A paid media audit should tell an e-commerce team what decision is safe to make next. A list of account settings is not enough. If store revenue, customer definitions, conversion goals, product availability, and contribution economics are unresolved, a technically correct campaign change can make the system more efficient at pursuing the wrong outcome.
This guide is for a DTC founder, head of growth, paid-media lead, or finance partner reviewing active spend. It orders the audit by dependency: establish commercial truth first, verify what the platforms are learning from, and only then judge campaigns, creative, and landing pages.
The short answer: audit the evidence chain, not the interface
An e-commerce paid media audit should follow eight gates:
preserve the evidence window and access;
reconcile store sales and reversals;
verify the conversion actions used for bidding;
define contribution and customer-acquisition boundaries;
check catalog, inventory, and offer eligibility;
assign each channel and campaign a commercial job;
inspect whether creative testing produces reusable learning;
test the path from click to purchase.
Do not score all findings equally. Stop budget changes when the business outcome or bidding input is unreliable. Repair eligibility and purchase-path defects before interpreting weak demand. Test channel or creative hypotheses only after the earlier gates are usable. The output should be a prioritized decision record, not a generic health score.
Define the audit decision before opening an ad account
An audit can support several different decisions:
whether to increase, hold, or reduce total media investment;
whether a channel deserves another test;
whether an agency or internal team is operating with adequate evidence;
whether poor reported performance comes from demand, measurement, merchandising, creative, or conversion;
whether an account rebuild is justified;
which constraint must be removed before a campaign can be evaluated fairly.
Write one decision at the top of the brief. “Improve performance” is not specific enough. A more useful question is: “Can we release another month of acquisition budget, or must we first repair purchase measurement, product availability, or creative learning?”
The decision determines the evidence window. A review of a seasonal promotion may need a comparable prior period and promotion calendar. A new-customer acquisition review needs a stable customer definition and enough time for cancellations and returns to mature. A creative review needs concept and launch records, not only asset-level delivery.
Gate 1: preserve a frozen evidence window
Before changing settings, record the current state. Otherwise the audit itself destroys the baseline it is trying to understand.
Capture:
audit date, market, currency, time zone, and comparison period;
store, analytics, ad-platform, product-feed, creative-library, and finance access available;
campaign budgets, bid strategies, conversion goals, attribution settings, exclusions, and major automated controls;
product availability, price, promotions, shipping terms, and known site incidents;
releases, tracking changes, consent changes, feed migrations, and agency handoffs during the period;
known reporting delays for returns, chargebacks, cost of goods, and fulfillment.
Use exports or screenshots where the platform permits them, but do not treat a screenshot as a data model. Keep the source, extraction time, scope, and owner beside every artifact. If the team lacks access to a critical source, record that as an audit finding rather than filling the gap with an estimate.
A frozen baseline does not mean the account must remain untouched during an emergency. Fraud, policy risk, broken checkout, runaway spend, or widespread product disapproval may require immediate containment. Record the containment action separately so later analysis does not confuse it with ordinary optimization.
Gate 2: reconcile commerce truth before platform attribution
Begin with orders and sales in the commerce system. Platform attribution cannot establish how much money the store actually kept.
For a representative period, reconcile at least:
Field Definition to settle Why it changes the audit --- --- --- Orders Placed, paid, fulfilled, or another governed state Pending and canceled orders can inflate an early view Gross sales Product value before discounts and reversals Useful for merchandising, weak as a profit boundary Net sales Gross sales less discounts and sales reversals Closer to realized product revenue Total sales Net sales plus shipping, tax, duties, and applicable fees Can differ materially from the value used in an ad platform New customer First-ever order, first order in a lookback, or another rule Changes acquisition CAC and cohort comparisons Returns and reversals Event date, original-order date, or cohort date Determines when weak product or customer quality appears Shopify currently defines net sales as gross sales minus discounts and sales reversals, while total sales also includes items such as tax, duties, shipping charges, and fees. It notes that reversals can appear on the date they are processed rather than the original order date (Shopify Help Center). That timing alone can make an ad-period report and a finance-period report disagree without either being fabricated.
