How to measure UGC ad performance and incremental sales
Report attributed performance first, then use a suitable holdout or lift design to estimate whether the campaign added sales.
Start with spend, reach or impressions, clicks, landing-page visits, and tracked purchases under the account’s stated attribution window. These show delivery and attributed performance. To estimate whether the ad added sales, compare an eligible group that can receive it with a suitable holdout. Attribution, discount codes, and return on ad spend do not by themselves show that the ad caused purchases.
First decide which question you need answered. “Does this ad beat our current creative?” differs from “Does this campaign create extra sales?” A creative comparison can guide the next production order without measuring the campaign's total incremental return.
Choose between a creative test and a lift test
A creative test compares two advertising options. One option might use the named creator and the other a suitable presenter, with the same offer and product evidence. The result applies to the versions and audience tested.
A lift test includes a group that does not receive the advertising being studied. Google describes user-based Conversion Lift as an experiment with a treatment group that can see the ads and a control group held back from them. The product is not available to every account, and eligibility varies. Google’s Conversion Lift measurement guide, setup and eligibility.
If both groups receive ads, do not report their difference as proof of all sales caused by advertising. A holdout that compares a whole creator campaign with no campaign does not isolate the creator's face from the script, media, or offer.
Write the question as a sentence that names the alternative. This keeps a useful test from supporting a claim it never examined.
Set the business outcome before launch
Pick a main outcome that matches the decision. An ecommerce brand may choose contribution from new customer orders. A subscription brand may care about retained paid accounts. A business selling to other businesses may need qualified opportunities rather than form submissions.
Keep attention and traffic measures as supporting information. They help explain where a result changed. An ad can earn clicks from the wrong buyers, so attention alone should not decide a sales claim.
Agree on the conversion window, how to treat cancelled orders, whether tax and shipping sit in revenue, and when delayed conversions will be counted. Use the same definitions in both groups. Record tracking failures so they remain visible when the team reads the result.
Build a comparison the team can interpret
Use a supported experiment tool or a measurement specialist when possible. Random assignment reduces the risk that the treatment group already differs in purchase intent. Keep the non-test settings stable and record changes that cannot wait.
Research by Gordon and colleagues compared observational measurement with large Facebook experiments. It found that observational approaches could fail to recover the experimental effect. The study supports caution about causal claims from campaign reports; it does not measure the value of AI creator identity. Marketing Science paper.
Consider what could contaminate the comparison. A creator may post the same message organically, an email campaign may reach both groups, or another campaign may show the tested ad to the holdout. Some overlap is unavoidable. Record it and narrow the conclusion when needed.
Calculate the result using the right denominator
Here is a simplified hypothetical example of a properly randomized test with equal-sized groups: 10,000 people can receive the ad and 10,000 are held out. The ad group records 240 purchases (2.4%); the holdout records 200 (2.0%). The raw difference is 40 additional purchases per 10,000 people, or a 0.4 percentage-point difference in purchase rate. This example gives no uncertainty interval, so it cannot establish that the difference is reliable.
If each estimated additional order contributes C$40 before campaign costs, the 40 orders represent C$1,600 in contribution. If the total added campaign cost is C$2,000, that is C$400 below cost for this hypothetical test period. Reporting all 240 purchases as incremental would overstate the result.
For unequal groups, compare conversion rates and scale them to the relevant population; do not subtract raw counts. Where a platform supplies a lift estimate and an uncertainty range, use its defined method rather than recreating a rough count calculation and presenting it as the official result.
Incremental contribution after campaign cost = estimated additional contribution − media − creative − other added campaign costs.
Avoid subtracting a cost twice if your contribution definition already includes it.
Record what the test supports
Use a short report to keep the conclusion tied to the design:
| Field | Record |
|---|---|
| Question and comparison | The campaign, creator version, alternative, and holdout, if used |
| Assignment | Who could receive each version and how the groups were assigned |
| Outcome and window | Purchase or contribution definition, cancellations, and conversion period |
| Estimate and uncertainty | The estimated difference and any uncertainty measure the tool provides |
| Limits and costs | Other campaign exposure, tracking issues, and all added campaign costs |
If the tool does not provide an uncertainty measure, say so. Do not turn a difference in raw counts into a confident claim by itself. For a creator-count question, see how to compare one creator with several; for a client-facing record, use the AI creative work report guide.
Give the test enough time and information
There is no universal budget that makes a creator test reliable. Required information depends on normal conversion rates, outcome variability, the smallest worthwhile gain, and the test design. Ask for a power or feasibility check before agreeing to a costly test.
Google's user-based lift guidance says smaller holdouts can require longer studies and that duration should account for conversion lag. A short window can miss purchases that take longer to occur. Google lift setup guidance.
Set a planned end date and rules for material tracking failures. Repeatedly checking for a winner and stopping on a favourable day can produce a misleading result unless the analysis accounts for that stopping rule. Report the experiment design and uncertainty in the client report guide.
What if the brand is too small for a lift study?
Use a controlled creative comparison to answer a narrower question. Combine its result with actual production cost, customer feedback, and operational reliability. Label the finding “better observed performance in this test,” with the limits that apply.
You can also test whether buyers understand the message or recognize the creator before funding a sales experiment. Those checks may help assess the message, but they do not substitute for sales evidence.
Use the result to decide whether to continue, improve the test, or stop. Record uncertainty when the data cannot show whether a meaningful difference exists.
Connect this decision to the campaign
The proposed pilot needs an evaluation design before launch: outcome, control, comparable offer and spend, reporting window and known limits. Keep platform attribution separate from added sales when sending results to the team.