Hypothesis
If the same video runs as a Spark Ad from the creator's account instead of a standard ad from the brand account, it will deliver results at a lower cost, because native identity and the post's social proof raise engagement; if not, the authorisation and creator fees buy nothing. The decision is a spending one: whether to pay creators for authorisation and run Spark Ads, or to buy only the video and run it from the brand.
The hypothesis isolates something most creator tests blur. When a brand compares creator videos with studio videos, it changes the content, the person, the style and the identity at once, and a win cannot say which mattered. Here the content is the same file. What differs is the handle above it, the profile a tap leads to, and the likes and comments that sit under a published post. The Spark Ads guide explains what carries over from the organic post into the ad and what does not, which is exactly the difference this test prices. Write the creator, the post, the fee and the result event into the experiment hypothesis template before any budget is set.
Metrics and guardrails
The primary metric is cost per result: spend plus the identity-related creator fee, divided by the results the ad group optimises for, whether purchases, leads or another event. It is a cost per acquisitionCost of each conversion. by another name, and the CPA calculator handles the arithmetic once the fee is added to spend. Check the platform's count against the business's own records before reading it, because the decision rests on customers who exist in the order system, not on events the pixel reported.
Three guardrails protect the account and the creator. Engagement RateShare of sessions that were engaged; on social, interactions relative to reach. per arm shows whether the identity is doing what the hypothesis claims; a Spark arm that wins on cost with no rise in engagement is winning for some other reason, and the reason should be found before money moves. Comment sentiment on the creator's post matters because the comments the ad attracts stay under the creator's name; a cheaper result bought with a thread of complaints is a cost to the relationship. The third guardrail is a hard condition rather than a number: every creator post in the test carries a clear paid-partnership disclosure. The FTC's endorsement guides answer common questions about when a material connection between a brand and an endorser has to be disclosed, and the TikTok help centre documents the platform's own branded content setting; neither replaces the other.
Design
Use the platform's split test so that each identity reaches its own randomised audience. Create one campaign with a split test on identity, as the campaign structure guide describes at the campaign level, and build two ad groups that are copies of each other in objective, optimisation event, targeting, placements, bid strategy and budget. One ad group runs the creator's post as a Spark Ad through an authorisation code; the other runs the same video file, uploaded to Ads Manager, under the brand's identity with the same caption and call to action.
| Design element | Choice | Reason |
|---|---|---|
| Unit of randomisation | User, assigned by the platform's split test | Prevents the delivery system favouring one arm |
| Arms | Spark Ad from the creator's account against a standard ad from the brand identity | Identity is the only change |
| Held constant | Video file, caption, call to action, landing page, targeting, budget, event | Isolates the handle and the post's social proof |
| Primary metric | Cost per result with the identity fee added to spend | The fee is what the decision buys |
| Guardrails | Engagement rate, comment tone, disclosure on every post | Protect the account, the creator and compliance |
| Authorisation | Period agreed in writing to cover the whole test and a margin | Expiry stops delivery mid-test |
Choose the video before the test, not after. Pick one that suits the objective and the result event, as the hooks and user-generated content guide describes for briefing creators for paid use, and avoid a post whose opening is weak, because the same weakness will drag both arms and shrink any identity effect. Running one video answers the question for that video; repeating the test with a second creator answers it for the programme.
Sample size reasoning
Engagement is plentiful but the decision rests on results, so size the test on the result rate and the smallest cost difference that would justify creator fees. Likes and views arrive in volume and will separate quickly, which is why they tempt teams into early calls. Results are rarer, and the test needs enough of them in each arm to tell a real cost gap from week-to-week noise. Work backwards from the fee: the smallest gap in cost per result that would pay for authorisation across the planned spend is the minimum effect worth detecting. The sample size and significance guide explains how the baseline result rate and that minimum effect set the volume each arm needs. If the budget cannot reach it, use a higher-volume result event as a proxy and say so in the plan.
Duration
Run through the learning phase and at least two full weekly cycles, within the authorisation period agreed with the creator. Set the learning period aside rather than reading it, since both ad groups start from nothing and settle at different speeds. Confirm the code's expiry date before launch and agree an authorisation period that covers the test, a buffer for delays and the scale-up that follows a win. Fix the end date in advance and do not extend it because one arm has drifted ahead.
Reading the results and the decision rule
Read once, at the planned end. Compare cost per result including fees between the arms with its interval, then check engagement, comment tone and disclosure. Record the post's engagement at launch so that the reading can say how much of the Spark arm's advantage came from proof the post already had. Write the decision rule into the plan before launch:
| Result | Decision |
|---|---|
| Spark arm's cost per result including fees is lower and guardrails hold | Fund Spark Ads for this creator and test the next one before scaling the programme |
| Spark arm is cheaper before fees but not after | Run the video from the brand account and keep creator budgets for content rather than identity |
| No reliable difference | Run the video from the brand account; the authorisation fee buys nothing measurable |
| Spark arm wins but comments turn hostile or a disclosure lapsed | Pause, fix the guardrail, and rerun before deciding |
Pitfalls
The organic post's existing engagement can favour the Spark arm. A post with a large organic following arrives in the feed with likes, comments and shares that a freshly uploaded ad cannot match. That is part of what a Spark Ad buys, but it means a win on an already popular post does not prove that every creator post will beat the brand. Prefer a post published shortly before the test, or record its engagement at launch and limit the conclusion accordingly.
Missing disclosure is the second pitfall and the most serious. A paid creator post without a clear disclosure is a compliance problem whatever the test shows, so check every post at launch and again during the run. The third is authorisation expiring mid-test: the creator's code lapses or the post is made private, and delivery in the Spark arm stops while the brand arm keeps spending. The troubleshooting page on a Spark Ads code that stops working lists the causes; check them before launch. Finally, watch for Creative FatigueThe decline in an ad's response as the same people see it again and again. in both arms on long runs, because the same video served to the same audience tires at a similar pace and can flatten the difference late in the test.
A skincare brand prices a creator's handle
Quillmoss Skincare, an invented online skincare brand, pays a creator £400 for a cleanser video and is offered a 60-day Spark authorisation for a further £600. It runs a split test for 21 days with £3,500 per arm, optimising for purchases. The Spark arm records 140 purchases and the brand arm 118, matched against the order system. Cost per purchase is £25.00 in the Spark arm and £29.66 in the brand arm before the authorisation fee; adding the £600 fee to the Spark arm's spend gives £29.29. The gap is inside the noise, engagement is higher in the Spark arm, and disclosure held throughout. Quillmoss runs the video from its own account and spends the next creator budget on two more videos instead of authorisation.
Common mistakes
Spark Ads tests go wrong when they change more than the identity or read the cheap signals instead of the expensive one. The ad creative hook test varies the video's opening instead, which is the natural next test once the identity question is settled, and the experiment library lists every design. Creators are often the source of user-generated contentContent about a brand made by customers or creators rather than by the brand itself. in the brand's other channels too, so a result here should inform how those rights are bought.
- Changing the video along with the identity. Re-edits or new captions make the test a creative test.
- Leaving the fee out of the cost. The fee is exactly what the decision buys.
- Reading likes and views. Engagement separates fast; results decide.
- Putting both ads in one ad group. The delivery system then chooses who sees which.
- Letting the authorisation lapse. An expired code ends one arm while the other keeps spending.
- Skipping the disclosure check. A win on an undisclosed post is not a result anyone can use.