Shopify marketing reports also let merchants change attribution models. The default marketing view uses last non-direct click, while other available views allocate credit differently. Shopify warns that an any-click view can award credit to more channels than the number of orders received and is best used for examining one channel or reconciling channel claims (Shopify Help Center). Attribution is therefore a lens on interactions, not the sales ledger.
Take a small sample of orders across paid channels and trace them through store, analytics, and platform reports. The goal is not perfect equality. It is an explained variance: different attribution windows, time zones, modeled conversions, identity coverage, canceled orders, refunds, taxes, shipping, and duplicate events should be visible rather than silently blended.
Gate 3: verify the signals that control bidding
An event can appear in a report without being used for optimization. Conversely, an apparently secondary action can influence bidding when it is included in a custom goal. Inspect the actual campaign-to-goal relationship.
Google Ads distinguishes primary conversion actions, which can enter the Conversions column and bidding when their goal is selected, from secondary actions intended for observation. Google documents an exception for secondary actions included in custom goals: those actions can be used for bidding (Google Ads Help). The audit must therefore check the campaign goal, conversion action, action setting, count method, value, and source together.
For every biddable action, record:
business meaning;
firing condition;
source and transport path;
transaction or event identifier;
count rule;
dynamic value and currency behavior;
refund or adjustment treatment;
consent and modeled-data boundaries;
delay from action to platform availability;
campaigns that use the action.
Then run a bounded purchase test in a safe test environment or use existing verified test orders. Do not place a real customer order or trigger a live advertising conversion merely to complete editorial or documentation work. An audit for an operating brand should use the account's approved QA procedure, preserve test identifiers, and ensure test commerce is excluded from business reporting.
Classify the result:
Verified: event meaning, value, deduplication, and campaign use are supported by evidence.
Partially verified: delivery is visible but value, coverage, or adjustment behavior is unresolved.
Conflicted: multiple sources claim the same purchase or different outcomes share one goal.
Unavailable: the team lacks access or a reliable test seam.
Do not tune automated bidding around a signal labeled conflicted or unavailable. That is a measurement repair, not a media optimization problem.
Gate 4: establish the economic boundary
A platform can report strong return while the store loses contribution after product cost, fulfillment, discounts, returns, payment fees, and agency or creative costs. The audit should not invent a universal acceptable ROAS. It should identify the economic definition that the business will use.
A simplified boundary is:
contribution before media = net revenue − variable product and fulfillment costs
contribution after media = contribution before media − paid media − other variable acquisition costs
The exact fields depend on the company. Document whether the decision uses first-order contribution, cohort contribution, or another finance-approved view. Keep forecasts separate from observed outcomes. Use the e-commerce contribution-margin framework when the business definition is not settled.
Next, reconcile customer mix. A blended platform CPA can fall while new-customer CAC rises if returning customers represent more attributed purchases. Shopify customer reports provide cohort, customer, order, marketing-channel, and one-time-versus-subscription views, but Shopify notes that some metrics may be unavailable for a particular cohort or period (Shopify Help Center). Missing fields should remain missing; they should not become zero.
The audit should show at least three layers:
platform-attributed cost and value;
store orders, net sales, and customer status;
finance-approved contribution and maturity.
Our blended CAC versus platform CPA guide covers the reconciliation between layers. The audit's job is to decide whether those layers are reliable enough to release budget.
Gate 5: check product, inventory, and offer eligibility
Weak delivery is not always weak demand. A product can be unavailable, disapproved, mapped to the wrong variant, excluded from a campaign, out of stock, uncompetitive after shipping, or sent to a landing page with inconsistent price or availability.
For Google Shopping inventory, inspect Merchant Center before judging campaign structure. The current Needs attention area surfaces account- and product-level issues, affected products, product status history, and issue-detail views. Google says suggested values and automatic fixes should be checked against the landing page before acceptance (Google Merchant Center Help).
Google's product-data specification says inaccurate or missing information can cause limited eligibility, disapprovals, or incorrect display, and it identifies variant, identifier, image, category, price, availability, and landing-page consistency among the relevant inputs (Google Merchant Center Help). Treat the store or governed catalog as the commercial source; do not let a convenient platform suggestion overwrite correct product truth.
