The account is spending as planned. The adverts are active. CPA appears stable. Yet net sales are falling, and the teams do not know whether the problem lies with demand, targeting, the landing page, tracking or sales follow-up.
That is precisely the role of an operational audit: locate the constraint, measure its impact and sequence the corrections.
No box-ticking exercise.
1. Definition: an operational acquisition audit prepares the system for improvement
A paid acquisition audit is a structured review of the resources, data and decisions governing investment across search, social, programmatic, retail media and other channels. It compares platform reporting with commercial outcomes and verifies whether the organisation can learn without putting its budget at risk.
Its purpose is not to assign blame to the agency, advertiser or algorithm. It should produce a shared working basis: verified facts, gaps, estimates of value at stake, recommendations, owners and an implementation sequence.
This guide covers the operational optimisation audit: auditors can work with the teams and prepare changes. It differs from an independent assurance engagement, presented later in the Logiks series, whose mandate prioritises evidence, impartiality and the absence of modifications during the review.
The distinction matters.
2. Key figures: investment is growing faster than operational control
- In Europe, the digital advertising market reached €131 billion in 2025, up 10.5%, according to IAB Europe. Video represented €34 billion, social media €35.5 billion and retail media €13.3 billion.
- In France, SRI and Udecam measured €12.4 billion in digital advertising revenue in 2025, an increase of +11%. Search accounted for €4.93 billion and social media €4.18 billion.
- The same report states that eight global participants capture 76% of the French market and 83% of its growth. This concentration strengthens the need for an advertiser-owned reference framework independent of proprietary interfaces.
- In its State of Data 2026 survey of more than 400 decision-makers, IAB reports that 60% to 75% of advanced-measurement users consider it inadequate in terms of rigour, speed, confidence or effectiveness.
- Google notes that some experiment results can remain inconclusive and sometimes recommends waiting four to six weeks before reaching a conclusion. Changing campaigns every three days therefore often makes it impossible to distinguish noise from effect.
Rising spend does not demonstrate equivalent operational maturity. The more automated the portfolio becomes, the more decisive the quality of objectives, exclusions, data and guardrails becomes.
3. The control tower: eight areas to audit in sequence
A media account does not operate in isolation. The audit follows the value chain from the financial outcome back to the settings, so that the team does not optimise a metric that no longer serves the business.
3.1. Area 1 — Economics and actual capacity
Start with the economic model: net revenue, contribution, variable costs, time to cash, purchase frequency, returns, sales capacity and stock. The acquisition threshold is calculated by segment rather than imposed as a universal average.
Consider an offer billed at €2,000 with a 55% contribution margin before media, a 20% lead-to-sale rate and 10% in additional sales costs. The contribution available before advertising is €900 per customer; a lead is worth no more than €180 before a safety margin, not the €400 suggested by expected revenue alone.
The audit also checks the fulfilment constraint. Generating 300 enquiries when the team can call back only 120 in time turns spend into a queue. The right correction may be a temporary reduction, different qualification or sales reorganisation, not more aggressive bidding.
Decision question: which outcome still deserves the next euro?
3.2. Area 2 — Offer, messaging and journey
A technically sound campaign can fail when it promises one thing, attracts a different audience and leads to a generic page. The auditor connects the query or context with the advert, proposition, evidence, friction and call to action.
They examine differences by device, audience, area and intent. A mobile conversion rate that is half as high does not immediately justify exclusion: it may reveal a broken form, excessive load time, customers who research on mobile then convert on desktop, or a poorly explained offer.
Creative assets are classified by angle, evidence, format and maturity, not merely by identifier. This taxonomy reveals whether the team is genuinely testing hypotheses or regularly changing a colour and headline without learning what changes the decision.
Output: message-to-journey map showing breaks and missing evidence.
3.3. Area 3 — Conversion, tracking and consent
Testing follows an event from the browser to the CRM and, where relevant, back to the platform. It verifies triggering, amount, currency, identifier, deduplication, rejection, cancellation, offline import, consent and delays.
Every bidding objective is qualified. A download, page view, lead, opportunity and sale cannot all be declared “primary” with arbitrary values without steering automation towards the easiest outcome. The audit proposes a hierarchy compatible with data maturity.
The report quantifies the difference between platform conversions, analytics outcomes, CRM records and financial sales. It does not force them to be equal, because windows and scopes differ, but explains the known share of the gap and creates controls to detect new breaks.
Output: events × systems × quality × owner matrix.
3.4. Area 4 — Account architecture
The structure is assessed on its ability to concentrate signal, isolate distinct economics and preserve necessary controls. Too many campaigns divide the data; too few mix brand, prospecting, retention, areas or ranges whose constraints are not comparable.
For search, the audit examines intent, actual queries, negatives, brands, areas, match types, adverts, URLs and partners. For social media, it examines audiences, exclusions, overlap, placements, frequency, fatigue, the creative library and the split between acquisition and retargeting.
For programmatic, the review adds inventory, supply path, fees, domains or applications, brand safety, viewability and frequency rules. Average CPM remains secondary if a high share of impressions is served in unsuitable environments or to users who are already saturated.
