Attribution supplier routes on one matrix, with the commercial unit kept blank

Define the evidence need first, then choose the one route that matches it. Everything below applies that rule.

This matrix sets out six attribution procurement routes open to a UK buyer and keeps every commercial cell blank by design. No supplier was contacted or tested, no price was verified and no ranking is given. Research closed on 6 September 2026.

Read the guide to planning around one buyer decision before the rows. A route chosen for the wrong job cannot be repaired by paperwork.

What to take away

Six procurement routes, each with a boundary

Products named in the table illustrate each route. They are examples, not endorsements, and none was tested.

HM Treasury's Quality in Policy Impact Evaluation explains why outcome monitoring differs from causal attribution in public policy, setting a method boundary rather than endorsing a supplier.

Example: one retailer, two evidence needs

An England-based homeware retailer wants to know whether paid social increased quarterly subscriptions. That evidence need is aggregate and long-window, so the matching route is MMM software, with spend and order data prepared in house.

The same retailer's app team asks which install source drove day-one purchases. That need is app-level and short-window, so it takes mobile measurement. One scorecard over both jobs would compare unlike evidence.

Identical evidence fields for every route

Record the legal entity, exact service or edition, UK availability evidence, inputs, outputs, model version, missingness, buyer workload and change history. Add data roles, sub-processors, transfer routes, security controls, IP rights and support, then export, deletion and exit.

The ICO's contract guidance lists the Article 28 processor topics to cover and was under DUAA review at the research date.

A demonstration shows that a control exists. Only the buyer's retained test evidence shows how the purchased configuration behaved. A blank attribution budget template applies the same field discipline to money rows.

Why every commercial cell stays blank

Paid routes are usually quoted rather than listed. LiveRamp, InfoSum, Recast, Analytic Partners and Kantar quote per contract. AppsFlyer, Adjust, Branch and Singular quote on request, with some entry tiers published for small app volumes.

Open-source options are free to install. Meta Robyn and Google Meridian are free MMM packages, and GeoLift is a free geo experiment library from Meta. Both MMM packages still need in-house data engineering.

Use common fields for currency, VAT and billing period, then minimum term, users, event volume, connectors and model runs. HMRC's VAT guidance does not settle an individual supply, so a tax reviewer confirms it.

Acceptance testing on one shared fixture

No weighted total should conceal a failed gate. The NCSC's SaaS guidance also requires the buyer to configure and maintain its own tenancy securely. Re-run affected checks after a change to a model, connector, term or sub-processor.

Common questions

Are the named vendors recommendations?

No. Google Analytics 4, LiveRamp, Recast and the others appear as examples of a route on the market. Nothing was tested, and the note on why no provider is named best sets out that stance in full.

Why are there no prices?

Paid routes such as InfoSum and Kantar quote per contract, and the free packages carry engineering cost instead of a licence fee. Published units are not compatible, so no total is built here.

What if a supplier publishes nothing at edition level?

Supplier claims stay attributed and unknown stays unknown. The buyer can request edition-level material under confidentiality, or keep the row on HOLD until it arrives.