Fianza

1. One-liner

Your reputation guarantee buys five problems. This spends them carefully and fights the ones that shouldn’t count.

2. Trend signal — why now?

Mercado Libre runs a programme almost nobody outside the platform writes tooling for. A seller sitting on red, orange or yellow reputation can post a cash guarantee in Mercado Pago and have their thermometer frozen at green while they trade their way back. It runs in Argentina, Brazil, Mexico, Chile and Colombia, and Mercado Libre exposes it through a developer API as RECOVERY_GRNTEE, with endpoints at /users/reputation/seller_recovery/status, /activate and /cancel_guarantee (Mercado Libre developers).

The terms are where it gets interesting. The benefit runs for roughly 120 days, the guarantee amount is negotiated per seller against their sales volume, and the seller is allowed a limited number of “ventas afectadas” — sales with a claim, a cancellation by the seller, or a delayed dispatch. Cross the limit and the benefit ends. And the money is refunded pro rata against that limit:

“Si el número de ventas afectadas es 5 y tienes 3 ventas afectadas, cuando finalice el beneficio te cobra el 60% y recuperarás el 40% restante.”

Exceed the limit and you lose 100% (UpSeller).

Read that again, because it is the whole idea: on a five-problem allowance, each individual problem costs the seller roughly 20% of their deposit, in cash. Not a score, not visibility — money, charged at the end of the window. A seller with a guarantee posted is running a budget of five, and every claim that lands is a withdrawal.

Two things make that budget harder to manage right now.

The measurement window just changed underneath small sellers. From 16 September 2026 in Argentina, Mercado Libre raised the reference threshold from 50 to 101 sales in 60 days. Sellers hitting 101+ are judged on the last 60 days; everyone below is judged on the last 365 days. Cancellation tolerance is 1% on the 60-day window and 1.5% on the 365-day one. Chile changed on 31 August 2026 (UpSeller Argentina; UpSeller Chile). For a small seller this is a trap with a long memory: a bad month sits in the denominator for a year, and the fast exit — crossing 101 sales in 60 days — is exactly what a damaged reputation makes harder.

And Mercado Libre publishes a rubric for getting problems removed. The reglas de exclusión list eleven qualifying grounds, live across Argentina, Mexico, Chile, Colombia and Uruguay: the buyer opened the claim by mistake; the buyer does not recognise the purchase; the product shows delivered but the buyer says otherwise; “el comprador se arrepiente de la compra, pero el producto está en perfectas condiciones”; “el comprador usó el reclamo como medio de contacto”; postal or Mercado Envíos delays where dispatch was inside the agreed plazo; claims tied to a global platform failure; claims opened “por competencia o venganza”; claims raised inside the “8hs hábiles desde el primer contacto del comprador” messaging window; size or model changes in autoparts, clothing, bags and footwear; and unauthorised account access, fraud or incorrect suspension by ML (Centro de Vendedores).

The route is Métricas → Atención a tus compradores → “Revisar todos” → Ventas con problemas — and Mercado Libre explicitly supports selecting many at once: “podés usar la barra superior para seleccionar varias a la vez.” An in-product batch button for a per-event argument is the platform admitting the volume exists.

The clocks are short, published, and automatic. On a claim with cancellation the seller must refund “dentro de los 2 días hábiles” or notify ML, or protection is lost. On a delayed package, “si despachas el producto en los 2 primeros días hábiles de iniciado el reclamo, tu reputación seguirá sin afectarse” — and for self-managed shipping you must also notify ML through the messages. Then the sentence that defines the product’s job: “la buena gestión del reclamo evitará la afectación de tu reputación del reclamo, pero no eliminará la que ya traes por la demora en el envío” (Centro de Vendedores MX). One event, two metrics, two separate fights — and handling one well does not undo the other.

Separately, the cancellations help page says postal failures and ML-caused cancellations should not hurt you — but documents no dispute route for them. The rule lives on one page; the button lives on another.

