Direct answer: there is no single cost that is correct for every decision. Use platform-delivered spend for intramonth pacing; invoiced or accrued cost for financial commitments; and an approved analytical cost — with fees, tax, FX, credits, and exclusions made explicit — for CAC and profitability. Connect those layers in a Media Cost Reconciliation Ledger with an as-of date, tolerance, source, and owner. AI may classify invoice lines and propose matches; it should not autonomously decide which number becomes financial truth.
A variance is not automatically a mistake. A platform may record delivery in a different time zone; an agency may consolidate accounts and add fees; Finance may recognize tax, FX, or accruals; an invalid-activity credit may arrive after the invoice. Google documents differences between campaign and billing views, billing transactions that include costs, credits, and adjustments, and credits issued after invoicing. Without states and dates, a performance review becomes a contest among extracts.
Start with the decision, not the cost column
Define the question each layer answers. Delivered spend asks whether a campaign is consuming budget now. Payable cost asks what the company owes a media owner or partner. Accounting cost asks what Finance recognized in the period and cost center. Analytical cost asks which comparable basis supports CAC, margin, or a forecast. One amount may serve two layers, but that must be a documented choice rather than a spreadsheet accident.
Suppose a campaign ends the month at $1,000,000 in the platform. The invoice is $985,000 after an adjustment, the agency adds an $80,000 fee, and Finance recognizes $1,095,000 after tax and FX. The platform figure can guide last-day pacing. Accounts payable follows the invoice and contract. For CAC, a committee might approve $1,065,000 by excluding institutional production but including the fee. This illustrates the structure; it is not a benchmark or accounting recommendation.
Build the Media Cost Reconciliation Ledger
- Identity: account, canonical campaign, native IDs, market, legal entity, cost center, and period.
- Delivered spend: amount, currency, time zone, extraction time, window, and platform source.
- Document: vendor, invoice number, line item, service period, due date, and billed amount.
- Adjustments: agency fee, technology, production, tax, FX, credit, rebate, invalid activity, and accrual.
- Policy: inclusion or exclusion by decision — pacing, accounts payable, CAC, margin, or forecast.
- Lineage: file or API, version, transformation, approver, and as-of date.
- Outcome: reconciled cost, absolute and percentage variance, state, reason, and owner.
Do not rebuild identity inside the ledger. Use https://makinai.co/insights/en/canonical-campaign-identity-media-analytics-crm-ai to attach each source line to the right initiative, then apply cost policy. Combining matching and accounting treatment in one rule obscures audits: a campaign can be correctly identified while its cost is unsettled, or financially closed while mapped to the wrong commercial initiative.
Use four reconciliation states
Classify each campaign-account-month unit as matched, timing difference, explainable adjustment, or unresolved. Matched is within the approved tolerance. A timing difference reflects close timing, time zone, accrual, or a late document. An explainable adjustment has evidence and policy, such as tax, fee, credit, or FX. An unresolved item lacks sufficient evidence and stays out of sensitive automated decisions. Never force variance to zero by spreading a residual without a recorded rule.
Every row needs an as-of date. A report issued November 3 may show provisional October cost; the November 12 close may include a delayed invoice or credit. Preserve versions so the team can reproduce the CAC used in the earlier meeting. Do not overwrite an actual with an unlabeled forecast, and make a retroactive correction touch only the affected window.
Where AI helps and where deterministic controls decide
AI is useful for extracting fields from inconsistent invoices, proposing campaign and cost-center matches, classifying variance causes, summarizing exceptions, and prioritizing investigation. Deterministic controls should validate document totals, currency, duplicates, period, tolerance, cardinality, authorization, and posting. High-confidence matches can enter batch review; ambiguity, a new document type, a policy change, or a material value requires human approval.
Google Analytics supports importing cost, clicks, and impressions from external platforms to compare marketing initiatives. That increases analytical coverage, but it does not make imported spend payable or accounting cost. Import the layer appropriate to the report and expose the policy name, currency, and as-of date. Otherwise, a technically successful integration can institutionalize the wrong basis.
Deliver a pilot in one close cycle
- Week 1 — select one market, legal entity, agency, and up to two platforms; baseline variance and manual effort.
- Week 2 — approve layers, treatment of fees and tax, currency, service period, tolerances, and owners.
- Week 3 — connect platform exports, invoices, and the ledger; implement IDs and idempotent keys.
- Week 4 — generate the reconciliation ledger and exception queue; run in shadow mode without changing payments or bids.
- Week 5 — follow the close with Marketing, agency, Data, and Finance; correct rules and replay.
- Week 6 — publish the runbook, approval trail, and comparison report; expand, constrain, or stop.
Scope includes a dictionary, source map, cost policy, connectors, document parser, versioned ledger, service-period and FX rules, tests, exception queue, approvals, monitoring, replay, and runbook. Cost rises with legal entities and currencies, local tax, rebates, agency buying, consolidated accounts, rate cards, allocated production, retroactive adjustments, inconsistent PDFs, poor campaign IDs, and near-real-time close expectations.
Accept the work on reconciliation and close speed
- Coverage: share of delivered and billed spend connected to account, campaign, period, and cost center.
- Variance: difference by cost layer before and after explainable adjustments.
- Exceptions: unresolved value and count, age, and concentration by vendor.
- Correctness: duplicates, idempotent reprocessing, and reproducible retroactive entries.
- Operations: hours to close media, exception resolution time, and manual-review rate.
- Decision readiness: CAC, pacing, and forecasts expose policy, currency, as-of date, and close state.
Once cost is reconciled, connect the approved basis to acquisition economics and payback at https://makinai.co/insights/en/ai-customer-acquisition-margin-payback-value, and put material exceptions into the weekly decision report at https://makinai.co/insights/en/ai-weekly-marketing-report-data-decisions. Reconciled cost still does not prove attribution or incrementality. It makes the denominator reproducible; media contribution requires a separate question and design.
Assign responsibility and make the next decision
Media owns delivery extracts and operating context. The agency supplies itemized invoices, accounts, periods, and contract logic. Marketing Operations owns campaign identity and budget. Data operates ingestion, tests, and replay. Finance decides service period, tax, FX, materiality, and approval. Procurement clarifies fees and rebates. An AI partner implements extraction, assisted matching, monitoring, and evaluation inside those decisions; it does not post payments or rewrite finance policy.
If your dashboard, agency invoice, and finance ledger close at three different totals, bring one month of exports, documents, and the current policy to a bounded discussion through https://makinai.co/services/en/digital-marketing-media-performance-growth-agency. MAKINAI can map the variance and estimate the smallest pilot before you automate pacing, CAC, or AI optimization.