Platform Fee Recovery Playbook for D2C: Razorpay, PayU, Marketplace Settlement Audit

A D2C finance lead at a brand doing ₹84 crore of annualised GMV sits with six months of Razorpay, PayU, Cashfree, and Stripe settlement files spread across four tabs of a working file. The aggregator-level summary at the bottom of each file shows neat percentages — 1.78% effective MDR on Razorpay, 1.82% on PayU, 1.74% on Cashfree, 2.94% on Stripe (cross-border). The board chair has asked whether the brand is leaking money on platform fees. The honest answer is that the finance lead does not know — because no one has reconciled the aggregator summary to the contracted slab at the instrument and transaction level. The drift, if it exists, is invisible at the summary level.

This is the playbook for finding and recovering it. Five audit tracks, a worked example, and the aggregator escalation matrix that converts variance into rupees back in the bank account. It sits alongside the broader Revenue Leakage Recovery Playbook — platform fees are class one (fee deduction) of the Seven Classes framework and the highest-recovery-probability class for any D2C operator.

Quick reference: the five leakage tracks

The bands are indicative for a brand with ₹50-150 Cr GMV, three to five aggregators, and 8-15% international sales. They tighten with per-brand calibration during baseline.

Track one — MDR audit per instrument and per aggregator

The contract with Razorpay, PayU, Cashfree, or any other aggregator specifies MDR by instrument and network. A standard schedule lists credit card domestic at one rate, credit card international at another, debit card by network (Visa, Mastercard, RuPay), UPI (often zero or near-zero with promotional overrides), netbanking, wallet (per partner), EMI (by tenor and bank), BNPL, and special schemes (no-cost EMI, network reward redemption). The audit recomputes the expected fee per transaction at the contracted rate and compares to the actual deduction line on the settlement file.

The mechanics. Pull the transaction-level settlement file for six months. For each transaction, identify the instrument, network or partner, transaction amount, and whether it falls under any promotional scheme. Look up the contracted MDR for that exact combination. Compute expected fee plus GST on fee. Compare to actual MDR plus GST deducted on the settlement file. The per-transaction delta aggregated by instrument, network, and month is the MDR drift line. Aggregate by aggregator for the renegotiation pack.

Common drift patterns. UPI promotional overrides not applied — many aggregators offer 0% MDR on UPI up to a monthly volume threshold, after which a standard rate kicks in; the application of this tier is frequently delayed by one to two billing cycles. RuPay debit special rates — RuPay debit is regulated, and aggregators sometimes apply Visa or Mastercard slabs in error. EMI tenor misapplication — 3-month EMI carries a different MDR from 9-month or 12-month; ops teams sometimes apply a blended rate. International card domestic-rate application — a transaction on an international card issued to an Indian customer is sometimes billed at international MDR despite the card being used domestically; this depends on BIN routing.

End-state. A per-aggregator monthly MDR scorecard with contracted MDR, effective MDR, drift in basis points, drift in rupees. The drift line by month becomes the dispute basis. A typical brand surfaces 8-16 basis points of MDR drift in the first audit cycle.

Track two — Settlement-vs-orders mismatch

Daily reconciliation of order count, gross order value, and expected net credit against actual settlement batches. The cadence is daily for the largest two aggregators (Razorpay, PayU, Cashfree in most D2C stacks) and weekly for tail aggregators and marketplaces.

The matching logic. Per day, total orders captured through the aggregator times average ticket size gives gross expected. Gross expected minus MDR minus GST on MDR minus aggregator-side adjustments (cashback, network discount) gives net expected. Compare net expected against the sum of settlement batches credited two banking days later (T+1 standard, T+2 for some categories, T+3 for international). Variance per day goes into a residual queue.

Common patterns. Missing settlement batches — a small percentage of batches are dropped at the bank-side bulk credit and need recharge. Timing mismatches — T+2 versus T+3 routing varies by instrument and risk category; cross-border transactions on Stripe and Cashfree International route T+5 to T+7 with intermediate FX leg. Held amounts — a percentage of settlement may be held against chargeback exposure; the held amount needs to be tracked as a receivable and aged.

Marketplaces are a separate cadence. Amazon Easy Ship, Flipkart F-Assured, and Meesho settle on T+7 to T+14 with deferred reserves of 3-7% held against returns. The reconciliation pack matches order ID to remittance ID to net credit with a separate reserve-release register. Reserve releases age out at 60-90 days; brands routinely lose 0.3-0.6% of marketplace GMV to unaged reserves.