Provenance:

• Signal 1 (demand): Mercado Libre’s reputation benefit escrows a negotiated cash guarantee against a limited allowance of affected sales and refunds it pro rata — 3 problems against a limit of 5 means ML keeps 60%, and exceeding the limit forfeits 100% — so each individual claim, cancellation or late dispatch carries a direct cash price — https://www.upseller.com/es/blog-article-72 — accessed 2026-09-17
• Signal 2 (feasibility): The programme is exposed via API (RECOVERY_GRNTEE, /users/reputation/seller_recovery/status) returning order counts, total problems and the individual counts of claims, cancellations and delayed dispatches, plus guarantee reserve and charge amounts, across MLA, MLB, MLM, MLC and MCO — https://developers.mercadolibre.com.ar/es_ar/recuperacion-reputacion — accessed 2026-09-17
• Signal 3 (economic): The reputation the guarantee protects controls real money — yellow or orange removes listings from relevant search rankings, excludes the seller from premium Mercado Libre Ads campaigns, drops them from Mercado Líder Platinum and cuts subsidised interest-free instalments, forcing paid ads to compensate and raising cost per sale; meanwhile certified ML agencies sell the manual version of this job over WhatsApp — “se arma un sistema de evidencia y un proceso para que tu reputación no se destruya por casos mal encuadrados” — https://algoritmodigital.com.ar/devoluciones-y-reclamos-patron-consultor-certificado-mercado-libre-argentina/ — 2025-12-03 Category: Platform shift

3. The opportunity

The tooling market around Mercado Libre is mature and aimed elsewhere. Real Trends sells a “Control de reputación” that promises to “analizá tus métricas para que tu termómetro esté siempre verde” — analysis, not action. Nubimetrics sells market and competition intelligence, and its own educational page on low reputation was still publishing the superseded “menos de 50 ventas” threshold as of 17 September 2026, a day after the rule changed. UpSeller, WooSync, ProfitOS, Base.com and Smartselling publish explainers. Astroselling, Multivende, Anymarket and Jumpseller sync listings and inventory and do not treat reputation as a feature at all.

Every one of them reports the score. None contests the events that produce it. Real Trends even blogged the exclusion process back in 2020 and still has not productised it.

The people actually doing this work are agencies, by hand, over WhatsApp. Algoritmo Digital, certified ML consultants in Argentina, describe the job precisely (2025-12-03): “Una agencia te ayuda a presentar evidencia y encuadrar el caso correctamente para proteger la reputación”, and “Devoluciones y reclamos no se arreglan con ‘buena voluntad’. Se arreglan con método, evidencia y proceso… Se arma un sistema de evidencia y un proceso para que tu reputación no se destruya por casos mal encuadrados.” No published pricing; contact is WhatsApp. When a job is monetised by agencies and by nobody in software, that is the signal.

And the sellers describe the failure mode themselves. One seller took four cancellations, at least two of which an ML support agent had instructed her to make, and found them counted anyway: “Mercadolibre toma las cancelaciones de los clientes para el cálculo de reputación, aún cuando el cliente haya experimentado problemas con la plataforma.” Another coordinated a delayed dispatch with the buyer and still got flagged, watching his delayed-shipment metric go from clean to 30% despite positive feedback. A third complains that buyers make fictitious purchases or use purchases as a contact channel, forcing seller cancellations that count against him — describing, without apparently knowing it, two textbook exclusion grounds (tuquejasuma.com). That last one is the whole opportunity in a sentence: the remedy is published and the affected seller cannot find it.

That leaves a specific, monetised gap. A seller in the recovery programme has:

• A budget with a price per unit. Five problems, roughly 20% of the deposit each. This is unusual and valuable: most marketplace-score products have to argue that a score matters eventually. Here the seller can be told, in pesos, what this morning’s claim just cost.
• A published rubric for removing problems. Eleven named grounds, a batch-submit review button, and — notably — no documented deadline, no stated limit on requests, and no specified evidence standard. Ambiguity the seller navigates per claim, under published two-day clocks, while the meter runs.
• A rule scattered across pages. Postal-failure and ML-caused cancellations supposedly do not count, but the dispute path for them is not written down next to the rule.