There is no ideal structure. The audit seeks the smallest architecture capable of supporting useful economic decisions and exposing enough data to verify what automation is doing.
Output: target architecture and elements to merge, isolate, rename or stop.
3.5. Area 5 — Budgets and bidding
Reconstruct changes in budgets, CPA or ROAS targets, strategies, limits and learning phases. A monthly average can hide ten days of under-delivery followed by rushed spending, or a tranche whose marginal cost already exceeds the financial threshold.
The audit compares average and marginal outcomes. If the first €40,000 produces 800 customers and the next €10,000 only 100, is the portfolio’s average CPA €100 and the last tranche’s CPA also €100? Let us calculate it correctly: €50,000 for 900 customers gives an average of €55.56, while the additional tranche costs €100 per customer. It is this second figure that informs the next increase.
Automated objectives are tested against signal quality and delay. A strategy receiving partial sales data each night, corrected three weeks later, is working with a moving target. It is better to stabilise imports and define a genuinely predictive early indicator.
Output: response curve, alert thresholds and base/high/low scenarios.
3.6. Area 6 — Audiences, queries and inventory
Search-term, segment and placement reports are analysed by value, not clicks alone. The team classifies what matches the offer, what deserves an isolated test, what consumes budget without evidence and what creates brand risk.
An exclusion can save money immediately. It can also block an exploratory query that assisted future sales. The audit therefore recommends an explicit rule, checks conflicts and measures the effect instead of accumulating inherited lists whose origin nobody knows.
For automated audiences, strict controls — legal age, geography, language, existing customers and unserved areas — are separated from suggestions that the platform may expand. A business constraint must not be presented as a mere optimisation preference.
Output: taxonomy of retain, explore, exclude and monitor.
3.7. Area 7 — Creative and speed of learning
The auditor measures creative throughput: new concepts, variants, production time, approval rate, time to fatigue and reused learning. A platform supplied with two near-identical visuals lacks enough diversity to learn the angle, even if it displays fifteen formats.
Every test receives a hypothesis and a unit of learning. “Video versus image” is often too vague; “opening demonstration of the problem versus customer testimonial, for a cold audience, with qualified visits as the primary metric” produces a more transferable conclusion.
Rights, mandatory notices, regulated sectors and brand rules are checked before scaling. A high-performing but non-compliant creative is a liability, not an asset.
Output: concepts × audiences × maturity matrix and production schedule.
3.8. Area 8 — Governance and decisions
The audit concludes with the management system: who changes what, on what evidence, at what frequency, with what ceiling and what record? Access, histories, naming conventions, alerts and emergency procedures should enable rapid intervention without making every change uncontrolled.
Reporting distinguishes daily operations, weekly learning and monthly or quarterly allocation. Asking the same dashboard to detect an outage at 10 a.m. and prove the causal effect of a channel over a year produces an overloaded document that serves neither decision.
Output: RACI, schedule, change log and escalation rules.
4. The Logiks 72-hour method for making the account intelligible
This sequence is a Logiks recommendation, not a universal standard. It is suitable for a first pass over an available scope; complex cycles then require longer analysis and observation.
4.1. Hours 0 to 8 — Secure
Check access, billing, caps, incidents, disapprovals, critical conversions, consent and abnormal spend. Any active leak receives a correction or approved stop procedure before historical analysis begins.
4.2. Hours 8 to 24 — Reconcile
Extract spend, delivery, conversions, CRM data and sales over a period compatible with the cycle. Document definitions, reconcile totals and identify tracking breaks or major changes.
4.3. Hours 24 to 48 — Segment
Analyse outcomes by channel, campaign, intent, audience, creative, page, device, area, new customer and period. Formulate possible causes without confusing correlation with a confirmed diagnosis.
4.4. Hours 48 to 64 — Quantify
Estimate the annual value at stake, confidence level, cost, dependencies and risk of every correction. A definite €500-a-year error should not outrank a well-supported hypothesis involving €200,000 in contribution.
4.5. Hours 64 to 72 — Decide
Present a short list: stop now, repair within two weeks, experiment, investigate further or accept. For every action, name the owner, metric, guardrail, date and expected evidence.
Fast does not mean superficial.
5. Prioritise by impact, confidence, effort and reversibility
A conventional impact/effort matrix overlooks certainty and risk. Logiks recommends scoring four dimensions out of five: economic impact, confidence in the diagnosis, ease of implementation and reversibility. An irreversible or legally sensitive action requires additional validation even if its raw score is high.
For example, correcting a purchase counted twice scores impact 4, confidence 5, ease 4 and reversibility 5. Rebuilding the entire architecture on the basis of one week might score impact 5, confidence 1, ease 1 and reversibility 2. The order becomes obvious.
Recommendations are grouped into waves.
- Wave 0 — Stop loss: leak, non-compliance, false promise or out-of-scope spend.
- Wave 1 — Reliability: conversions, CRM, definitions, alerts and access.
- Wave 2 — Returns: exclusions, budgets, pages, bidding and fulfilment.
- Wave 3 — Learning: new audiences, concepts, channels and lift.
- Wave 4 — Scale: automation, expansion and marginal reallocation.