End-state. A daily variance queue routed to AR ops with 48-hour resolution SLA on items above a defined rupee threshold (typically ₹10,000-25,000 per batch).

Track three — Refund leakage

Three sub-tracks: refund initiated but not credited; original transaction fee not refunded on refund; partial refund handled at full fee.

Refund initiated but not credited. The aggregator API returns success on the refund instruction, but the customer credit takes 5-14 days depending on issuer and instrument. The corresponding settlement debit is staggered. Without per-refund tracking, these go missing — finance closes the refund in books on instruction, the customer credit lands later, and the settlement debit lands later still. A per-refund register with refund ID, instruction date, expected customer credit date, expected settlement debit date, and three-way match field closes this.

Original transaction fee on refund. Some aggregator contracts refund the MDR on the original transaction when the transaction is refunded within a stated window (often 30-60 days). Others retain MDR regardless. Many contracts have a middle position — refund MDR on customer-initiated refund, retain on merchant-initiated. The audit checks every refund against the contractual treatment and identifies cases where MDR should have been refunded but was not.

Partial refunds. A partial refund of, say, 40% of the original transaction should attract 40% of the original MDR refund (where MDR is refundable). In practice, aggregator systems sometimes apply full or zero MDR refund on partial events. Per-refund check against the contractual proration logic closes this.

End-state. A monthly refund leakage decomposition by aggregator and sub-track, feeding a dispute pack at month-end.

Track four — Chargeback recovery and representment

The unmanaged chargeback win-rate for D2C brands sits at 20-30% — many disputes are accepted without representment because the operational discipline does not exist. A structured program moves this to 55-70% within two quarters.

The five components. Twenty-four-hour intake — the aggregator notifies the chargeback via dashboard or webhook; the brand needs an intake within 24 hours because most schemes give a 7-to-10-day representment window. Per-reason-code evidence library — Visa Compelling Evidence 3.0 (CE 3.0), Mastercard Consumer Clarity Dispute Resolution Network (CDRN), RuPay reason codes, and American Express reason codes each require different evidence stacks (delivery confirmation, IP and device fingerprint, refund history, AVS match, prior transaction history). Automated evidence assembly — pulling from order management, warehouse management, shipment tracking, and customer support tools into a template per reason code. Representment workflow with deadline tracking. Win-rate dashboard by reason code, aggregator, and product category, used to refine evidence templates and identify systematic abuse patterns.

Common patterns. Friendly fraud on high-AOV categories (cosmetics, apparel, electronics accessories) where the customer disputes after receipt — beaten with delivery confirmation, IP and device match. Subscription disputes — beaten with subscription consent record and prior usage. Familial fraud (child uses parent’s card) — beaten with usage pattern. Genuine fraud — accepted, but pattern logged for issuer escalation.

End-state. A weekly chargeback queue with representment status per dispute, weekly win-rate review, quarterly contract review with the disputes lead.

Track five — FX drift on international card transactions

For brands with international sales, cross-border PG fees apply (typically 2.5-3.5% MDR versus 1.6-2.0% domestic) plus an FX markup applied to the conversion. The audit recomputes expected INR settlement using the published reference rate for the transaction date and the contracted FX markup.

The mechanics. Pull every international transaction over six months. For each, identify the transaction currency, transaction amount in foreign currency, and the conversion timestamp. Look up the published reference rate for the date (RBI reference rate or the contracted forex source). Apply the contracted FX markup (often 2-4% above mid-market). Compute expected INR. Compare to actual INR credit. Variance per transaction aggregated across the period is the FX drift.

Common patterns. Stale reference rates — aggregators sometimes use a rate from one or two banking days earlier rather than the transaction date. Undisclosed markup — markup beyond the contracted band. Conversion timing — transactions captured at one rate and settled at another with the spread retained by the aggregator.

End-state. A monthly FX drift line per cross-border aggregator, feeding the contract renegotiation pack.

PA-PG aggregator escalation matrix

Every aggregator contract has a dispute and escalation clause. The escalation matrix names the operational, account-management, and senior counterparts and the time bound for each level.

Level one — operational. Daily variance queue items, settlement-batch follow-ups, refund tracking. The aggregator-side counterpart is the ops desk; turnaround 24-48 hours.

Level two — account management. MDR drift disputes, refund-fee retention disputes, FX drift disputes. The aggregator-side counterpart is the named key account manager; turnaround 5-10 business days.

Level three — senior escalation. Systematic drift, contract interpretation disputes, repeated SLA breach. The aggregator-side counterpart is the head of merchant services or VP merchant relations; turnaround 10-15 business days.