So the product is not “watch your thermometer”. It is: treat the guarantee as a budget, price every incoming problem against it, and automatically identify and argue the ones that qualify for exclusion. Every successful exclusion is worth roughly a fifth of the deposit, which makes the value proposition arithmetic rather than rhetorical.

There is a second, sharper reason this is unbuilt. The honest first output of this product is often “do not activate the guarantee yet” — because a seller with a shaky dispatch process will burn five problems in three weeks and forfeit everything. That is advice against transacting, which neither Mercado Libre nor the integration vendors whose pricing scales with the seller’s sales volume have any incentive to give.

4. Target market

• Primary customer: Mercado Libre sellers in Argentina, Mexico, Brazil, Chile or Colombia who are currently in the recovery programme with a guarantee posted, or who are red/orange/yellow and considering it. Typically small operations — one to ten people, often a family business or a single operator with a warehouse — with enough volume that the reputation drop is costing real sales.
• Why they buy: They have cash locked in Mercado Pago that they lose in 20% increments. Nothing else they own tells them what a given problem costs, and nothing helps them argue that a particular claim should never have counted.
• Secondary customer: the seller sitting at yellow deciding whether to post a guarantee at all, who needs the readiness answer before committing money.
• Rough TAM reasoning: soft, and I will not dress it up — Mercado Libre does not publish seller counts, nor how many sit in the recovery programme. What is known: ML reported roughly USD 28.9B of revenue in 2025, with Argentina GMV up 42% year on year and Mexico units up 45% (Infobae, 2026-02-24). Across five countries, a programme with a dedicated API and a negotiated-guarantee sales motion implies a population in at least the tens of thousands. At US$39/month and 2,200 paying sellers that is ~$1M ARR. Establishing the real denominator is the first job of the validation sprint.
• Why now for them: the 16 September 2026 Argentine change lengthened the penalty window for exactly the small sellers most likely to need recovery, and Chile moved two weeks earlier.

5. Product sketch (MVP)

• Budget meter in pesos, not points. Connects via the seller’s own Mercado Libre authorisation and reads the recovery status endpoint: problems used, problems remaining, days left, and what each remaining problem is worth against the posted guarantee.
• Exclusion triage, ranked by what it is worth. Each claim, cancellation and late dispatch is checked against the eleven published grounds and the cancellation carve-outs. Mercado Libre’s own dashboard already flags whether a case is eligible — what it never does is tell you which of forty eligible claims is the one holding you at yellow. The seller reputation API returns, per metric, both the headline rate and an excluded object carrying real_value and real_rate, so the gap between them is computable and every candidate exclusion can be ranked by the movement it would produce.
• Proof in the seller’s own numbers. “These six exclusions took your claims rate from 1.9% to 1.2% — amarillo to verde” is a report the API can substantiate, which is a different conversation from a dashboard asking to be trusted.
• Two-clock alarms. The 2-business-day refund window, the 2-business-day dispatch window and the 8-business-hour messaging window are where protection is won or lost automatically — and because good claim handling “no eliminará la que ya traes por la demora en el envío”, the shipping-delay hit needs its own separate fight, flagged separately.
• The exclusion request, drafted. For qualifying cases, the argument and the supporting evidence — messages, tracking events, timestamps — assembled in the form the review expects, ready to submit through Métricas.
• Readiness check before you post the guarantee. Reads the last 120 days of claims, cancellations and dispatch timing and estimates how many problems the seller would realistically incur in a window. If the honest answer is “more than five”, it says so and names what to fix first.
• Window-rule awareness. Tells a sub-101 seller how the 365-day window is holding old problems against them, and how far they are from the 101-sale threshold that switches them to the 60-day measure.
• Outcome ledger. Which exclusion requests were filed, which were granted, and the guarantee cash recovered as a result — the number the seller renews on.

6. AI angle — what’s load-bearing

The load-bearing work is arguing an ambiguous case from unstructured evidence, not arithmetic.