The numbering protects the sequence. Testing an audience while sales are duplicated produces a quick but unusable conclusion.
6. Worked example: why apparent CPA was sending the team in the wrong direction
A services company spends €60,000 per quarter and reports 600 leads, or €100 per enquiry. The media dashboard appears to meet its €120 target. Reconciliation, however, reveals 72 duplicates, 108 out-of-area contacts and 60 unreachable enquiries, leaving 360 usable cases.
Sales accepts 144 opportunities, closes 36 and generates €2,400 in contribution before advertising per contract. Total contribution reaches €86,400 for €60,000 in media, before creative and tools: the real cost is €1,667 per customer, not €100, and the economic cushion is far narrower than the first screen suggests.
Segmentation then shows that €20,000 allocated to precise intents generates 20 contracts, while €15,000 in retargeting claims 10 contracts but the experiment confirms only three additional sales. The remaining €25,000 produces thirteen signatures across heterogeneous sets, but three are already included in the retargeting credit.
The plan is therefore not to reduce cost per form uniformly. It corrects geography, deduplicates imports, sends CRM qualification back to bidding, caps retargeting, protects the core intent and reserves a budget for two new messages. Every correction has a baseline and a commercial guardrail.
After eight weeks, management will assess usable cases, net cost per customer and marginal contribution, not only improvement in reported CPA. Even if the raw number of leads falls, the system will have improved if valuable sales and economics strengthen.
The example is deliberately simplified. Its logic remains robust: clean the denominator, eliminate duplicate credit, follow the sales progression and decide at the level where the business actually creates value.
7. What a professional deliverable should contain
An actionable audit includes, at a minimum:
- an executive summary with three to five decisions;
- the scope, period, access and limitations;
- metric definitions and source reconciliation;
- findings supported by reproducible evidence;
- an impact estimate with assumptions;
- a roadmap organised into waves;
- an experiment register;
- a table of owners and deadlines;
- the technical appendices required for reproduction.
A recommendations file alone is not enough. The auditor must explain why a decision is needed now, what uncertainty remains and how genuine improvement will be recognised.
8. Logiks recommendations: twelve signs of a superficial audit
- the report judges the account solely by the platform’s automated recommendations;
- net sales, contribution and CRM data are never requested;
- every discrepancy is labelled an error without an impact estimate;
- brand and non-brand remain combined;
- conversions from different platforms are added together;
- one week is used to redesign a seasonal account;
- no rejection or refund case is tested;
- creative analysis stops at CTR;
- recommendations have neither an owner nor a date;
- no consent or data risk is identified;
- the auditor promises a growth percentage without a protocol;
- the deliverable states no limitations.
A good audit reduces uncertainty. It does not disguise it.
9. FAQ
9.1. How long does a paid acquisition audit take?
An initial pass can secure and orient the account in three days, but a complete review often takes two to six weeks depending on the channels, sales cycle, access and data quality. Some hypotheses then require a longer experiment.
9.2. Should campaigns be paused during the audit?
Not as a matter of course. Only a leak, risk or clearly out-of-scope delivery is stopped, in line with agreed authorisations. Everything else is observed, documented and then changed in a controlled sequence.
9.3. Can the incumbent agency take part?
Yes, in an operational audit. Its history accelerates understanding and allows explanations to be tested. The report must nevertheless distinguish verified facts, statements and recommendations.
9.4. What period should be analysed?
Long enough to cover seasonality, conversion delay and major changes — often six to twelve months, with shorter close-up periods. A new or highly volatile business requires different comparisons.
9.5. Does a good optimisation score guarantee a good account?
No. It measures adoption of recommendations defined by the platform, not incremental profitability, CRM quality or fit with the advertiser’s strategy.
9.6. How does this differ from a Logiks independent audit?
This service aims to improve operations and can prepare changes with the teams. An independent audit formalises the mandate, evidence, impartiality, severity levels and separation between review and remediation more rigorously.
10. Conclusion
Auditing paid acquisition means tracing contribution back to every decision that produced it. Economics sets the threshold. Measurement establishes what is observable. Accounts, messages and pages explain the journey. Governance finally turns the finding into action.
Secure first. Then reconcile. Segment, quantify and decide.
The account is only the interface.
11. Main sources
- IAB Europe, AdEx Benchmark 2025, 7 July 2026: https://iabeurope.eu/knowledge_hub/iab-europe-adex-benchmark-2025-report/
- SRI and Udecam, 35th Online Advertising Observatory, 10 February 2026: https://www.sri-france.org/observatoire-epub/35eme-observatoire-de-le-pub/
- IAB, State of Data 2026, 2 February 2026: https://www.iab.com/insights/2026-state-of-data-report/
- Google Ads Help, About campaign experiments: https://support.google.com/google-ads/answer/10682377
- Google Ads Help, About data-driven attribution: https://support.google.com/google-ads/answer/6394265
- Google Ads Help, geographic incrementality measurement: https://support.google.com/google-ads/answer/14102986
- CNIL, Cookie and other tracker recommendation — consolidated version, January 2026: https://www.cnil.fr/sites/default/files/2026-01/recommandation_cookies_consolidee.pdf