FullRecoup

1. One-liner

Catches the units Mercado Livre lost, damaged or short-received, and files before your 20- and 60-day windows shut.

2. Trend signal — why now?

Three things landed on Mercado Livre Full sellers in 2026, and they compound.

The platform started charging per unit for inventory discrepancies. New Full rules impose fines of up to R$9 per unit where declared and actual stock diverge, and R$5 per unit for missed receiving appointments. Divergence used to be an annoyance. In 2026 it is a line item with a price on it.

The cost of holding anything in Full went up across the board. Daily storage fees rose 7.6% for medium and large items, withdrawal fees rose 5%, collection rose 5–10%, and from 2 March 2026 a variable pricing model replaced fixed fees on items under R$79. Items unsold at 90 days trigger a 30-day sell-or-withdraw window; items paused six months pick up a 6.4% surcharge. Every one of those meters runs on inventory that may not physically be there.

Meanwhile the recovery side is entirely manual and deadline-gated. Sellers have up to 20 days from goods being received at the fulfilment centre to claim divergences or missing units that never entered sellable inventory — and if items are not registered to the account within 3 days of delivery confirmation, the seller is instructed to contact Mercado Livre immediately. Separately, sellers get up to 60 calendar days after an inspection result to contest a return decision or claim reimbursement. Miss it and the capital loss is permanent.

The asymmetry is the whole opportunity. The fines are automatic and per-unit. The reimbursements are manual, seller-initiated, and expire. Mercado Livre is not going to email you that it lost eleven units and you have nine days left to say so.

And the commission hike raised the stakes on every unit: Mercado Libre Brazil moved the standard commission to 17% in Q3 2026, on top of ~2.99% payment processing. A lost unit does not just cost COGS — the platform fee was already taken.

Provenance:

• Signal 1 (demand): Full sellers have up to 20 days to claim receiving divergences/missing units and up to 60 calendar days after inspection to contest returns or claim reimbursement; missing the window means permanent capital loss — https://sellsync.ai/pt/blog/erros-mercado-livre-full-prejudicam-lucro/ and https://www.mercadolivre.com.br/ajuda/Termos-e-condi%C3%A7%C3%B5es-do-MercadoEnvios-Full_2982 — observed 2026-09-04
• Signal 2 (feasibility): Mercado Libre’s developer platform exposes REST APIs for orders, shipments, claims and returns, including a /claims resource to “get claim details and start actions via API” — https://global-selling.mercadolibre.com/devsite/manage-claims and https://developers.mercadolivre.com.br/en_us/ruby/working-with-claims — observed 2026-09-04
• Signal 3 (economic): 2026 Full rules add fines up to R$9/unit for inventory divergence and R$5/unit for receiving non-conformity, storage +7.6%, withdrawal +5%; Brazil commission moved to 17% in Q3 2026 — https://base.com/pt-BR/blog/novas-regras-full-mercado-livre-2026/ and https://www.rewarx.com/blogs/mercado-libre-brazil-2026-q3-commission-hike-to-17-seller-cost-pressure — observed 2026-09-04 Category: Platform shift

3. The opportunity

There is a mature category for exactly this on Amazon — FBA reimbursement recovery, dozens of vendors, 15–25% contingency fees, a well-understood business. On Mercado Livre, the equivalent does not exist, despite Full being the dominant fulfilment rail in the largest e-commerce market in LATAM.

What does exist is a shelf of analytics tools. ProfitOS calculates true net profit per sale by breaking down commissions, shipping, VAT and COGS via the official API. Magiic does BI, Full stock coverage, replenishment suggestions and Ads ROAS. Jaguar Sheet syncs 44 columns of order data — including billing, commissions, withholdings and claims — into Google Sheets.

Every one of these tells you what happened. Not one of them tells you what you are still owed and how many days you have left to ask. Magiic’s own site makes no mention of detecting lost inventory or filing reimbursement claims; it is optimisation and analytics, not dispute resolution.