Deciding that a claim was opened “por competencia o venganza”, or that a buyer regretted a purchase and the item is in perfect condition, or that a postal delay fell inside the agreed dispatch window, means reading buyer–seller message threads in Spanish or Portuguese, cross-referencing tracking events and timestamps, and matching the result to a rubric that Mercado Libre publishes as prose with no evidence standard attached. Then it means writing the request that a reviewer will accept.

That is language work on messy multilingual evidence at a per-claim cost that has to stay well under the ~20%-of-deposit value of winning. Strip it out and you have a dashboard showing a counter that Mercado Libre’s own API already returns — which is precisely why the existing vendors, who are excellent at dashboards, have not built this.

The readiness estimate is the honest exception: that is ordinary statistics over the seller’s own history, and it should not be dressed up as anything cleverer.

7. Localization angle

This is a LATAM-only product and the localisation is the moat. The exclusion rules, the guarantee mechanics, the thermometer and the 101-sale window rule are Mercado Libre constructs that differ by country — Argentina changed on 16 September 2026, Chile on 31 August, and Mexico and Brazil run their own variants. It needs Spanish and Brazilian Portuguese, and it needs them at the level of arguing a case, not translating a UI.

Pricing has to survive Argentine inflation and currency volatility, which argues for USD pricing with local payment rails through Mercado Pago. No global seller-tools vendor is going to do this work for a programme that exists in five countries and is documented mostly in Spanish.

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

• Pricing: US$39/month while a guarantee is active, which is the window where the value is undeniable. US$19/month maintenance outside it. A 20% success fee on guarantee cash recovered through granted exclusions as the alternative for sellers who prefer outcome pricing.
• ACV: ~$300–470/year for a seller who stays subscribed through and past a recovery window.
• Rough math to $1M ARR: ~2,200 sellers at $39/month, or fewer with the success fee on larger guarantees. Across five countries this is a small share of the addressable base — if the base is what the programme’s existence implies.
• Rough math to $5M ARR: needs the product to outlive the guarantee window. That means becoming the seller’s standing claims-defence layer — every claim triaged against the exclusion rules whether or not money is escrowed — plus expansion into the other Mercado Libre penalty surfaces.
• Expansion path: guarantee window → permanent claims defence → the sub-101 window management → the same exclusion logic applied to Mercado Envíos dispatch penalties.

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

1. The recovery programme is a self-identifying segment with a public marker. A seller in the programme shows green while their history says otherwise, and their listing and history are public. Sellers with recent claim patterns and a freshly green thermometer are findable at scale, and they are precisely the people with cash at stake this month.
2. Sell the readiness check free, to people about to post money. “Before you deposit a guarantee, find out how many problems you would actually incur” is a genuinely useful free tool, requires connecting the account that the paid product needs, and reaches the seller at the moment of maximum attention — before the deposit, not after.
3. The Mercado Libre consultant ecosystem is organised and monetised. There is an established layer of certified consultants and course-sellers — Algoritmo Digital, Néstor Arranz and similar operators with real audiences on TikTok and YouTube teaching people to sell on ML. They already answer “my reputation dropped, what do I do” for free. A revenue-share tool that answers it properly is an easy partner pitch.
4. Answer the reputation threads in ML’s own community. comunidades.mercadolibre.com.ar carries a continuous stream of sellers asking how to get an unfair claim removed. Answer with the actual exclusion ground that applies to their case. Slow, unscalable, and the fastest route to knowing whether the rubric works in practice.
5. Publish the exclusion rules properly. Mercado Libre documents the eleven grounds in prose across several help pages; nobody has turned them into a clear per-scenario guide. That guide is both the SEO asset in a language and niche with thin competition, and the product’s own logic made visible.

10. Build complexity — justification

Low, unusually so for this catalogue. There is an official API with the exact fields needed — problem counts by type, guarantee reserve and charge amounts, window dates — and OAuth for seller authorisation. No hardware, no scraping, no credential storage. The evidence for a claim is already in the ML message threads and order records the API exposes.