Build a product eligibility map:
Product or group In stock? Margin known? Feed eligible? Landing page accurate? Active media role --- --- --- --- --- --- Hero products Acquire, defend, or test High-margin products Scale if demand and capacity support it Low-stock products Constrain or exclude New products Learn with bounded budget Return-prone products Investigate quality and audience fit Do not optimize only the products that already receive spend. Inspect eligible products receiving no spend, spend on unavailable products, and products whose reported revenue hides weak contribution or high reversals.
Gate 6: assign every channel and campaign a job
Channel comparisons fail when two systems receive different jobs and are judged by one attribution column. Search may capture existing intent. Paid social may create and retarget demand. Shopping may match products to queries. Video may support discovery or creative learning. The audit should define the intended role before comparing outcomes.
Use a campaign contract:
Field Audit question --- --- Buyer state What does the customer already know or want? Commercial job Acquire, capture, recover, defend, launch, or learn? Eligible products Which products and inventory states can participate? Primary signal Which verified action controls bidding? Budget boundary Which economic limit or experiment cap applies? Evidence window When is the result mature enough to review? Stop condition Which evidence pauses or changes the job? Look for duplicate jobs, unexplained overlap, brand demand masking acquisition, and campaigns that cannot state what they are allowed to learn. Do not require one universal account structure. Consolidation can help an automated system learn, but separation can be necessary when economics, geography, inventory, offers, or decision rights differ.
Use the Google Ads versus Meta Ads decision framework when the unresolved question is channel mix. Use the Performance Max audit when the issue is specifically whether a PMax campaign should be kept, fixed, constrained, or replaced.
Gate 7: audit creative learning, not asset volume
A folder full of ads does not prove a creative system exists. Group assets by concept, audience problem, product promise, proof type, offer, format, and landing destination. Then ask whether the team can explain what was tested and what changed because of the result.
Separate:
a concept, the core reason or story being tested;
an execution, the specific hook, visual, copy, creator, or edit;
a variation, a bounded change inside an execution;
an export, a resized or reformatted file that may add no new hypothesis.
For each concept, record launch date, intended audience, spend opportunity, delivery, destination, product availability, result maturity, and next decision. A concept that received little delivery is not automatically a failed message. An ad with strong attributed return is not automatically incremental or reusable.
The e-commerce creative-testing framework provides a deeper learning-system design. In this audit, the pass condition is simpler: creative history must support a decision about what to repeat, revise, retire, or test next. If it cannot, the finding is an operating-process gap.
Gate 8: test the post-click purchase path
Do not stop at the ad account when the purchase happens elsewhere. Review representative mobile and desktop journeys from ad to product page, variant selection, cart, checkout, payment, confirmation, and any subscription or bundle choice.
Check:
message and offer continuity from ad to destination;
price, promotion, variant, stock, shipping, and return clarity;
page speed and functional errors on real target devices;
consent and tracking behavior across relevant states;
navigation or pop-ups that obscure the intended action;
checkout payment and error recovery;
confirmation behavior and duplicate-event risk;
accessibility barriers that prevent selection, form completion, or payment.
A landing-page issue should be supported by observed friction, not by a generic preference for shorter copy or a different color. Our paid-traffic CRO guide explains how to connect message match and conversion work to qualified outcomes rather than form or click volume.
Keep an evidence ledger for every finding
Use one record format across the audit:
Field What to record --- --- Observation What was directly seen? Source Store, finance, platform, feed, analytics, QA, or interview Scope Product, campaign, market, device, date range, and customer state Risk Which decision could be wrong? Confidence Verified, partial, conflicted, or unavailable Owner Who can verify or change it? Next action Keep, verify, repair, test, constrain, or stop Acceptance evidence What would close the finding? This prevents recommendations from outrunning evidence. “Meta is inefficient” is not an observation. “Store new-customer orders declined while Meta-attributed purchases rose, and the customer-status join is only available for 62% of orders” is an observation with a defined gap. The next action is to repair or bound the join before moving budget on the basis of that comparison.
Prioritize findings by decision risk and reversibility
A long audit becomes useful only when the team knows what happens first. Use five action classes.