That is the gap, and it is a structural one rather than an oversight. A profitability dashboard is a read-only product — it ingests reports and renders them. A recovery product has to do three things a dashboard never does: reconstruct what should have happened from independent event streams, decide that a discrepancy is claimable, and act inside a deadline. The incumbents sell insight to a seller who then does nothing with it. We sell the money back.

The second-order point: the 2026 rule changes made the platform’s own accounting more consequential, not less. When storage is metered daily and divergence is fined per unit, a phantom unit in your Full inventory is charging you rent and exposing you to a fine simultaneously. Sellers are now paying to store inventory Mercado Livre has already lost.

4. Target market

Primary customer: Brazilian Mercado Livre sellers running Mercado Envios Full with roughly R$150K–R$3M in monthly GMV — the band with enough unit volume that discrepancies are statistically certain, but no dedicated operations analyst to chase them. Typically 3–25 employees, an owner-operator or an analista de e-commerce who also handles ads, listings and customer service. Secondary: Mexican and Argentine Full sellers in the same band.

Why they buy: Because the loss is invisible until someone reconciles, and by then the window has closed. In their framing, the money was never theirs to miss — it silently never arrived. The published guidance to sellers is blunt about the consequence: miss the 60-day contest window and it is permanent capital loss on product sitting unsellable in limbo. The R$100-cost / R$200-sale example circulating in Brazilian seller content works out to roughly R$227 of total loss once commission, freight, packaging, opportunity cost, dispute labour and reputation damage are added — the platform fee having already been charged before anything went wrong.

Rough TAM reasoning: Mercado Libre reports millions of active sellers across LATAM; the Full-using, mid-volume Brazilian subset is plausibly in the low tens of thousands, with Mexico adding a meaningful second tranche. I do not have a defensible published count of Full-enrolled sellers in this GMV band, and I am not going to invent one — this is the single biggest number I would want nailed down in validation. At 1,500 paying accounts the business clears $1M ARR, which is a small fraction of any reasonable estimate of that population.

Why now for them: Three cost increases and a new per-unit fine regime landed in 2026, and the 17% commission means every lost unit is a larger absolute loss than it was last year. Margin pressure is the thing that makes a seller open a spreadsheet and discover the leak.

5. Product sketch (MVP)

• Inbound reconciliation — for every shipment sent to a Full centre, compare units declared against units actually registered to the account, and flag shortfalls within the 3-day registration expectation rather than after the 20-day claim window has run.
• Expiry ledger — a running list of every claimable event with its own countdown: receiving divergences on the 20-day clock, return and damage contests on the 60-day clock. Sorted by days remaining and rands at risk, not by date discovered.
• Claim dossier generation — for each flagged event, assemble the shipment reference, declared-vs-received counts, inspection result, timestamps and product value into the format the claim actually requires, in Portuguese or Spanish.
• File and track — submit through the claims API where the resource supports it, fall back to a guided one-click flow where it does not, then track the claim to resolution or denial.
• Denial rebuttal — when a claim comes back short or refused, produce the counter-argument with the evidence attached rather than making the seller start over.
• Phantom-inventory alarm — surface units you are being charged daily storage on that have no plausible physical existence given the receiving record.
• Recovery statement — a monthly plain-language statement of what was found, filed, recovered and lost to expiry, which doubles as the renewal justification.

6. AI angle — what’s load-bearing

Two places, both genuinely load-bearing.

The first is adjudicating the discrepancy. A shortfall between declared and received units is not automatically a claim. It could be a mis-scan, a split shipment still in transit, a repack, a units-per-box declaration error by the seller, or an actual loss. Deciding which — from partial, inconsistent event records across shipments, orders, returns and billing, in Portuguese, with the platform’s own vocabulary — is exactly the judgment call that ate the seller’s afternoon and is why the work never gets done. Rules-only logic produces a flood of false positives, which is worse than nothing because it burns the seller’s trust on the first week.

The second is building the argument. A claim that gets paid reads differently from one that gets closed. The system has to write the case in the platform’s own terms, cite the right references, and — when the response is a denial or a lowball — construct the rebuttal against the specific reason given. Brazilian seller reporting on the LATAM side of this is explicit that support will push back, question the seller’s valuation, and suggest lower figures. Negotiating text at volume is language work.

