a metric is defined differently by the two sides, the dispute will surface at the worst possible moment.

Platform guidance can help here. LinkedIn publishes ad tips and best practices for campaign optimisation that set out how its own reporting fields behave, which is useful when you are agreeing what a lead means on that channel.

Agree what happens on exit as well. Ad accounts, creative files, tag configurations and historical reports should transfer without negotiation. Measurement you cannot take with you is not really yours.

Keep a measurement glossary

Shared vocabulary prevents most reporting arguments. Keep the definitions short and put them in the appendix of every report.

Common questions

How often should an agency report?

Match the cadence to the decision cycle. Weekly summaries suit optimisation teams, monthly reviews suit budget holders and quarterly deep dives suit boards. More frequent reporting rarely improves decisions unless someone acts on it.

Who owns the analytics account?

The client should own the ad accounts, analytics property and tag management container, with the agency given access. That arrangement protects continuity if the relationship ends and makes the data auditable.

Can we measure without cookies?

Partly. Consent-based measurement, modelled conversions and incrementality testing all work alongside cookie data. The ICO's online tracking guidance sets the boundaries, so treat consent design as part of measurement design.

What if the agency's dashboard disagrees with ours?

Agree in advance which system governs the reported figure, then reconcile the other to it each period. Documenting the difference is better than quietly picking whichever number looks stronger.