Every marketing platform claims the conversion

A customer sees a social ad, searches the brand, clicks a paid result, returns directly and buys. Social reports a view-through conversion. Search reports a click-through conversion. Analytics credits a channel according to its model. The CRM records one order.

Adding the platform claims together can produce more conversions than the business made. That does not prove that every platform is fraudulent. It proves they are not maintaining a shared ledger.

Marketing attribution becomes governable when we stop asking one number to answer three different questions:

• What happened?
• What helped us optimise delivery?
• What happened because of the investment?

Platform attribution is an optimisation signal

Ad platforms observe their own impressions, clicks and eligible identities. They apply configurable lookback windows and attribution rules. Google documents that its conversion window determines how long after an interaction a conversion can be recorded; LinkedIn similarly exposes click-through and view-through windows.

These numbers help each platform's algorithm and operators compare activity inside that environment. They are not a de-duplicated company-wide sales report. Two platforms can legitimately touch the same buyer and each claim credit under its own rules.

The problem begins when an optimisation signal is presented as incremental business impact.

Keep three measurement layers

The outcome ledger

The commerce, billing or CRM system should record the canonical business outcome: accepted order, qualified opportunity, activated subscription or settled revenue. Define reversals, duplicates, fraud, tax, refunds and currency consistently. This answers what happened.

The governed journey view

Analytics or a warehouse can connect observable interactions across channels using an explicit identity and attribution model. It explains paths and contribution under stated limitations. This answers what we observed about the journey.

Causal evidence

Holdouts, geo experiments, conversion-lift studies and, at a broader level, marketing-mix models estimate what would not have happened without an intervention. No click-path model can establish that counterfactual by itself. This answers what the investment caused, with uncertainty.

Leaders can still have a concise scorecard. The layers should not be collapsed into a fictional universal truth.

Build a canonical conversion contract

Start with the business event, not the channel:

• a stable conversion identifier;
• event type, timestamp and value;
• lifecycle state and reversal rules;
• customer or account identity where lawful;
• currency and value policy;
• consent and permitted destinations;
• deduplication behaviour; and
• the system responsible for finality.

Pass the same event identifier to eligible platforms for deduplication where supported. Keep a delivery log showing which version of which event was sent, without exposing more customer data than necessary.

Server-side collection can improve control and resilience; it does not create incrementality or remove privacy obligations. It can faithfully send the wrong event just as efficiently as the right one.

Govern windows and models as business rules

A 30-day click window and a 1-day view window encode different beliefs about influence. Last-click, first-click and data-driven models distribute credit differently. Changing either can move reported performance without changing one advertisement or sale.

For every report, disclose:

• click and view windows;
• attribution model and version;
• identity scope;
• included conversion states;
• observed versus modelled data;
• reporting lag; and
• known blind spots such as offline or cross-device activity.

Version material changes and annotate trend lines. Otherwise, a measurement-policy change can be mistaken for business growth.

Use platform data without letting it grade the whole business

Platform reporting remains valuable. It is close to the auction and often has signals unavailable elsewhere. Use it to optimise campaigns, audiences and creative within that platform. Monitor the relationship between claimed conversions and canonical outcomes rather than expecting equality.

For budget allocation, combine:

• backend revenue, margin or qualified pipeline;
• blended acquisition efficiency;
• governed cross-channel journey evidence;
• lead and customer quality;
• saturation and diminishing returns; and
• experimental lift where stakes justify it.

A third-party attribution product may make collection and analysis easier. It still embodies assumptions. Buying it does not remove the need for definitions, identity governance or experiments.

Privacy and identity are measurement constraints

Consent, browser restrictions and platform identity boundaries mean the journey will be incomplete. Trying to close every gap can create disproportionate privacy and security risk.

Collect only what has a clear purpose. Document legal basis and retention. Restrict raw user-level joins. Prefer aggregate evidence when individual identity is unnecessary. Explain where modelling is used and avoid presenting estimated behaviour as observed fact.

Good governance accepts that some uncertainty is a feature of responsible measurement, not an engineering defect to bypass.

A practical operating rhythm

1. Daily: monitor event delivery, duplicates, rejected payloads and unusual claim-to-outcome ratios.
2. Weekly: review spend, platform signals, canonical outcomes and lead quality together.
3. Monthly: reconcile windows and lifecycle changes; inspect channel overlap and blended economics.
4. Quarterly: test a material assumption with a holdout or geo design where feasible.
5. On change: version the conversion contract, attribution settings and dashboards.

Marketing, data, finance, privacy and engineering need named decision rights. A committee that discusses everything but owns nothing is not governance.

What to measure

• platform-claimed conversions as a ratio of canonical outcomes;
• percentage of conversion events with stable identifiers;
• duplicate and rejected event rates;
• time to reconcile a material discrepancy;
• share of spend covered by an incrementality test;
• blended CAC or cost per qualified outcome;
• downstream quality and margin by acquisition cohort; and
• attribution-rule changes without documented approval.

The counterargument: attribution is too imperfect to be useful

It is imperfect. Abandoning it would also be a mistake. Teams need feedback to operate campaigns and understand journeys.

The answer is to match evidence to the decision. Platform attribution can direct tactical optimisation. Governed analytics can describe observable paths. Experiments can support causal budget decisions. Backend data can close the financial ledger.

The mature position is neither blind trust nor total cynicism.

Adovo’s view

Attribution meetings become unproductive when everybody defends their favourite dashboard. Start with the decision. Then choose the evidence that can support it and name the uncertainty that remains.

Every platform can claim a role in the conversion. Only the enterprise can govern what the conversion means.

Primary documentation and further reading

• Adobe marketing channel processing rules
• Microsoft Power BI star schema guidance