The creator who produced a Rp 119 million video also produced one that sold nothing.
Same person. Same product line. Same three-month window. One video moved 5,580 units. The other moved zero across 4,161 views.
We found that pattern repeatedly while auditing affiliate performance across nine FMCG product lines in Indonesia. Of 92 creators in the sampled set, 23 appear in both the top-GMV list and the zero-revenue list. A quarter of the roster produced the company's best results and its worst results.
That number breaks the assumption most affiliate programs are built on.
The Roster Theory
Almost every affiliate program treats creator selection as the primary lever. Recruit better creators, vet harder, build a tier list, cut the bottom performers. The logic assumes that revenue variance sits between creators: some are good at this, some are not, and your job is to find the good ones.
The data does not support that assumption. Revenue variance sits inside individual creators at roughly the same scale it sits between them.
| Creator | Top-GMV videos | Combined GMV | Zero-revenue videos | Product lines won |
|---|---|---|---|---|
| Creator A | 7 | Rp 84,749,838 | 3 | 4 |
| Creator B | 4 | Rp 50,508,664 | 3 | 3 |
| Creator C | 3 | Rp 34,070,404 | 2 | 2 |
| Creator D | 2 | Rp 17,294,956 | 2 | 2 |
| Creator E | 2 | Rp 12,248,840 | 3 | 1 |
| Creator F | 1 | Rp 119,104,499 | 1 | 1 |
Creator A is the clearest case. Seven videos in the top-GMV lists across four different product categories, worth Rp 84.7 million combined. Also three videos that sold nothing. If you built a tier list from the winners, Creator A sits at the top. If you built it from the failures, Creator A is on the cut list.
Creator F is the extreme version. One video at Rp 119 million, the single highest-earning asset in the entire dataset. One video at zero.
What Concentration Actually Tells You
The revenue distribution looks like every affiliate program you have seen. Ten creators, 17.9% of the ranked pool, produced 78.2% of tracked GMV.
The usual read on that number is that talent is concentrated. Find the 18%, pay them more, stop wasting budget on everyone else.
But 6 of those 10 top-earning creators also appear in the zero-revenue sample. Their concentration is real at the portfolio level and unstable at the video level. Paying them more does not buy consistency, because their own output is not consistent.
Ten creators did win across two or more product lines, and one won across four. That looks like transferable skill, and some of it probably is. But those same creators failed inside the categories they otherwise won.
What Changes Inside a Creator
If the person is constant and the outcome swings from Rp 119 million to zero, the variable is not the person.
The teardowns pointed at three things that changed between a creator's winners and their failures.
The product they were given. Creator A won on pest control and home care, and produced zeros on lines with longer consideration cycles. The same delivery style that closes a rat-glue purchase does not close a baby-care purchase.
The offer they were handed. Winners carried an exact price, a bundle, or a stated promo window. The failures said "check my cart" and left it there. A creator cannot invent a bundle price. If nobody supplies one, the call to action defaults to generic.
What they opened with. The winning videos put the product in the first frame. The failures opened on the problem, which is standard copywriting advice and costs you category clarity in the first three seconds of a scroll.
All three are decisions made before the creator opens a camera. None of them are creator attributes.
How to Score a Creator Properly
Ranking by total GMV rewards whoever happened to receive the best briefs and the best products. Ranking by views rewards whoever makes the most watchable content, which is a related but different skill. Both produce tier lists that reshuffle the moment conditions change.
Three measurements that survive the variance:
Hit rate, not total. How many of a creator's videos cleared a revenue floor, as a percentage of everything they posted for you. Creator A's seven winners look different once you know how many attempts produced them. Total GMV hides the denominator.
Performance against the same brief. Compare creators who received identical product, identical offer, and identical opening instruction. That isolates delivery skill from briefing quality. Most programs never run this comparison because briefs are inconsistent by default.
Category fit, tracked separately. A creator who converts on pest control and fails on baby care is not inconsistent. They are correctly matched to one category and wrongly matched to another. Scoring them on a blended average hides both facts.
Where the Leverage Sits
Recruitment gets attention because it feels like the decision with the most upside. Sign a big creator, get big numbers. And the ceiling does move with creator quality, which is why the top of this dataset contains people who won across four categories.
The floor moves with something else. The 23 creators who produced both winners and zeros were capable of the winning version every time. They did not deliver it every time, and the difference tracked to what they were handed rather than what they are able to do.
For how those briefing decisions show up in the video itself, see the analysis of 5,871 FMCG affiliate videos. For the commission structures that determine which creators take your brief seriously in the first place, see TikTok Shop affiliate commission structures for 2026.
The Uncomfortable Version
If a quarter of your roster is producing both your best and worst results, then a quarter of your roster is telling you the brief is doing more work than the byline.
That is better news than the alternative. Creator quality is expensive to change and slow to fix. Briefs you can rewrite this afternoon.
