What our Commercial Audit actually looks like

Section 01

Account snapshot & tracking integrity findings

Before any structural work, we establish whether the numbers the account is optimising against are true. In this account they were not. Reported conversion value was overstated by roughly 18%, which means every bid strategy in the account had been trained on inflated feedback for at least nine months.

• Finding 1.1 — Duplicate purchase events

  The GA4 purchase event and the Google Ads global site tag were both firing on the order confirmation page, with a second fire on browser back-navigation. Roughly 11% of orders were counted twice. Deduplication via transaction_id was absent.

• Finding 1.2 — Conversion value includes VAT, shipping and discounts

  Purchase value was passed at gross order total. Stripping VAT, delivery income and applied discount codes reduces true trading value by 16.4%. Every ROAS target in the account was therefore set against a number that does not exist in the P&L.

• Finding 1.3 — Refunds and returns never fed back

  Return rate across the catalogue is 9.8%, rising to 21% in the lighting category. No refund adjustment was uploaded to Google Ads, so Smart Bidding continued to bid up the highest-returning products.

• Finding 1.4 — Enhanced conversions inactive

  Enhanced conversions for web was enabled but not receiving hashed customer data, so match rates sat at 0%. Estimated 6 to 9% of conversions unattributed on iOS.

• Finding 1.5 — Three primary conversion actions counted together

  Purchase, newsletter signup and 'add to basket' were all set to primary. Bidding was optimising towards a blended value in which a free email signup was worth £14 of modelled value.

• Finding 1.6 — No margin data in the feed

  No custom label carried cost of goods. Without it, no bid strategy in the account can distinguish a 62% margin candle from a 9% margin air fryer.

Section 02

Break-even ROAS by margin band

A single account-wide ROAS target of 5.0x was applied to a catalogue whose gross margin ranges from 9% to 64%. Break-even ROAS is simply 1 divided by gross margin, adjusted for returns and payment fees. Once calculated, the account's flat target is revealed as simultaneously far too soft on high-margin lines and structurally loss-making on low-margin ones.

• Finding 2.1 — 40% of spend sits below break-even

  Bands D and E together consume £18,400 per month and return revenue at a ROAS beneath the point at which the order makes money. Every incremental sale in band E costs the business roughly £6.20 in gross profit before overhead.

• Finding 2.2 — The best products are being starved

  Band A clears break-even by 3.8x and is impression-share limited at 41% lost to budget. This is the single largest uncaptured profit pool in the account.

• Finding 2.3 — Returns are not priced into targets

  Lighting looks acceptable at 5.1x until a 21% return rate is applied. Net of returns it delivers 4.03x against a 4.1x break-even, which is a rounding error away from zero.

Section 03

SKU-level profit analysis

Every SKU is given a commercial job under BOI® (Bid On Intent): Scale, Profit, Protect, Recovery or Gateway. The job determines the target, not the other way round. Below is a representative extract from the full 2,300-line workbook you receive alongside this document. Figures are 90-day.

• Finding 3.1 — Two SKUs consume 18% of total spend at a loss

  The 5L and 8L air fryers absorb £103,100 of 90-day spend and return £75,000 of gross profit. They are treated as hero products because they top the revenue report. On profit they rank 1,847th and 2,011th of 2,300.

• Finding 3.2 — The top ten profit SKUs receive 6% of spend

  Ranked by absolute gross profit per pound of spend, the top ten lines are collectively budget-constrained. Nine of the ten sit in Shopping campaigns with no priority separation.

• Finding 3.3 — 612 SKUs have spent with zero conversions in 90 days

  £31,400 of cumulative spend, no orders. 340 of these are disapproved or have missing GTINs and should never have been served.

• Finding 3.4 — Gateway logic is untested

  Air fryers are defended as a customer acquisition route. Cohort analysis of 2024 first-order air fryer buyers shows a 12-month repeat rate of 9.1% against a site average of 27.4%. They are not a gateway, they are a discount magnet.

Section 04

The three biggest profit leaks, costed

Every leak below is stated in monthly gross profit, not revenue, and every one is recoverable inside 90 days without additional budget.

• Leak 1 — Brand cannibalisation, £9,400 per month

  Performance Max is claiming 71% of brand search conversions at a £0.94 average CPC on terms that historically converted at 14x through exact-match brand Search. The incremental value of paying for a customer who typed the brand name is close to zero. Excluding brand from PMax and holding it in a capped brand Search campaign recovers the difference immediately.

• Leak 2 — Structural loss-making spend, £7,900 per month

  Bands D and E deliver revenue at a ROAS below break-even. This is not an optimisation problem, it is a mandate problem: the account was told to buy revenue. Reallocating this spend to band A and B ceilings, which are impression-share limited, converts the same media budget into positive contribution.

• Leak 3 — Return-blind bidding, £4,600 per month

  Lighting returns at 21%. Smart Bidding sees the sale, never the refund, and compounds the error by bidding harder on the products most likely to come back. Uploading refund adjustments and re-baselining targets on net revenue corrects the feedback loop.

Section 05

Brand-traffic split in Performance Max

Performance Max reports a 9.2x ROAS and is described in the current agency's monthly report as the account's strongest channel. Separating brand from non-brand tells a different story.

• Finding 5.1 — 58% of PMax revenue is brand

  Brand queries account for £121,600 of the £209,300 attributed to PMax. Strip them out and the campaign's reported 9.2x becomes 5.5x on the non-brand portion, before incrementality adjustment.

• Finding 5.2 — Brand exclusions were never applied

  No brand exclusion list exists at account or campaign level. This is a two-click change that has been available for the entire duration of the engagement.

• Finding 5.3 — Display and video placements are draining the asset group

  15% of PMax spend is going to low-intent placements returning 2.3x. There is no separate reporting on this in the monthly deck.

• Finding 5.4 — Search Partners and Display are inflating the top line

  Removing brand and low-intent placements would drop reported account ROAS from 5.9x to roughly 4.1x on paper, while increasing actual contribution. Any agency unwilling to make a number look worse in order to make the business more money is optimising for the report.

Section 06

Structural recommendations

Sequenced across 90 days, lowest risk first. Nothing in weeks 1 to 4 touches bid strategies during a trading peak; measurement and exclusions come before restructure.

• What we would not do

  We would not restructure the account during peak trading weeks, would not test Demand Gen until measurement is trustworthy, and would not chase the reported ROAS number back up. It should fall before contribution rises.

• What needs to come from your side

  Cost of goods at SKU level, refreshed monthly. Return rates by category. Confirmation of the contribution margin floor the business needs after overhead. Without COGS, no agency can do this work honestly, including us.

Section 07

The straight recommendation