What this is, and how to use it
The Research Desk answers one narrow question well: of all the products I could film next, which ones are worth the camera time? It answers it from the reports you already export, with a formula you can read, and then it measures whether the answer was right.
Reading time: about six minutes. You do not need an account to read this page, the method or the data contract.
The problem it solves
Amazon Influencer commission does not come from the number of videos. It comes from filming products that sell, have few creators already reviewing them, carry a funded campaign, and sit in a category this account already earns in. Left to feel, that judgement is unreliable: the list of products "everyone says are hot" is the same list every creator films.
The desk replaces the feeling with arithmetic. You give it three reports. It scores every product on six signals, shows the working for each row, and ranks them. You choose what to film. Weeks later, it joins the commission back and tells you whether the ranking was better than picking an average product.
What it is not
- Not a marketplace or a course. There is nothing to buy here, no upsell, no affiliate code.
- Not a model making the decision. The ranking is a deterministic formula with a version number. The same data always produces the same order, and every row shows its inputs.
- Not an autoposter. It never messages a brand, never uploads a video, never spends money. It records where your outreach got to.
- Not a promise. A score is a prediction from incomplete data. That is exactly why every row carries a coverage figure and why the desk measures itself later instead of asking you to trust it.
The mental model
One loop, five moves. Everything on the site is one of these:
| Move | What happens | Where |
|---|---|---|
| Import | Three CSVs land in the database and the ranking is recomputed | Data |
| Read | A score, a coverage figure and one headline flag per ASIN | Ranked products |
| Act | Set an outreach state; copy the ASINs; film | Outreach |
| Freeze | Record the exact list you are acting on, with its version and date | Measurement |
| Measure | At 30/60/90 days, real commission is compared with the prediction | Measurement, Brief |
1 · Gather three files
All three are exports you already have, or a spreadsheet you can build. Full column detail and alias names are on the data contract page.
| File, and where it comes from | What it powers |
|---|---|
|
Product sheet — one row per ASIN. Any research tool's export, or a spreadsheet you fill in: ASIN, price, monthly units, creator-video count, category, rating, reviews, parent ASIN. |
Sales, competition, and — with campaigns — commission per sale. This is the file that makes the ranking possible. |
|
Campaign history — Creator Connections. The campaign report downloaded from your influencer dashboard. |
Commission rate, budget health, sample likelihood, brand grouping, and campaigns about to expire. |
|
Earnings report — Associates. Your earnings report, one row per ASIN and period. |
Category adjacency, the "already on my storefront" flag, and the baseline your prediction is measured against. |
Coverage is weighted, so the number of files you bring changes what you see: product sheet alone is 50% coverage and every row is flagged as missing evidence; plus campaigns reaches 85%; all three reaches 100%. The desk says so on the row rather than quietly guessing.
2 · Import
Sign in, open Data, pick the report type for each file and upload. Limits and behaviour:
- 12 MB per file; commas, tabs, semicolons and pipes are all detected.
- Headers are matched by alias —
Product ASIN,units sold,commission rateand similar all work. Anything unrecognised is listed back to you, so a wrong report is caught before it corrupts a ranking. - The identical file uploaded twice is detected by SHA-256 and changes nothing.
- A products import recomputes the ranking immediately. Every import is recorded in the run log.
3 · Read the ranking
Each row has three things: a score (0–100), a coverage figure (how much of the score had real evidence) and a flag that says what to do with it.
| Flag | Plain meaning | What to do |
|---|---|---|
| high_potential | Strong score, real sales, few creators, an open funded campaign | Film, or request a sample today |
| low_hanging_fruit | Sells well, almost nobody has reviewed it | Film even without a sample |
| on_my_storefront | This ASIN already pays you | Consider a second angle, or a retag |
| variation_i_own | A variation of a product you already earn on | Check the variant before filming, then retag |
| data_missing | Less than 60% of the score had evidence | Fix the import, not the product |
| dead_weight | No sales, or a campaign that has spent nothing | Skip |
| watch | Nothing strong either way | Leave it in the reserve list |
Open a row and you can see the arithmetic: what each of the six signals contributed, what its input was, and why. The method page has the weights and the normalisation.
4 · Act on it
The desk does not send anything. It gives you the list and remembers where each product got to: queued → sent → sample → filmed → placed → earning → dead. Brands are grouped, so a sample request can cover several ASINs in one message, and Copy ASINs puts the current view on your clipboard one per line.
5 · Freeze a list
When you decide what to film, freeze the top of the ranking into a cohort. That is a copy with the score version and the date on it, and it never changes. This is the only way the later comparison means anything: a list that keeps moving always "wins".
Freezing is one click on the Measurement page, and you can freeze as many cohorts as you like — each one is measured separately.
6 · Measure
At 30, 60 and 90 days the desk joins the commission that actually arrived back onto the frozen list, and reports:
- Baseline — the median commission per ASIN across the whole account in that window: what an average filmed product earns.
- Precision — the share of your frozen top rows that beat that baseline.
- Lift — the mean commission of the frozen rows divided by the baseline. Below 1× the list was worse than filming at random.
Before any commission exists the window reads collecting, not 0%. A monthly earnings row counts only if its period starts on or after the freeze date — no proration, no invented days.
7 · The Monday brief
Every Monday at 08:00 the desk builds a short brief: how fresh your import is, whether coverage has slipped, which campaigns end within 30 days with budget left, three named actions, and whether the last cohort is proving out. It is saved to data/briefs/, shown on the Brief page, and posted to Telegram once a bot token is configured.
When it looks empty — and why that is honest
| What you see | What it means |
|---|---|
| Watchdog says needs attention, database warning | No products imported yet. Nothing is faked green. |
Every row data_missing | Probably the product sheet alone at 50% coverage. Add campaigns. |
Measurement reads collecting | No commission has landed in that window yet. Wait; do not read it as failure. |
| Brief says the import is stale | Campaign budgets and creator counts move. Re-export weekly. |
| No cohort yet | You have not frozen a list, so there is nothing to measure. |
No data yet?
Then the desk will honestly tell you it has nothing to rank — and there are still three useful things here:
- Read the method. The six signals and their weights are the transferable part; they work whether you use this tool or a spreadsheet.
- Build the product sheet now. Columns: ASIN, title, brand, price, monthly units, creator videos, rating, reviews, category, parent ASIN. Twenty rows is enough to see the ranking behave.
- Look at the worked example on this site. The method page shows the arithmetic; the gallery has a console rendering built from synthetic ASINs — and a measured cohort — so you can see the shape before your own data exists.
Questions
Do I need a paid data vendor? No. The three files are yours. A vendor export only saves you the spreadsheet.
How often should I import? Weekly. Budgets and creator counts change; the brief flags an import older than 14 days.
Can I use it for something other than Amazon? The signals are generic (demand, payout, competition, adjacency, budget, samples) but the flags and the Creator Connections fields are Amazon-specific. It was built for one Amazon Influencer account, on purpose.
Who can see my numbers? Only Pocket ID operators. The method, the data contract and this guide are public; earnings and campaigns are not. See privacy.
What if my export has different headers? Send one and it will be mapped, or rename the columns to the aliases on the data contract page. The import tells you what it could not recognise.
Does it write my outreach? No. You send; it records the state and groups brands so one message can cover several ASINs.
Ready to look at the console?
The data is operator-only: Pocket ID passkey in the operators group, no password on this site.