The Number That Made Me Suspicious
A creator partnerships lead at a beauty brand in Singapore sent me a screenshot of her Q1 ROI report. The number was 6.2x — for every dollar spent on creator commissions and samples, the attributed revenue was $6.20. The brand’s CEO was pleased. The partnerships lead was suspicious.
She was right to be suspicious. The 6.2x was counting shared sales twice. A customer who had bought from three creators’ links and then bought from a fourth creator’s link showed up in all four creators’ columns. As a result, the ROI calculation ran against the union of those attributions rather than the unique share. The real number, after the deduplication, was closer to 2.4x — still good, but not 6.2x.
The mistake is not rare. It is the default in most spreadsheets that calculate creator ROI from a TikTok Shop export, because the export shows each creator’s attributed orders, and the natural instinct is to sum the attributed revenue. The sum double-counts every order that touched more than one creator. The right calculation deduplicates first and sums second.
This article is for the operator who runs a TikTok Shop creator program and reports creator ROI upward, and who is not sure whether the number they are reporting is real or is the result of a double-count. The calculation below is the one that produces a number you can defend.
It is a step-by-step formula, not a calculator. The point is that you can run it in a spreadsheet in twenty minutes and know whether your number is honest.
Where the Double-Count Comes From
The TikTok Shop export — the one that lives in the DAMI store data view, or in TikTok Shop Seller Center, or in whatever the platform gives you — shows orders attributed to each creator. The export is per creator, not per order. That is the design of the export, and the design is fine for understanding which creator drove which order. It is not fine for summing.
When you sum the attributed revenue across creators, an order that is attributed to creator A, creator B, and creator C is counted three times. If the order was for $100, the sum shows $300 in attributed revenue, and the ROI calculation divides the $300 by the cost of the program.
The real number is $100 in revenue, attributed to one creator’s share and two others’ assist. The correct attribution is a model that decides how to split the $100 — by first-touch, by last-touch, by weighted share, or by a model that the platform uses internally. Most operators do not run a model; they just sum, and the sum is wrong.
The cost of being wrong is not just a bad number. The cost is that the operator who reports 6.2x ROI is making decisions on the basis of a 2.4x number. The decisions include which creators get more budget, which products get more creator attention, and which markets are worth expanding. A 2.5x mistake in the reported number produces decisions that are worse than they look.
The Four Steps That Produce a Defensible Number
The four steps run in a spreadsheet and take twenty minutes once the export is in hand. The export is per creator, per order, with the order value, the order date, and the creator’s commission. The DAMI store data view produces this export directly, with each row tagged to the order ID and the creator handle.
Step 1: Build the per-order view from the per-creator view. Every row in the per-creator export has an order ID, a creator ID, and an order value. Pivot the export so that the rows are order IDs and the columns are creators, with a 1 in the cell if the order touched that creator and the order value in a separate column. The pivot produces a table where each order appears once, with a list of the creators who touched it.
Step 2: Assign the attribution model. The model decides how to split the order value among the creators who touched it. The simplest model is last-touch: the entire order value goes to the last creator who touched the order. A more nuanced model is weighted: the first touch gets 40%, the last touch gets 40%, and any middle touches split the remaining 20%. The platform’s internal model is a black box, but the last-touch and weighted models are close to what most platforms do.
Steps 3 and 4: Sum and Divide
Step 3: Calculate the unique attributed revenue. Apply the attribution model first, then sum the order values. The model counts each order once and splits the value according to the rule. The total is the deduplicated attributed revenue, and the total is the number you should be reporting.
Step 4: Divide by the program cost. The cost is the sum of commissions paid, samples shipped, and any direct program expenses (creator events, agency fees, content production). The divided number is the ROI. The divided number is the number you report.
| Step | Input | Output | Common mistake |
|---|---|---|---|
| 1. Pivot per-creator export | Per-creator rows | Per-order rows with creator list | Forgetting to keep the order value |
| 2. Choose attribution model | Per-order creator list | Per-order value split | Defaulting to “all equal” without thinking |
| 3. Sum deduplicated revenue | Per-order split values | Single attributed revenue number | Summing the per-creator export instead |
| 4. Divide by cost | Attributed revenue, program cost | ROI multiple | Using gross commission instead of total cost |
A Walk-Through with Real Numbers
The Singapore lead’s Q1 export had 1,840 orders across 14 creators. The per-creator export summed to $312,000 in attributed revenue. The cost was $50,000 in commissions and samples. The naive ROI was 6.2x.
She pivoted the export to per-order. Of the 1,840 orders, 410 carried more than one creator. The average order value was $78. The sum of order values across unique orders was $143,520 — about 46% of the per-creator sum.
She applied a weighted model: first touch 40%, last touch 40%, middle touches 20%. The deduplicated attributed revenue came out to about $118,000. The cost was $50,000. The real ROI was 2.36x.