The real work is the exclusion rubric: encoding eleven grounds across five countries, learning what reviewers actually accept versus what the help page says, and keeping up as rules move (two changed in the last three weeks). That is domain accumulation, not engineering. A technical founder with Spanish ships v1 for one country in 6–8 weeks.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Operates on the seller’s own data through Mercado Libre’s official API with their authorisation. Must stay within ML’s API terms — verify third-party access to the seller_recovery endpoints before building.
Ethical — no harm / dark patterns	⚠️ Pass with a condition	This files exclusion requests. Filing weak or fabricated grounds to dodge legitimate buyer complaints would corrupt a system that exists to protect buyers, and would poison the reviewer’s trust in every request. Only file where the published ground genuinely applies, keep the evidence honest, and let “no grounds” be a common answer. The readiness check must also be willing to say “don’t post the guarantee”.
Market exists (evidence above)	✅	An official programme with escrowed cash, an API, and five countries.
1–5 person team can build this	✅	One or two people, 6–8 weeks for the first country.
Launchable with <$50K / ₹40L	✅	API integration and content; no capital requirement beyond time.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Cash is escrowed and drains in ~20% increments per problem, on a 120-day clock, with 100% forfeited past the limit. That is about as sharp and immediate as marketplace pain gets. Docked because it only bites the subset who have actually posted a guarantee — outside that window the pain reverts to an ordinary score worry.
Demand evidence	15	13/15	An official programme with a dedicated API, eleven published exclusion rules with a batch-submit UI, a documented pro-rata forfeiture formula, dated rule changes in two countries, certified agencies selling the manual version over WhatsApp, and collective seller complaints including one describing two textbook exclusion grounds without knowing the remedy exists. Docked because the sourced seller quotes predate the 2026-09-16 change and ML publishes no seller-count denominator.
Build feasibility	15	12/15	Official API with exactly the right fields, OAuth, no scraping. Docked for multi-country rubric encoding and for the unverified question of whether third parties may call the seller_recovery endpoints.
Distribution clarity	15	11/15	Self-identifying segment, a free readiness check at the moment of payment, an organised consultant ecosystem, and ML’s own community forum. Docked because identifying who is in the programme is inferential rather than queryable, and the consultant channel needs relationship-building.
Revenue mechanics	15	10/15	$39/month against a guarantee worth several times that is easy to justify, and the success fee aligns perfectly. Docked hard on churn: the acute need lasts 120 days, so this is structurally a subscription with a built-in expiry unless the permanent claims-defence layer works.
Time to first revenue	10	7/10	Low build complexity and an urgent, self-identifying customer. Realistically 8–10 weeks to first paid seller.
Defensibility	10	4/10	Thin. The API is public, the rules are published, and Real Trends or Nubimetrics could add this to an existing distribution base in a quarter if it proves out. The only durable moat is accumulated knowledge of what reviewers actually accept, plus speed.
Total	100	73/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · content-heavy — with native or fluent Spanish non-negotiable and Brazilian Portuguese needed for MLB. The content half matters more than usual: the distribution plan runs on being genuinely useful in forums and to consultants, in the seller’s language.

Key assumptions to validate (3–5)

1. Assumption: Third parties can call the seller_recovery endpoints with seller authorisation, and ML’s API terms permit this use. How to test: Register a developer application and attempt the calls against a test seller before writing anything else. If this is blocked or restricted, the product falls back to manual data entry and gets much worse — verify first.
2. Assumption: Exclusion requests are granted often enough to be worth automating. How to test: Recruit ten sellers, triage their last 90 days of claims against the eleven grounds, file the qualifying ones, and measure the grant rate. Below roughly a third and the value proposition weakens badly.
3. Assumption: Excluded claims actually stop counting toward the guarantee’s affected-sales limit. How to test: This is the single most important unknown and it is not documented anywhere I could find. Confirm with Mercado Libre support and against a real seller’s status endpoint before and after a granted exclusion. If exclusions do not reduce the guarantee counter, the core pitch collapses to reputation-only value.
4. Assumption: Enough sellers are in the programme to build on. How to test: Sample seller histories for the green-thermometer-with-bad-history signature, and ask ML consultants how many of their clients have posted guarantees.
5. Assumption: There is a deadline on exclusion requests. How to test: No exclusion page states a plazo — this is genuinely unsourced. Check the seller UI for an expiry on old claims and ask support directly. If requests are accepted indefinitely, the urgency wedge weakens and the product leans entirely on the published two-day protection windows and the guarantee clock instead, which changes the marketing but not the mechanics.