Strip the AI out and you have a spreadsheet diff that generates noise and hands the seller a to-do list they already had. The product’s entire promise is that a human does not have to read the records.

7. Localization angle

This is a LATAM-native play and could not be a global product wearing a translation.

The rules are Brazil-specific and change on Brazilian timelines — the March 2026 variable pricing shift, the R$3–R$9 divergence fines, the 120-day and 60-day long-term storage thresholds. The claim windows are set by Mercado Livre’s own T&C, not by any statute a global tool would track. The seller-facing UI, the claim vocabulary, and the support interaction are all Portuguese, with Spanish for Mexico and Argentina.

Pricing has to be local too. A recovery tool priced at $99/mo reads as expensive to a seller doing R$200K/month with thin margins; the same seller will happily give up a share of money they had already written off. Contingency pricing is culturally the easier sell here and removes the “will this pay for itself” objection entirely.

There is also a real localization moat in the boring direction: the fee schedule, the fine table and the claim windows differ by country and get revised. Keeping that current for BR, MX and AR is unglamorous maintenance work that a US-based Amazon-recovery incumbent will not prioritise.

8. Business model — path to $1M–$5M ARR

Pricing: Hybrid, and the hybrid is deliberate. A R$297/mo (~$55) floor for the monitoring, ledger and alarms, plus 20% contingency on recovered amounts. The floor keeps the business from being purely at the mercy of how much the platform loses in a given month; the contingency is what closes the sale, because the seller is risking almost nothing.

Contingency at 20% sits inside the 15–25% band the equivalent Amazon FBA recovery category has already established, so it is a benchmarked number rather than a guess.

ACV: Target ~$1,100/yr blended — roughly $660 subscription plus ~$450 average annual contingency. That contingency figure assumes a mid-band seller recovering a few thousand reais a year, which is the assumption I would most want to test before believing this model.

Rough math to $1M ARR: ~900 sellers at $1,100 ACV. Alternatively 1,500 sellers if contingency underperforms and it is closer to a pure $55/mo subscription.

Rough math to $5M ARR: ~4,500 accounts, which realistically means Mexico and Argentina carrying a third of the base, plus an upmarket tier for sellers and agencies managing multiple storefronts. Agencies are the accelerant — one agency relationship can bring 20–40 seller accounts under a single contract.

Expansion path: Start with Full inventory claims. Extend into shipping-fee and commission overcharge auditing on the same event data — the reconciliation engine is already reading billing records, and sellers are already reconciling fees manually. Then multi-account and agency tiers. Then Mexico and Argentina on the same rails.

9. Go-to-market wedge — first 100 customers

• The free audit, which is the whole wedge. Read-only API connection, 90-day retrospective scan, output a single number: “R$14,300 recoverable, R$3,100 of it expires in 11 days.” This is the strongest possible cold opener because it is specific, verifiable, and has a countdown on it. It also self-qualifies — sellers with nothing recoverable are not customers and disqualify themselves for free.
• Brazilian seller YouTube and the consultancy layer. There is an established Portuguese-language ecosystem of Full-focused operators publishing rules explainers — Base, SellSync, Hunter HUB, ConectaADS and similar — precisely because the rules keep changing. These are content businesses that need material and have the exact audience. Offer them the audit as a segment for their channel and a revenue share; a single well-placed video in this niche reaches thousands of qualified Full sellers.
• Agencies and accountants first, sellers second. Mercado Livre agencies manage portfolios of sellers and are measured on client margin. Sign 10 agencies and the audit runs across their whole book in a week. The contingency model makes this trivially easy to pitch — the agency looks good for free.
• Target the 2026 rule change directly. Sellers who got hit with the new R$3–R$9 divergence fines have already discovered their inventory records disagree with the platform’s. That fine is the moment of awareness. Content and outbound aimed at “you were fined for divergence — you may also be owed for it” converts a punishment into a lead.
• Full-focused Facebook and WhatsApp groups. Brazilian seller communities live there rather than on Reddit, which is why English-language complaint mining turns up so little on this. Show up with recovered-amount screenshots, which is the only credential that matters in those rooms.