The 2.36x is still a good number. It justified the program. The 6.2x would have justified a doubling of the program budget, which would have been a mistake because the second $50,000 would not have produced a second $118,000 — the second batch of creators would have overlapped less with the first, and the deduplicated revenue would have been lower per dollar.
What the Model Choice Does to the Number
The choice of attribution model is not free. Last-touch is the simplest and most aggressive — it gives all the credit to the last creator who touched the order, which usually over-rewards the closer and under-rewards the awareness builder. Weighted is closer to what the platform does internally and is the better default for cross-creator programs.
Time-decay is a third option: the closer the touch is to the order, the more credit it gets. Time-decay is what most analytics platforms use by default, and it is closer to the platform’s internal model than the simple weighted model.
The model you pick is the model you report. If you use last-touch, you should say so. If you use weighted, you should say so. The number without the model is a number that cannot be defended, and the operator who reports a number without the model is the operator who will get caught when the CEO asks “how do you know.”
The right model for most TikTok Shop creator programs is weighted, with first and last at 40% each and middle touches at 20%. The model is close enough to the platform’s internal model that the difference is in the noise, and the model is simple enough that the operator can explain it on one page.
Why the Cost Side Is Usually Underestimated
The cost side of the ROI calculation is where the second common mistake lives. The naive calculation uses commission only — the percentage paid to creators on attributed orders. The right calculation uses commission plus samples plus direct expenses.
The sample cost is the largest of the three for a beauty, supplement, or apparel brand. A $15 sample shipped to 200 creators is $3,000 in cost, which is 6% of the $50,000 example above. Across a year, the sample cost is the difference between a 2.36x ROI and a 2.2x ROI. In fact, that difference matters most when the operator defends the program in a budget review.
The direct expense cost is the smallest but the most volatile. Creator events, agency fees, content production, and any contracted services belong here. The cost is the line that surprises the operator who has been tracking commission only.
The cost side is harder to get right than the revenue side because the cost side lives in operational data, not in a clean export. The fix is to maintain a per-month program cost log that captures commission, samples, and direct expenses, and to refresh the log every month rather than at the end of the quarter. The log is the difference between a defensible ROI and a fuzzy one.
When the Number Is Real but the Conclusion Is Wrong
A defensible ROI is not the same as a defensible conclusion. A 2.36x ROI is real, but the conclusion “we should double the program” is not automatic. The conclusion needs three additional pieces of information: the saturation point, the marginal ROI, and the program cost ceiling.
Saturation point. The ROI is a function of how many creators are in the program. A 14-creator program might be at 2.36x, but a 28-creator program might be at 1.8x because the additional 14 creators overlap with the first 14 in their audiences. The saturation point is the creator count at which the marginal ROI drops below the cost of capital.
Marginal ROI. The ROI on the most recent dollar spent is the number that matters for the next dollar spent. A 2.36x average ROI might be hiding a marginal ROI of 0.8x on the last $10,000. The marginal ROI is the number that tells you whether the next creator is worth adding.
Program cost ceiling. The program is bounded by the budget. A 2.36x ROI on $50,000 is $118,000 in revenue. A 2.36x ROI on $200,000 is $472,000 in revenue, but the marginal ROI on the second $50,000 is probably 1.4x and the marginal ROI on the fourth $50,000 is probably 0.6x. The program cost ceiling is the budget at which the marginal ROI crosses the cost of capital.
These three numbers are not in the per-creator export. They are calculated by running the program at different sizes and observing the curve. The first two months of a new program are the calibration period; the third month onward is the operational period.
Frequently Asked Questions
Is the platform’s own ROI number reliable? Usually no. The platform’s ROI is calculated with the platform’s attribution model, and the platform’s model usually over-counts. Run your own calculation, deduplicate, and use a model you can explain.
What’s the simplest model that does not double-count? Weighted: first touch 40%, last touch 40%, middle touches 20%. It is close to the platform’s internal model, it is simple to explain, and it does not require tooling beyond a spreadsheet.
What if I have no per-order data? You cannot run a real attribution without per-order data. Ask TikTok Shop Seller Center for the per-order export, or use the DAMI store data view to get a per-order view of the attributed orders.
What cost should I include in the denominator? Commission, samples shipped, and any direct program expenses. Exclude internal headcount, agency retainers that are not specific to this program, and platform fees that exist regardless of the program.
How often should the calculation run? Monthly. A weekly calculation is too volatile, a quarterly calculation is too late. The monthly cadence lets the operator see the trend without being misled by a single week’s noise.
Run the Number Before the CEO Does
A defensible creator ROI is a number you can run in twenty minutes and explain in two pages. The naive sum is a number you can run in three minutes and cannot explain at all. The difference is whether the program is funded next quarter on the basis of a real number or a double-counted one. The DAMI store data view is where the per-order export comes from, and the DAMI per-creator view is the row-level source. For the KPI definitions this calculation depends on, the data dashboard guide sets out which metrics to fix before you calculate anything. See how the DAMI workspace surfaces the per-order attribution on the product page, or create your account to run the four steps on your next export.
