How to measure whether managed CSS adds sales

Ask two questions, rather than treating attribution as proof

First, did the managed route deliver the sales that qualify for commission under your agreement? Second, did adding it increase the business’s total sales or contribution compared with what would otherwise have happened? Both matter, but they need different evidence.

An affiliate network can credit an order to a campaign under its attribution rules. That does not show whether the customer would have purchased through another route. Equally, overlap with your existing catalogue does not by itself prove that the sale would have happened anyway.

GMVLY funds and manages a separate Shopping route alongside your core campaigns. You pay the agreed commission on completed sales attributed to that route. The long-tail focus describes the campaign strategy; it is not a finding that every attributed order is incremental.

Define the result your business needs to keep

Choose the outcome before launch. For a commercial review, record approved order count and sales value separately from pending orders, cancellations and returns. Agree how long orders need to mature before a result is considered final.

Your affiliate manager manages validation and commission in the network; its ecommerce and finance teams provide total order and margin data. GMVLY manages its campaigns and uses the reported outcomes for that route. Use a consistent basis for tax, discounts, refunds and currency. Evaluate contribution after product cost, fulfilment, other relevant variable costs and commission. A higher revenue total can still be a poor commercial result if the product mix leaves too little margin.

For example, £10,000 of approved sales with £3,000 contribution before marketing and £1,000 commission leaves £2,000 before any other remaining costs. This is an illustrative contribution calculation, not GMVLY pricing or evidence that £2,000 is incremental profit. A causal assessment also needs the appropriate comparison and relevant costs across the business.

• Primary outcome: total approved sales or contribution for the agreed scope.
• Supporting measures: attributed sales, commission, reversals and order volume.
• Review window: trading period plus the agreed order-validation delay.
• Decision: the scale and contribution that would justify continuing or adjusting the route.

Keep a baseline, but do not mistake it for a control

Record performance before launch so everyone understands the starting position. Include total retailer sales for the agreed scope, existing Shopping spend and results, product availability and the managed route’s launch dates.

A rise after launch is useful context, not proof of cause. Promotions, seasonality, price changes, stock recovery, changes to your own bids and other marketing can all move the result. Keep a dated change log and include these explanations in the review.

Avoid comparing a promotional week with an ordinary week or reporting only the best-performing products after seeing the results. Agree the scope and review dates beforehand, then report the full agreed period.

Choose a comparison that can answer the question

A controlled test compares a group where the additional route runs with a suitable group where it does not. Random assignment, where feasible, helps separate the campaign’s effect from differences in customers or markets. Decide the design with the people who own targeting, tracking and commercial approvals.

Google describes Conversion Lift studies using exposed and held-out groups. Its geography-based approach compares regions and checks whether a study is likely to produce precise results. Availability and eligibility depend on the account and setup; this is a measurement option to investigate, not a GMVLY feature or a promise that every CSS campaign qualifies.

A geographic test needs comparable regions and stable activity elsewhere. Travel, cross-region purchases or other advertising reaching the control can blur the difference. Check targeting and the sales data’s geographic basis before interpreting the result.

A product-level holdout may be worth exploring if campaign scope can be controlled. Similar products are not automatically interchangeable: demand, price, stock and margins differ, and shoppers may buy a substitute in the other group. Evaluate total relevant category outcomes as well as individual product sales, and treat this design’s limitations explicitly.

Give the evidence time to become useful

Agree the test duration and the smallest commercially useful change before starting. Low order counts and normal trading variation can leave a wide range of plausible effects. Report the estimate with its uncertainty rather than presenting a single uplift percentage as a certain outcome.

An inconclusive result does not establish zero impact. It may mean the test lacked enough information to distinguish an effect. Avoid stopping early because one week looks positive, and allow for conversion delays and returns before reviewing the final business outcome.

Where a credible controlled test is not feasible, use the commercial review to decide whether the route merits further evaluation. Describe the result as observed or attributed performance, with the known confounders, rather than claiming measured incremental growth.

Take this worksheet to your launch discussion

Bring the PPC owner, affiliate programme owner and whoever can validate sales and margin data into the same discussion. Confirm what information each party can provide; do not assume that campaign, network and retailer reports use the same definitions.

Agree a feasible measurement plan before launch. The questions below help you identify gaps and decide whether the first review will be a commercial performance review or an incrementality study.

• What Shopping target are we pursuing, and what contribution must we retain?
• Which products, markets and dates are included in the review?
• Which attribution rules, commission terms and return-validation dates apply?
• Has the retailer assigned network validation to your affiliate manager, store data to ecommerce, margin inputs to finance and core campaign records to PPC?
• What comparison could estimate the outcome without the added route?
• Can exposure be controlled, and where could spillover or substitution occur?
• Is the expected order volume sufficient for the effect we need to detect?
• What would lead us to continue, adjust, investigate further or stop?