10. Build complexity — justification

Medium. The Mercado Libre developer platform exposes the REST APIs this needs — orders, shipments, claims, returns, billing — and a documented /claims resource that supports both reading claim details and initiating actions, so this is integration work rather than scraping. The reconciliation engine and expiry clocks are ordinary application logic.

What makes it Medium rather than Low: OAuth against a marketplace with country-specific behaviour, event data that arrives inconsistently and has to be stitched into a defensible per-unit narrative, and a fine/fee table that varies by country and gets revised. The claim-filing path also needs a graceful manual fallback wherever the API does not fully cover submission.

Realistic v1 for a technical pair: 12–14 weeks for Brazil only. Mexico and Argentina add roughly 4 weeks each, mostly rules-table and language work rather than architecture.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Filing claims a seller is contractually entitled to file, via the platform’s official API, with the seller’s authorisation.
Ethical — no harm / dark patterns	✅	Recovers money the seller is owed under published terms. Contingency pricing means the seller only pays on success. Must not file speculative claims — see risk flags.
Market exists (evidence above)	✅	Published claim windows with permanent-loss consequences, a per-unit fine regime, and a directly analogous mature category on Amazon.
1–5 person team can build this	✅	Documented REST APIs, standard stack, technical pair in ~3 months.
Launchable with <$50K / ₹40L	✅	API access, inference costs, one Portuguese-speaking ops/sales hire.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Real money, permanently lost, on a clock — and 2026 cost increases sharpened it. Docked because it is a silent leak: painful once seen, easy to ignore before that. Not quite hair-on-fire until the first audit runs.
Demand evidence	15	12/15	Strong structural evidence — published deadlines, per-unit fines, an analogous mature category on Amazon, incumbents who explicitly stop at reporting. Docked hard because I could not surface direct verbatim seller complaints in the volume I want; Brazilian sellers are in Facebook/WhatsApp groups, not indexable forums.
Build feasibility	15	11/15	Documented APIs including /claims, but multi-country rules tables, messy event stitching and a manual filing fallback push this to ~3 months.
Distribution clarity	15	12/15	The free retrospective audit with an expiry countdown is an unusually strong opener, and the agency channel compounds. Docked because the Brazilian seller community is harder to reach cold than an English-language equivalent.
Revenue mechanics	15	12/15	Contingency benchmarked at a rate the Amazon equivalent has proven, plus a subscription floor. Docked because average recoverable-per-seller is genuinely unknown and the whole ACV rests on it.
Time to first revenue	10	8/10	Contingency pricing means a seller can say yes in one call with no budget approval. First recovery cheque plausibly inside 8 weeks of launch.
Defensibility	10	6/10	Soft moat: accumulated knowledge of what claims actually get paid, per-country rules tables, and recovery history that compounds into better adjudication. But a determined Amazon-recovery incumbent could enter LATAM.
Total	100	77/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · sales-heavy

Technical because the reconciliation engine and API integration are the product. Sales-heavy because the agency channel and the Brazilian seller community are relationship businesses, not SEO. A Portuguese-speaking co-founder or first hire is not optional.

Key assumptions to validate

1. Assumption: The average mid-band Full seller has enough recoverable value per year to make 20% contingency meaningful — I am assuming a few thousand reais, and the entire ACV depends on it. How to test: Run the retrospective audit free for 20 sellers across the GMV band and measure actual recoverable amounts. This is the number that decides whether the business exists.
2. Assumption: Claims filed programmatically get paid at a rate comparable to well-argued manual claims. How to test: File 50 claims across categories and measure approval rate, partial-payment rate and time to resolution. If Mercado Livre systematically lowballs automated claims, the rebuttal layer becomes the product rather than a feature.
3. Assumption: The /claims API supports enough of the submission path that filing is not overwhelmingly manual. How to test: Build against the sandbox in week one and map exactly which claim types can be initiated via API versus dashboard-only. If it is read-only in practice, unit economics change materially.
4. Assumption: Sellers will grant API access to a new vendor that touches claims on their behalf. How to test: Offer the read-only audit to 30 sellers and measure connection rate. Read-only first is deliberate for this reason.
5. Assumption: Agencies will bring portfolios rather than treating this as competitive with their own services. How to test: Pitch 10 agencies with a revenue-share and measure how many connect more than five client accounts.

