Marketing attribution and commercial reconciliation

Start with the disagreement that matters

Marketing reports a strong return while finance sees little change in revenue. Several advertising platforms each claim the same order. Sales says the cheapest leads are the least likely to progress. These disagreements cannot be resolved by choosing the most flattering attribution dashboard. Emerge builds a commercial reconciliation model that explains what each system observes and how the business should interpret it.

The engagement begins with one decision, such as reallocating a market budget or comparing acquisition quality across channels. We identify the business outcome relevant to that decision and the evidence available to connect it to acquisition activity. Attribution assigns credit within a chosen model; it does not by itself prove that advertising caused the outcome.

Establish the business outcome and its lifecycle

An ecommerce model needs explicit handling of orders, refunds, cancellations, shipping and tax. A lead-generation model needs accepted leads, qualification, opportunity stages and eventual wins or losses. We agree which date and value represent each stage and which system owns the record.

The model preserves the difference between a customer, transaction and opportunity. One customer may place several orders or participate in multiple buying processes. Joining every later outcome to a single first contact can overstate the role of the original channel and obscure retention activity.

We also assess data completeness. If sales teams inconsistently maintain a qualification field, an apparently poor channel may simply have a different reporting habit. The first deliverable can therefore include CRM process repairs alongside analytics engineering.

Build a channel taxonomy that survives real traffic

Campaign parameters and platform identifiers are normalised into a maintained channel model. Paid search, organic search, referrals, email, social and messaging links receive clear classification rules. The rules retain the original source fields so an analyst can investigate a classification rather than relying on an irreversible label.

Redirects, link shorteners, booking hosts and application transitions are reviewed for attribution continuity. We test whether approved campaign parameters survive the actual journey. Unattributed traffic remains an explicit category; it is not redistributed into preferred channels merely to make a report look complete.

Campaign naming guidance is accompanied by validation. A taxonomy that exists only in a document will drift as teams launch new activity. We add checks for unknown sources and missing campaign fields at the point where those gaps can still be repaired.

Use the warehouse to make assumptions inspectable

BigQuery can combine permitted analytics events with advertising costs and business records at an agreed grain. Emerge defines the joining keys, lookback policy, time-zone treatment and deduplication rules before implementing the attribution calculation. Source tables and derived models remain distinguishable.

A straightforward observed-touch model can be a useful starting point because teams can inspect its behaviour. More complex models require enough representative data and a clear reason for the additional complexity. We compare model outputs against known journeys and document how credit changes when assumptions change.

Platform-reported and warehouse-calculated attribution may differ. The model explains those differences through data scope, identity availability, windows, processing and modelling. It does not claim access to information that a platform does not export or that the visitor did not permit the organisation to collect.

Separate attribution from incrementality

An attributed conversion can occur without the campaign being the reason the customer bought. Emerge presents that limitation directly in the interpretation. Where the decision requires causal evidence, we assess a separate experimental or incrementality approach rather than labelling a credit-allocation model as proof of lift.

The choice depends on business volume, geography, campaign design and the ability to maintain a credible comparison. A model can still be operationally useful without answering every causal question. The reporting should make clear which question it answers and which conclusions would exceed the evidence.

Validate the records and the interpretation

Acceptance follows sample orders or leads from their source into the warehouse and final reporting view. Checks cover duplicate identifiers, missing costs, delayed status changes, refunds and multiple currencies. We reconcile the business total before interpreting how the model distributes channel credit.

Sensitivity review shows how results change under reasonable alternative windows or attribution rules. A channel ranking that reverses under a small assumption change should be communicated as uncertain, not presented as a precise instruction to move budget immediately.

The commercial team reviews representative findings and the action they would take. This prevents the implementation from ending with technically valid SQL that nobody trusts or understands well enough to use.

Maintain the model as the business changes

Handover includes the channel dictionary, source contracts, model logic, reconciliation tests and a record of known coverage limits. Owners are assigned for campaign naming, CRM stage quality, data ingestion and interpretation. Changes to the sales cycle or commerce platform trigger a review of the model’s assumptions.

The connected Looker Studio service can make the analysis usable in operating meetings. Emerge’s anonymised measurement work provides relevant context, while every new attribution programme establishes its own baseline. The outcome is a defensible decision process that can explain its numbers and recognise where evidence is incomplete.