Build a versioned client data dictionary before the first advertising report

Define meaning, not just field names

Create one entry per decision metric, not a list of every field a platform can export. Each entry needs a business name, technical field or event, source, unit, inclusion rule, exclusions, date basis, currency rule, deduplication rule, availability caveat and definition owner. Include an example that a colleague can classify without asking the author. An entry for qualified enquiries should explain who qualifies them and which status counts, rather than merely repeat the label qualified.

For a teaching example, define enquiry as a submitted request, qualified enquiry as a request accepted for sales follow-up, and sale as a completed order net of the agreed cancellation rule. Ten requests, six qualified requests and two sales are three stages of one funnel, not eighteen business outcomes. If platform conversions also contain a button click, describe it as a diagnostic action. Do not combine it with sales to create an apparently stronger denominator for CPA.

Version the dictionary with an effective date and a short change note. Suppose the business starts excluding duplicate requests next month. Keep the previous definition and state which periods use each version. Recalculate old periods only when the underlying records support the new rule; otherwise label the break in comparability. AdAce Ads supports a client KPI event and results by event. Those settings supply reporting context, but your dictionary also covers CRM qualification and business rules that are not automatically determined by an ad account.

Assign semantic ownership

• A definition owner answers semantic questions; a data owner answers extraction questions. Naming both avoids treating a report developer as the authority on what constitutes a sale or a qualified customer.
• Add an allowed-use field: optimization diagnosis, management reporting or invoicing reconciliation, for example. A metric adequate for directional monitoring may be unsuitable for a precise business revenue claim.
• Store a small set of boundary examples beside the definition. A duplicate, a reopened enquiry and a refunded order often expose more disagreement than a paragraph describing the normal case.

Publish and maintain a definition version

1. Start with decisions the client expects to make. For each decision, identify the numerator and denominator, their sources and whether they use compatible dates. Define an unavailable value separately from a genuine zero.
2. Ask marketing and sales to classify the same small teaching dataset. Record where their answers differ. Resolve the counting rule before calculating a target, rather than averaging incompatible counts or choosing one team's terminology.
3. Document time boundaries: account reporting date, lead creation date or sale completion date. Include conversion maturity and late updates. A date label alone is insufficient if one source follows clicks and another follows completed orders.
4. Publish an initial version and attach its identifier to reports. Keep an amendment log stating what changed, why, who approved the definition and whether historical values were restated. Avoid replacing the old entry without a trace.
5. Before every major reporting change, rerun the boundary examples. Check that exports, client KPI settings and written report labels still match the agreed meaning. Open a reconciliation issue when implementation and definition diverge.

Agreement is not attribution equivalence

Sources and further reading

• AdAce Ads client KPI event
• AdAce Ads results by event