Action Use when Example --- --- --- Stop Continuing creates financial, policy, customer, or data risk Checkout is broken or bidding uses duplicate purchase values Repair A prerequisite is defective Product price and availability conflict with the feed Verify A material claim lacks evidence New-customer reporting cannot be reconciled to store records Constrain The system can continue inside a safer boundary Keep a campaign live with a capped test budget while returns mature Test Prerequisites are usable and a decision hypothesis remains Compare two creative concepts under a defined product and audience job Rank within each class using business exposure, confidence, reversibility, and time to evidence. Do not rank by how many checklist items a finding touches. One duplicate purchase event can matter more than twenty naming or organization inconsistencies.
A fictional triage example
Consider a hypothetical apparel store whose platform ROAS improved while finance reports weaker contribution. This is an illustrative example, not a Sharply Labs client result or benchmark.
The audit freezes an eight-week window and finds four issues:
the store's total sales include shipping and tax, while one platform receives product revenue before those fields;
a second purchase conversion is primary in several Google campaigns;
the highest-spend product group has frequent stock gaps and a high sales-reversal rate;
the creative report counts 60 exports but only three distinct concepts.
The team does not start by moving budget between Google and Meta. It first maps the two purchase actions, verifies which campaigns use them, and removes only the confirmed duplicate from bidding through the approved account process. It reconciles a representative order sample and establishes one contribution definition. It constrains spend on the unstable product group while merchandising addresses stock and returns. Finally, it rebuilds the creative log around concepts rather than exports.
Only after those repairs does the channel comparison become decision-ready. The audit has not promised better performance. It has made the next budget decision less dependent on mislabeled evidence.
A 30-day remediation order
Days 1–5: protect and reconcile
Freeze the evidence window, confirm access, record material changes, trace representative orders, define customer status, and agree on sales and contribution fields. Contain only urgent risk.
Days 6–10: repair bidding and eligibility inputs
Map campaign goals and biddable actions, validate value and deduplication, inspect product/account issues, reconcile price and availability, and identify products that should not receive spend.
Days 11–20: clarify jobs and learning
Assign channel and campaign contracts, separate acquisition from demand capture where evidence supports it, rebuild the creative concept log, and map destinations to offers and products.
Days 21–30: run bounded changes
Ship the smallest high-confidence repairs, define acceptance evidence, and launch only the tests whose prerequisites are usable. Record what remains unavailable. Set the next review date based on outcome maturity rather than a universal cadence.
What a useful agency audit deliverable should contain
A buyer should receive more than a presentation. A decision-ready package includes:
the audit question and scope;
access and evidence limitations;
commerce and contribution definitions;
campaign-to-conversion-goal map;
product eligibility and inventory exceptions;
channel and campaign job map;
creative concept and destination map;
evidence ledger with owners and confidence;
prioritized stop, repair, verify, constrain, and test actions;
acceptance evidence for each material action;
a change log and handoff path.
Reject universal scores that hide missing data. Challenge recommendations that change many settings at once, erase the baseline, or rely only on platform-attributed return. An audit can recommend no immediate campaign change when the evidence says measurement, merchandising, or checkout must be repaired first.
When the audit cannot conclude
Keep the conclusion conditional when:
return and cancellation data has not matured;
new-customer identity is incomplete or inconsistent;
cost of goods or fulfillment inputs are unavailable;
a major promotion, site incident, stockout, or tracking migration changed the period;
consent or attribution coverage differs materially by market;
an account lacks reliable change history or creative records;
a platform or store report changed definition during the comparison;
spend or conversion volume is too limited for the proposed cut.
The correct output may be a verification plan, not a channel verdict. “Unknown” is safer than false precision and more useful when it names the owner and evidence required.
The practical next step
For established DTC and e-commerce teams with active paid media, a Sharply Labs e-commerce growth review can examine the store and finance definitions, campaign bidding inputs, catalog eligibility, channel jobs, creative learning, and purchase path together.
The output is a prioritized decision map: what to keep, what to verify, what to repair, what to constrain, and which bounded test can run next. It does not promise lower CAC, higher MER or ROAS, incremental lift, or revenue growth.
Before the review, export one stable period of store orders and reversals, campaign and conversion-goal settings, product issues, creative history, and the finance definition that currently authorizes spend. The gaps in that packet are already part of the audit.
Sources
Marketing reports — Shopify Help Center
Sales reports — Shopify Help Center
Customers reports — Shopify Help Center
About primary and secondary conversion actions — Google Ads Help