Impressions Do Not Pay Invoices: Building a Direct Marketing Measurement Stack for UK Firms in 2026

For twenty years, UK marketing measurement was shaped by what was easy to count rather than what was true. Clicks, impressions and platform-attributed conversions were plentiful, cheap to collect and, crucially, flattering. That arrangement has now collapsed, and a good deal of British boardroom reporting has not caught up.

The argument of this piece is straightforward: most firms do not have an attribution problem, they have a decision problem. They are collecting metrics that describe activity rather than metrics that inform allocation. Fixing that requires fewer dashboards and more discipline.

What actually changed by 2026

Three shifts have combined. Third-party cookies are no longer a dependable basis for cross-site identity. Mobile operating systems have degraded open-rate and click data to the point where email engagement reporting is indicative at best. And UK regulators have continued to scrutinise advertising technology and data practices, with the Information Commissioner's Office showing a sustained interest in how consent and legitimate interests are evidenced.

The practical consequence is that the identity graphs many firms relied on are thinner, the platform dashboards are more self-referential, and the gap between reported performance and commercial outcome has widened. A paid social platform that claims credit for a sale it merely observed is not lying; it is doing exactly what its attribution window was designed to do.

Impressions measure delivery, not effect

An impression confirms that an advert was served, subject to whatever viewability standard the platform applies. It says nothing about persuasion, memory or intent. Yet impressions remain a headline metric in a surprising number of UK marketing reports, largely because they are large, available and move in the right direction when spend increases.

Direct mail has an analogous trap in the response rate. A one per cent response rate is neither good nor bad until you know what the same audience would have done without the mailing. That is the whole argument for incrementality, and it is the argument most firms still avoid because the answer might be uncomfortable.

First-party data is the foundation, not the strategy

Much of the current commentary treats first-party data as a solution. It is better understood as plumbing. What matters is whether the data is permissioned, accurate, deduplicated and joined to commercial outcomes.

In practice, that means a customer record that carries source, channel, first touch, most recent interaction and response history. For a direct mail programme, it means capturing the response code or unique URL at the point of reply and appending it to the record, not reconciling it manually three months later. For a B2B firm, it means a CRM that is treated as an asset rather than an administrative burden.

Under UK GDPR, the lawful basis for each processing activity should be documented, and data minimisation is not optional. Retention schedules should be enforced rather than written. None of this is glamorous, and all of it determines whether any attribution model built on top will be worth reading.

Incrementality: the only test that answers the question

Incrementality testing asks what would have happened without the activity. For direct mail, the method is a randomised control cell: withhold the mailing from a randomly selected slice of the audience, typically 10%, and compare outcomes over an appropriate window. Eight to twelve weeks is realistic for most considered purchases; four to six weeks may suffice for fast-moving retail.

For channels where individual-level holdouts are impractical, matched-market or geo tests work: run the activity in a set of comparable regions and compare against a matched control set. The discipline required is consistency of measurement, not statistical sophistication.

Two practical rules. First, size the test properly, since a control cell too small to detect a realistic effect will produce noise dressed as insight. Second, agree the success metric before the test begins, because the temptation to reinterpret results afterwards is considerable.

Marketing mix modelling is no longer enterprise-only

Econometric modelling, once the preserve of large advertisers with dedicated analytics teams, has become accessible to mid-market firms through lower-cost providers and better tooling. It requires a few years of weekly data across channels and enough variation in spend to identify effects. It will not tell you which creative worked, but it will tell you where the marginal pound is best placed.

The sensible posture is complementary: incrementality tests for specific questions, marketing mix modelling for portfolio allocation, and neither treated as a precise accounting instrument.

A three-tier stack an SME can run

Tier one, non-negotiable. Unique response codes or URLs for every direct marketing drop; call tracking where the telephone is a meaningful channel; a CRM with source and channel tagging applied at capture; one control cell per major activity.

Tier two, twice yearly. Structured incrementality tests on the two channels with the largest spend and the least certainty. Report results to the board in plain language.

Tier three, annually. A marketing mix model or equivalent econometric review to inform budget setting for the year ahead.

Underpinning all three: a single reporting view agreed with finance, and a written contract about which metric each function is accountable for. Most attribution disputes are actually disputes about who owns a number.

The UK compliance guardrails

Any measurement stack processes personal data, and the design has to reflect that. PECR governs electronic marketing to individuals, with its own consent and soft opt-in rules; postal marketing sits on a different footing, reliant on legitimate interests with a clear and simple opt-out and screening against the Mailing Preference Service. Profiling and automated decision-making may require a data protection impact assessment. Website analytics cookies generally still require consent, although the direction of UK data reform suggests some relaxations for low-risk analytics, which firms should monitor rather than assume.

The ICO's direct marketing guidance remains the reference point, and the industry codes published by the DMA provide a practical overlay. Firms that treat compliance as an obstacle to measurement tend to discover that it is cheaper to build the guardrails in at the start.

What to stop doing

Stop reporting impressions as a performance metric. Stop attributing direct mail on last click. Stop treating platform-reported conversions as revenue. Stop buying contact lists, which is both poor data practice and a compliance exposure. And stop commissioning dashboards that nobody uses to make a decision.

The conclusion

The measurement question in 2026 is not how to track more, but how to decide better. A modest, disciplined stack built on clean first-party data, a consistent control methodology and an annual econometric check will outperform an elaborate multi-touch model built on signals that no longer exist. Impressions are cheap to count and expensive to believe.