Quick calculator
Scenario gap = actual Amazon ordered product sales − your supplied expected sales without TikTok. Lift = gap ÷ expected sales. The baseline drives this answer; the calculator cannot verify it from two sales totals.
If you enter a baseline range, the gap range is actual − high baseline through actual − low baseline. It is a user-defined scenario range, not an 80% or 95% confidence interval. Previous-period sales alone do not establish a counterfactual.
Daily data model
A signed linear model uses trend, weekday effects and one delayed TikTok view series. It evaluates 12 prespecified kernels with lags 0–28 days, decay 0.5/0.7/0.85 and peak 0/3/7/14. At 730 days of input history, annual Fourier terms are included. Optional control_ columns are standardized and held fixed under the comparison.
The point estimate uses the lowest-AIC candidate. The counterfactual removes all measured exposure, including warmup. It is a whole-program comparison, not a campaign effect. Fitted sales and actual sales differ by the model residual; the contribution is fitted actual-condition sales minus fitted zero-exposure sales.
Each of the 12 candidates receives 1,000 HC2-adjusted block-wild bootstrap replicates, with 14-day calendar blocks and seed 41017. Aggregate 80% and 95% ranges are the union of candidate percentile ranges. They describe uncertainty conditional on these modeling assumptions; they do not cover all omitted causes or establish real-shop causal validity.
Requirements and limits
At least 140 complete outcome days plus 28 exposure warmup days; continuous daily inputs; exposure CV ≥0.3; selected exposure VIF ≤5; at least five observations per parameter. This simple runner requires 100% complete outcomes and limits input to 1,096 outcome days. A block choice of 7 or 28 days can be run as sensitivity after the main result.
Price, stock, other marketing and seasonal changes can explain the same sales movements. Causal support, placebo/negative-control tests, broader simulation validation and independent review are not completed by this tool. Daily model results are internal draft estimates, never automatically approved customer claims.
What to upload
Amazon: date and ordered product sales in one marketplace and currency. TikTok: date and verified daily total views, starting 28 days earlier. Use the same approved products and source-day boundaries. If data is missing, leave it blank; it won’t become zero.
No customer names, order addresses, payment data or private creator contact information are needed. Every export retains inputs, scope, method and result status. Program-return scenarios require matched costs; they are not media iROAS or profit.