Costs and pricing
Marketing automation return on investment: a platform review method
A method for assessing marketing automation return on investment in platform reviews, using contribution, full costs, counterfactuals and evidence quality.
For a platform review, Marketing automation return on investment is defensible only when cost is complete and the benefit is both contribution-based and supported by an evaluation design. A dashboard can attribute an order to a journey; it cannot, by itself, show that automation caused the order.
What to take away
- Marketing automation ROI needs complete costs and contribution-based benefits supported by an evaluation design.
- Choose a proportionate evaluation method before launch to establish whether automation caused the outcome.
- Use eligible orders and approved contribution treatment, not gross revenue or vendor pipeline estimates.
- Include all relevant costs such as licences, labour, assurance and exit in the economic cost.
- Check the named platform's pricing tier and its attribution reporting before treating either as an input.
- Report a range with assumptions and evidence quality, not a single ROI figure.
Define the decision and counterfactual
Choose one decision, audience and period. State what changed, when it began and what would have happened without it. HM Treasury's Magenta Book sets out the difference between observing an outcome and establishing whether an intervention caused it. It sorts designs into three families: experimental (randomised controlled trials and holdout tests), quasi-experimental (difference-in-differences, regression discontinuity, synthetic control and matched comparisons) and theory-based (logic models and contribution tracing). Each carries its own assumptions and limits.
Choose evaluation method
Is causal confidence strong?
Report realised ROI
Report attributed contribution scenario
Ask an evaluation specialist to select a proportionate method before launch. Preserve comparable pre-periods or groups where appropriate, contamination risks, concurrent campaigns and data exclusions. If causal confidence is weak, report an attributed contribution scenario, not realised ROI.
Move from revenue to contribution
Use eligible orders that match the stated geography, channel, customer group and window. Remove cancellations and refunds according to the finance policy. Define contribution before you calculate it: eligible order value less the variable costs of serving that order — packaging, fulfilment, payment fees, service and any other cost the margin policy treats as variable. Apply the organisation's approved contribution treatment and reconcile it to the order ledger. Do not substitute gross revenue or a vendor's estimated pipeline value.
Revenue to contribution
- Use eligible orders matching scope
- Remove cancellations and refunds
- Apply approved contribution treatment
- Include all relevant economic costs
- Do not substitute gross revenue
Costs should include the relevant share of licences, usage, implementation, migration, integrations, loaded internal labour, support, training, legal and privacy work, security and accessibility assurance, operation and exit. VAT belongs in economic cost only to the extent it is not recoverable: HMRC's VAT guidance, including VAT Notice 700 (The VAT Guide) and, where the business has exempt income, VAT Notice 706 (Partial exemption), makes recovery conditional and evidence-based.
Where the platform changes the maths
Platforms are priced differently, and each model moves a different part of the calculation. HubSpot, Klaviyo, ActiveCampaign, Mailchimp and Customer.io typically scale on contacts, active profiles or message volume; Marketo Engage, Braze and Salesforce Marketing Cloud usually add seat, edition or usage charges on top of a database-size tier; Marketing Cloud Account Engagement (formerly Pardot) is bought inside a wider Salesforce contract. The denominator must carry the tier the programme actually needs, not the entry tier.
Attribution reporting is the second trap. In-platform attribution — HubSpot's attribution reports or Klaviyo's attributed revenue, for example — reports platform-attributed gross revenue. That is an input to the numerator, never the numerator itself. Check whether an order is counted once per campaign, whether refunds are netted off, and whether the figure still holds once a contact has been exposed to several journeys.
Use a controlled return card
Equation: ROI for the evaluation window = (causally supported change in contribution minus total economic cost) divided by total economic cost.
Return card inclusions vs exclusions
Inclusions
- Audiences
- Named audiences
- Costs
- Approved variable costs
- Attribution
- Supported attribution
- Tax
- Non-recoverable VAT
- Timing
- Benefits in window
Exclusions
- Audiences
- Unrelated sales
- Costs
- Financing flows
- Attribution
- Unsupported attribution
- Tax
- Recoverable VAT
- Timing
- Sunk costs excluded
Controlled return card
- Currency, unit and periodGBP contribution and GBP cost for one named window; output is a ratio or percentage.
- Numeratorevaluated contribution change less the cost in the denominator.
- Denominatorcomplete economic cost for the same scope and period.
- Inclusionsnamed audiences, channels, eligible orders, approved variable costs, external charges, loaded labour, assurance and accountant-approved allocations.
- Exclusionsunrelated sales, unsupported attribution, financing flows, recoverable VAT, sunk costs excluded by policy and benefits outside the window.
- Owner and sourcesevaluation owner and finance approver; order ledger, margin policy, refund data, test or comparison dataset, delivery log, invoices and time records.
- Sensitivityvary causal effect, contribution rate, customer volume, implementation hours, supplier usage and allocation assumptions separately.
- Stop or reviewwithhold ROI if the denominator is incomplete, the counterfactual is not credible, groups are materially contaminated, contribution cannot be reconciled or a plausible downside changes the decision.
Report more than one number
Show the central estimate only beside its range, assumptions and evidence quality. Add cash payback separately if decision-makers need liquidity information; it is not the same measure as economic ROI. Label supplier-dashboard attribution, observational association and causally supported effect as different outputs.
Log missing orders, delayed revenue, changed margin rules and people exposed to more than one campaign. State whether each issue changes the numerator, the comparison or only precision.
If analysts alter the window or eligible population after seeing results, keep the original specification and require fresh approval. A favourable result may otherwise reflect the revised method, not the intervention.
The decision note should say what happens if evidence is inconclusive. Options might be a longer authorised test, a narrower deployment or no expansion, but none should be selected automatically. Record the accountable owner and the next evidence date.
Privacy and channel restrictions may also affect the evaluated population. The Privacy and Electronic Communications Regulations 2003 (PECR) — regulation 22 on unsolicited electronic mail in particular — and the ICO's direct marketing guidance determine who may be contacted and on what basis; they do not validate an experiment or a marketing outcome. Qualified legal and evaluation reviewers should approve the design and publication wording. A result that cannot survive those boundaries should remain an internal scenario, not a performance claim.
Before you act
- Define one decision, audience and period before launch.
- Select a proportionate evaluation method with an evaluation specialist.
- Use eligible orders and approved contribution treatment.
- Include all relevant economic costs in the denominator.
- Report a range with assumptions and evidence quality.
- Keep an exception log for missing orders and changed rules.
Common questions
What is the difference between revenue and contribution in marketing automation ROI?
Revenue is the total income from orders, while contribution deducts approved variable costs such as packaging, fulfilment, payment fees and service. ROI has to be built on contribution, not gross revenue or a vendor's pipeline estimate, because only contribution shows what is left after the costs the sale brings with it.
Why is a counterfactual necessary for evaluating marketing automation?
A counterfactual is what would have happened without the automation, and it is what separates causation from association. A platform dashboard can attribute an order to a journey, but only a design chosen before launch — a randomised controlled trial or holdout, difference-in-differences, regression discontinuity or synthetic control — can show whether the automation caused it.
What should be done if the evidence for ROI is inconclusive?
The decision note should state what happens if evidence is inconclusive, such as a longer authorised test, narrower deployment or no expansion. None should be selected automatically, and the accountable owner and next evidence date must be recorded.