Risk flags

1. Platform dependency — severe and structural. The entire product lives inside Mercado Libre’s API and terms. They can restrict claim-related API access, change the windows, or build detection natively. Amazon has never closed the FBA recovery category despite obvious ability to, which is mild reassurance rather than a guarantee. Assume the rules table changes at least twice a year and build for it.
2. The platform is the counterparty. We are filing claims against the company whose API we depend on and whose support team is reported to push back on seller valuations. Aggressive or speculative filing is the fastest way to get the integration killed. Discipline about only filing defensible claims is a business constraint, not just an ethical one.
3. Demand evidence is thinner than I would like. The structural case is strong and the deadlines are documented, but I did not get the volume of direct seller quotes that would let me call this High confidence. The evidence gap is a language-and-platform artefact, not proof of absence — but it is an unvalidated assumption either way.
4. Contingency revenue is lumpy and uncontrollable. Revenue depends on how much the platform loses, which is not a variable we influence. The subscription floor exists specifically to blunt this, but a quarter where Full runs clean is a bad quarter.
5. Copyable by an incumbent with a head start. Amazon FBA recovery vendors have the playbook, the pricing model and the operational muscle. Their absence from LATAM is an opportunity with a shelf life.

14. Structured verdict

Score:                  77/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical pair, at least one Portuguese-speaking, ideally with
                        marketplace-seller or fulfilment-operations background
Time to revenue:        8-12 weeks post-launch (contingency pricing removes budget friction)
Capital to launch:      $15-25K (API access, inference, one PT-speaking ops/sales hire)
Top 3 assumptions to validate first:
  1. Average recoverable value per mid-band Full seller per year — free retrospective
     audit across 20 sellers. This single number decides the business.
  2. Approval rate on programmatically filed claims — file 50, measure paid vs denied
     vs lowballed.
  3. Actual write-coverage of the /claims API — build against sandbox in week one,
     map API-initiable claim types vs dashboard-only.
Kill criteria:
  - Abandon if median recoverable value across 20 audited sellers is under R$2,000/yr —
    20% contingency on that will not support the CAC.
  - Abandon if claim approval rate on filed claims falls below 40%, or if Mercado Livre
    systematically pays automated claims below manual equivalents.
  - Abandon if Mercado Livre restricts claims API access for third parties, or ships
    native discrepancy detection with automatic reimbursement.
  - Abandon if fewer than 8 of 30 approached sellers will grant read-only API access.

15. Next step — 1-week validation sprint

• Day 1–2: Build the read-only retrospective scanner against the Mercado Libre sandbox and one live seller account. Map exactly which claim types the /claims API can initiate versus which are dashboard-only. Pull the current fee, fine and claim-window tables for Brazil into a rules file.
• Day 3–4: Recruit 20 Brazilian Full sellers in the R$150K–R$3M/month band through Full-focused Facebook groups and two agency contacts. Run the free 90-day retrospective audit on each. Record, per seller: total recoverable value found, how much of it has already expired past the 20- and 60-day windows, and how much expires within 30 days.
• Day 5: Decide on one number. Go if median recoverable value across the 20 audited sellers exceeds R$2,000/yr AND at least 12 of 20 sellers say yes to a 20% contingency arrangement on the spot. Anything less and the contingency model does not carry the CAC, and I would rather learn that in a week than after a 14-week build.

The audit output is the falsifiable artefact. Twenty real numbers from twenty real accounts either clear the bar or they do not — no interpretation required.