Risk flags

1. Assumption 3 is a live kill risk, not a nuance. The entire “each problem costs 20% of your deposit, so fight it” argument depends on a granted exclusion removing a problem from the guarantee count. Nothing in the published material confirms it. Test it before building.
2. Platform dependency, in the strongest form. Mercado Libre owns the API, the rules, the review decision and the customer relationship. It changed reputation rules in two countries within three weeks. It could publish its own exclusion assistant, restrict the endpoints, or retire the guarantee programme, and any of those ends the business.
3. Built-in churn. The 120-day window is the product’s own expiry date. Without the permanent claims-defence layer, this is a business that has to re-acquire its customer base several times a year.
4. Weak defensibility against incumbents with distribution. Real Trends and Nubimetrics have the sellers already. The head start is measured in months.
5. Seller voice is real but stale, and the denominator is unknown. The quotes sourced are genuine collective complaints, but they predate the 16 September 2026 change — no reaction to the 101-sale rule itself was found, one day in. More importantly, Mercado Libre publishes no seller counts, so the size of the affected band is an open question rather than a number. ML’s own community forum (comunidades.mercadolibre.com.ar) resists automated fetching and is the richest unmined vein; work it with a browser before committing.

14. Structured verdict

Score:                  73/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Spanish-fluent technical founder comfortable in marketplace
                        APIs and in LATAM seller communities; Portuguese for MLB later
Time to revenue:        8–10 weeks
Capital to launch:      US$8–15K
Top 3 assumptions to validate first:
  1. Granted exclusions actually decrement the guarantee's affected-sales counter —
     test against a real seller's status endpoint before and after. If not, stop.
  2. Third-party API access to the seller_recovery endpoints is permitted under
     ML's terms — register an app and try the calls
  3. Exclusion grant rate is high enough to matter — triage and file for 10 sellers,
     measure what proportion is granted
Kill criteria:
  - Abandon if granted exclusions do not reduce the guarantee counter — the cash
    argument is the product, and without it this is another dashboard
  - Abandon if fewer than a third of well-grounded exclusion requests are granted
  - Abandon if ML restricts third-party access to the recovery endpoints
  - Abandon if Real Trends or Nubimetrics ships exclusion triage before v1

15. Next step — 1-week validation sprint

• Day 1: Register a Mercado Libre developer application. Call /users/reputation/seller_recovery/status against a consenting seller. Confirm the fields, the permission model, and that the terms allow this. If the endpoint is closed to third parties, stop and rethink the shape now rather than after a build.
• Day 2–3: Recruit ten sellers from the ML community forums and the consultant ecosystem — ideally a mix of in-programme and considering-it. Triage their last 90 days of claims and cancellations against the eleven published exclusion grounds by hand. Count how many qualify.
• Day 4: File the qualifying requests through Métricas. While waiting, put the decisive question to Mercado Libre support in writing, in Spanish: does a granted exclusion reduce the affected-sales count under an active reputation guarantee?
• Day 5: Measure grant rate and read the answer.

Falsifiable outcome: proceed only if (a) the API is accessible to third parties with seller consent, (b) at least a third of well-grounded requests are granted, and (c) Mercado Libre confirms — or a real seller’s status endpoint demonstrates — that exclusions decrement the guarantee counter. Fail (c) and this is a reputation tool competing with incumbents on their turf, which is a materially worse business and should be scored again from scratch.