Level four — regulatory and contractual. Repeated material drift, non-recovery of disputed amounts, contract breach. The route is a formal notice under the contract dispute clause; the regulator-aligned reference is the Reserve Bank of India PA-PG framework for settlement timelines and transparency.

The matrix lives in the playbook binder. Each level has a templated dispute letter, the evidence pack expected, and the response SLA. Brands that operationalise levels one and two recover 60-75% of identified drift; brands that operationalise three and four push to 80-90%.

Worked example — a ₹84 Cr GMV D2C brand, four aggregators

The brand routes 62% of revenue through Razorpay, 18% through PayU, 10% through Cashfree, and 10% through Stripe (international). The annual baseline audit surfaces ₹38 lakh of identified leakage across the five tracks.

Decomposition. ₹14 lakh of MDR drift, concentrated on UPI promotional overrides not applied on Razorpay (volume tier moved from 0% to 0.4% three months late) and on RuPay debit applied at Visa/Mastercard slab on PayU. ₹9 lakh of refund fee retention on partial refunds across Razorpay and Cashfree where contract specified prorated refund but full-original-MDR was retained. ₹8 lakh of chargeback non-representment — 320 of 540 annual disputes were accepted without representment with average loss per dispute of ₹2,500. ₹4 lakh of FX markup drift on Stripe — markup applied at 3.4% against contracted 2.8%. ₹3 lakh of missing settlement batches across all four aggregators, identified through daily reconciliation.

Recovery program. Track one MDR — operational dispute filed on Razorpay UPI volume tier (₹6 lakh accepted within 90 days); account-management escalation on PayU RuPay slab (₹4 lakh accepted with prospective correction). Track three refunds — operational dispute on Razorpay partial refunds (₹4 lakh accepted); contract clarification on Cashfree (₹2 lakh accepted, prospective). Track four chargebacks — 24-hour intake and per-reason-code evidence library deployed Q2; win-rate moved from 22% to 58% over two quarters; ₹5 lakh recovered on Q3 and Q4 dispute volume. Track five FX — markup dispute on Stripe (₹2 lakh accepted, prospective correction to 2.8%). Track two settlements — ₹3 lakh recovered through batch recharge.

Total recovered. ₹26 lakh of the ₹38 lakh identified, recovery rate 68%, deployment effort 240 finance-team hours plus aggregator account-management cycles. The contract renegotiation pack at next aggregator renewal cycle is expected to compound the run-rate saving by tightening MDR slabs and FX markup language.

Common pitfalls

One — auditing at the aggregator-summary level instead of the transaction level. The effective MDR percentage at the bottom of the settlement file hides per-instrument drift; only transaction-level reconciliation surfaces it.

Two — ignoring promotional MDR overrides. UPI, RuPay debit, and no-cost EMI carry promotional rates that change quarterly; without tracking the override schedule, the application of the override is invisible.

Three — treating marketplaces as aggregators. Amazon, Flipkart, Meesho, and Ajio operate on different settlement timelines (T+7 to T+14), different reserve structures (3-7% held against returns), and different reconciliation primitives (remittance ID rather than settlement batch). A unified aggregator template under-recovers on marketplace leakage.

Four — under-investing in chargeback representment. Brands that treat chargebacks as a cost of doing business and skip representment leave the highest-recovery-probability track unrun. A 24-hour intake and per-reason-code evidence library is a four-to-six-week build that compounds for years.

Five — running the audit once a year instead of monthly. The compound drift over twelve months is 8-12 times the compound drift over one month; the dispute-window discipline (30-60 days per most contracts) means once-a-year audits time out a material portion of disputable leakage.

Six — not feeding the audit into the contract renegotiation cycle. The MDR scorecard, FX drift line, and refund-treatment evidence are the negotiating asset at renewal. Brands that recover in dispute but do not tighten the contract pay the same drift again next year.

Closing — platform fees as class one of the Seven Classes

Platform fee leakage is class one (fee deduction) of the Seven Classes framework and the highest-recovery-probability class for any D2C operator routing meaningful GMV through aggregators and marketplaces. The five-track audit, the PA-PG escalation matrix, and the monthly Discovered Money register convert what is otherwise an opaque deduction line into a board-track recovery program. It sits alongside class two (tax deduction), class three (discount), and class five (short settlement) as the four leakage classes a D2C finance team should prioritise in the first year of the recovery playbook deployment. The board case article covers how to package the audit findings for the audit committee and unlock the operating budget that sustains the program quarter over quarter.