Customer data platforms return on investment: a method

The buyer question ‘customer data platforms return on investment’ is answered only when a customer data platform buyer can estimate causal incremental contribution and divide it by full incremental economic cost. Profile counts, match rates, audience exports and supplier-attributed conversions are not returns.

This method is for a buyer-owned England journey and contains no example values. Named UK accounting, tax, statistics, privacy and legal reviewers must approve actual inputs and interpretation.

What to take away

Define the intervention and alternative

Describe the change narrowly. For example: introduce a specified identity and suppression workflow for authorised membership events, with production activation controlled. State the eligible population, period, geography, outcome, implementation date and service version.

Then document the counterfactual: what would have happened through the current manual process or existing approved systems. It is not automatically zero. HM Treasury's Green Book 2026 is government appraisal guidance, but its distinction between business as usual, additionality, costs and risks is a useful discipline. Private buyers should not treat it as a prescribed corporate accounting method.

Keep four evidence layers apart

A dashboard may display all four without proving the fourth. Do not add supplier-attributed amounts from different systems; overlapping credit can exceed the underlying outcome.

Design the counterfactual before launch

Choose a proportionate method with a qualified statistician. A controlled experiment may be possible where allocation is ethical, lawful and operationally credible. Where it is not, a carefully designed quasi-experiment or other evaluation may reduce some biases, but assumptions and residual confounding must remain visible.

Specify the design by name. A randomised holdout or control-group test can estimate lift where assignment is ethical, lawful and operationally credible. If randomisation is not feasible, options include difference-in-differences, synthetic control and propensity-score matching; state and test the assumptions each requires. Marketing mix modelling (MMM) is an aggregate, model-based approach, not a substitute for a suitable holdout when the question is the incremental effect of a specific CDP intervention. The Royal Statistical Society (RSS) is a UK professional body; choose a reviewer with relevant causal-evaluation expertise.

The Magenta Book separates monitoring, process, impact and value-for-money evaluation. It explains that impact assessment asks whether changes can be attributed to an intervention. The guidance is for government evaluation; it does not validate a commercial CDP study.

Predeclare eligibility, allocation, outcome event, observation period, sample exclusions, missingness treatment, stopping rule and analysis version. A test that changes targeting, creative and customer-data processing together cannot isolate the platform contribution without further design.

Build contribution from observed accounts

Use reconciled buyer records. Define:

Incremental completed outcomes = observed outcome under intervention - estimated counterfactual outcome

Contribution per outcome = recognised revenue - refunds - cost of goods or delivery - other approved variable costs - applicable reviewed adjustments

Incremental contribution = incremental completed outcomes x contribution per outcome

Recognise revenue under the buyer's applicable accounting framework, such as FRS 102 or IFRS 15 where applicable, and apply the corresponding policy to costs. The buyer's accountant should approve the treatment and contribution definition. For VAT and other tax treatment, confirm the basis against HMRC guidance, including VAT Notice 700 for VAT, and document whether VAT is recoverable. Do not use gross turnover where contribution is the intended numerator.

Use the full denominator

[Incremental economic cost](/customer-data-platforms-hidden-costs-in-england) = supplier cash + implementation + migration + data and identity work + assurance + operation + monitoring + rights handling + renewal + exit - evidenced avoided baseline cost

For a CDP, check whether supplier charges vary by event or profile volume, identity-resolution compute, activation or export, or consent-management modules. Include applicable amounts once, under supplier cash or the relevant work category.

Only evidenced, causal avoided cost belongs in the subtraction. Staff time saved in one task may reappear as governance, exception handling or monitoring. Record hours and approved loaded rates rather than asserting productivity.

For supplier-control assurance, identify whether an ISAE 3402 report or SOC 2 report covers the relevant service and period, and review its scope, exceptions and complementary user-entity controls. Such assurance can support due diligence but does not establish causal uplift or replace the buyer's own assessment.

ROI = (incremental contribution - incremental economic cost) / incremental economic cost

State currency, VAT and tax basis, period, population, evidence IDs, uncertainty and exclusions beside the result. If the counterfactual, numerator or denominator is unfit, report ROI as not established.

Set the decision rule

Give the decision owner a sensitivity table with buyer-approved inputs and no automatic optimism adjustment. Define what would cause pause, redesign, further evidence collection or withdrawal. A positive estimate cannot compensate for a failed privacy, PECR, security, accessibility or contract gate. Preserve negative and inconclusive findings, then recheck the analysis when the journey, model, population or costs change.

For personal-data processing, apply the UK GDPR and Data Protection Act 2018, and check guidance from the Information Commissioner's Office (ICO), the UK's data protection regulator. The Privacy and Electronic Communications Regulations (PECR) also apply where relevant, including to electronic marketing and cookies. Where customer-facing digital content or interfaces are in scope, assess accessibility against WCAG 2.2.

Before you act

Common questions

What evidence layers must be kept apart when assessing customer data platform ROI?

The article lists four: system output, supplier attribution, financial reconciliation and causal effect. A dashboard may display all four without proving the fourth. Do not add supplier-attributed amounts from different systems because overlapping credit can exceed the underlying outcome.

How should the counterfactual be designed before launch?

Choose a proportionate method with a qualified statistician. A randomised holdout or control-group test may be possible where allocation is ethical, lawful and operationally credible. If not, consider difference-in-differences, synthetic control or propensity-score matching, with assumptions and residual confounding made visible. Marketing mix modelling (MMM) is an aggregate model-based approach, so it should not be presented as proof of the effect of a specific CDP intervention without a suitable design. Predeclare eligibility, allocation, outcome event and analysis version.

What belongs in the incremental economic cost denominator?

Supplier cash, implementation, migration, data work, assurance and operation, minus evidenced avoided baseline cost. Only evidenced, causal avoided cost belongs in the subtraction. Staff time saved in one task may reappear as governance, exception handling or monitoring.