Nobody audits their marketplace deductions because “audit” sounds like a finance project — a weekend, a pivot table, a headache. So it never happens, and the double-charged return fee from February quietly becomes permanent. Here is the reframe: a deduction audit at 5–25 orders/day is a 30-minute monthly ritual— four documents, five comparisons, one sorting decision. The time-box is not a compromise. It is the design. An audit that fits inside a chai break is an audit that happens twelve times a year.
This is the exact checklist we run with sellers on AJIO and Meesho. It assumes nothing fancier than a download folder and a findings sheet. Every share and rupee figure below is illustrative — your category and volume move the numbers, not the method.
Minute 0–5: pull the four documents
The audit is only as good as its inputs, and there are exactly four: every settlement statement the marketplace issued in the month; the bank statement for the same period; your order export (every order with status and dispatch timestamp); and your returns list (every return and RTO with its order ID). Pull them into one folder, dated. Resist the urge to start reading — the pull phase is mechanical on purpose. If any document takes more than a minute to locate, that is your first finding: fix the filing habit before next month.
Minute 5–15: run the five comparisons
The matching phase asks five questions of those four documents:
- Return charges vs returns list. Group every return-related deduction by order ID. One return, one charge — any order ID appearing twice is a finding. This single check finds money more often than the other four combined.
- Penalties vs your timestamps. Every SLA or compliance penalty asserts an event. Check the order's dispatch and manifest timestamps. No matching late event means a finding.
- Approved claims vs credited reversals. List every claim or dispute the marketplace approved in past months. Confirm each reversal credit actually appears in a settlement. Approved-but-never-credited is a finding.
- Applied rates vs your rate card. Recompute commission and fees on five random orders. A rate that does not match your agreement — common after category re-mappings — is a finding.
- Delivered orders vs settlement coverage. Walk your order export and confirm every delivered order appears in some settlement. An order that never settled anywhere is the most expensive finding on this list, and the only one with no deduction line to notice.
What you will probably find
Across the sellers we've sat with, the findings repeat with boring consistency. Double-charged return and RTO fees top the list, followed by penalties with no matching event, then reversal credits that never arrived after an approved claim. Behind those: rate drift against the signed rate card, delivered orders absent from every settlement, and tax lines that never reached the government portals. The encouraging part: the top three are documentary disputes — the evidence is two statement lines, a timestamp, or an approval message, all already in your folder. These have strong win rates when filed inside the window, which is precisely why the audit must be monthly and not quarterly.
Two patterns deserve a special mention because they compound. Rate drift is small per order but applies to every order after the change — a one-percentage-point shift on 400 orders a month is real money on autopilot. And the missing-settlement case is the opposite shape: rare, but each instance is the full value of an order, not a fee on it. Between them they justify the two slowest checks on the list. The fast checks find the frequent errors; the slow checks find the expensive ones.
Minute 15–25: classify with the dispute-or-accept matrix
Every finding gets sorted on two axes: evidence strength and rupee value. Strong evidence plus high value: dispute now, individually, full evidence attached. Strong evidence plus low value: batch them and file weekly — a ₹23 double-charge is not worth its own session, but eleven of them are. Weak evidence plus high value: investigate before you argue; the answer is usually an upstream process fix, not a claim. Weak evidence plus low value: accept, and log it anyway. The logging matters more than it looks — three months of logged accept-losses is how you discover that one courier lane or one SKU is generating all of them.
Minute 25–30: file and log
Disputes go in immediately — windows on AJIO and Meesho events are measured in days, and a finding aged past its window is just trivia. Each filing gets the minimal documentary package: order ID, the statement line(s), the contradicting record. Then everything — disputed, accepted, fix-upstream — goes into the findings sheet with the date and amount. Twelve months of that sheet is the most honest report your business has: it tells you your real deduction error rate, your recovery rate, and whether the leak is shrinking.
A note on what a good dispute filing looks like, because quality beats volume here. One finding per ticket — bundled grievances get bundled rejections. State the claim in one sentence (“return fee charged twice for order X, see lines A and B”), attach the two or three documents that prove it, and say what you expect (“reversal of ₹118”). Skip the editorial about how often this happens — the findings sheet holds that for the escalation conversation, not the first ticket. And track every filing to its bank-line conclusion: an approved dispute whose credit never lands is just finding number three on next month's audit.
The findings sheet: six columns, no more
The audit's memory lives in one sheet, and the sheet earns its keep by staying small. Six columns: date found, order ID, amount, finding type (one of the six categories above), action (disputed / accepted / fix-upstream), and outcome (recovered, rejected, pending — filled in later). Resist every urge to add columns. The sheet is not a finance system; it is a pattern detector, and patterns show up in small tables. After three audits you can answer questions that almost no seller at this scale can: what is my real monthly deduction error rate? Which finding type repeats? Is my recovery rate improving? When a marketplace support conversation goes in circles, the sheet is also your receipts — “this is the fourth double-charged return fee in three months, here are the order IDs” lands very differently from a vague complaint.
Two tips for the first run. Expect it to take an hour, not thirty minutes — you are building the muscle and probably discovering your documents are scattered. And do not try to audit six months retroactively: most of what you would find is past its dispute window, which makes for an afternoon of well-documented frustration. Start with last month, file what is still in window, and let the ritual compound forward.
Scaling the ritual as volume grows
The 30-minute box holds comfortably to roughly 10–15 orders/day. Past that, the matching phase — the ten minutes of comparisons — grows linearly with order count while your patience does not. The progression we see: sellers first move the audit weekly (smaller batches, fresher memory, more windows still open), then they stop doing the matching by hand entirely and keep only the classification and filing as human work. That split is the right one. Matching is mechanical — tie line to order, flag what has no event — and machines are better at it. Classification is judgment — is this evidence strong, is this deduction actually correct — and judgment is the part worth your minutes. Whatever tooling you use, protect the monthly review of the findings sheet itself: automation finds individual errors, but only a human reading three months of patterns decides to change a courier, fix a listing, or renegotiate a rate.
Where Robnu fits
Everything above is doable by hand, and at low volume you should do it by hand at least twice — there is no better way to learn your own deduction anatomy. Past that, the matching phase is exactly what software is for. Robnu is an agentic OMS with AJIO and Meesho live: it ties every settlement line to its order continuously, flags deductions with no matching event, and routes dispute candidates into claims filed for you — with the audit trail the dispute needs already attached. The monthly ritual becomes reviewing what the system flagged, which is the five-minute version. The full method behind the matching is covered in our payment reconciliation guide.
Robnu is free for everyone right now — every feature, every order, no card, no trial timer — while we figure out paid pricing. When it launches, sellers under 25 orders/day stay free forever, and today's users get grandfathered locked rates. This Sunday, pull last month's four documents and run the five comparisons once. The first audit almost always pays for the next twelve.