Cifrato (YC Winter 2025): teardown vs XMarts

In one line: Cifrato sells AI agents that book supplier invoices, match them to accounts and taxes, and post them into the accounting software, mainly for accounting firms in Colombia and Mexico [site] [yc]. Threat: 2, because it is the closest live example of our W5 idea (one operating layer for one type of small firm) and claims real revenue, but it only sells in Colombia and Mexico today [site] [press].

1. Fact file

Item	Finding	Source
What they say	Site: "The agentic engine for accounting". YC line: "AI Accountant for Accounting Firms".	[site] [yc]
Founded, team, funding	YC lists 2025 as founding year, Bogotá, 17 people. Forbes Colombia said 13 in March 2026, so the team grew or sources differ. Seed of "over $1M" announced 25 to 26 Mar 2026 from Y Combinator, Fen Ventures, Decelera Ventures and Kuiper.	[yc] [press]
Founders	Juan Pisco, CEO, 21. Started writing machine learning code at 16, sold his first company at 19, built the travel planner Triplanner (they claim 3M users, sold to Layla). Yerson Cacua, ex VP Technology at Rappi, co-founder of Sumer. What YC likely saw: a young technical founder with a prior exit plus a senior operator from a large Latin American company (inference). The idea came from watching invoices typed by hand at a family restaurant in Bogotá [press].	[yc] [press]
Customers	Logos on the site: Grupo Gordo, Le Grand, Beer, Russell Bedford, Todos Comemos, Onco Life. A testimonial is named for "EMG - Asesorias contables", but the text did not load. They claim "3.5k+ companies". Forbes: more than 270 direct clients, indirectly serving about 3,500 firms. The two claims count different things (inference).	[site] [press]
Traction (all they claim)	About $1M ARR in under 12 months; over 1 million invoices worth over $1B processed; 98.7% accuracy; over 75% would be "very disappointed" without it (Sean Ellis test); operating breakeven expected the month after March 2026; 2026 goal $5M ARR and 10,000+ businesses.	[press]
Price	Not public. Forbes: monthly subscription based on invoice volume and number of integrations, no standard plans. One blog post explains how to pay the subscription, with no figures. Revenue per direct client is roughly $3.7k a year if both claims hold (inference: $1M ARR over 270 clients).	[press] [blog]
Path	Triplanner (consumer travel) sold; Cifrato founded 2025; YC W25; positioned as an "accounting copilot" in Nov 2025 blog; seed March 2026; site now says "agentic engine". YC job page says "Automating finops for businesses", the YC tagline says accounting firms, and the site also lists retail and restaurants. No rename found.	[blog] [site] [yc] [press]

2. Their proposition, reverse engineered

Part	What they do
Who buys	Accounting firms first. A typical firm manages 3 to 20 companies and 500 to 1,500 purchase invoices a month [press]. The site also lists retail and restaurants [site].
The pain	Typing invoices into the ERP by hand; firms win clients but margin does not improve and tax filings slip. Their blog calls this "the invisible ceiling" of accounting firms (paraphrase) [blog].
The promise	"Grow your client base, not your team." An invoice in 0 to a few seconds against 360 seconds by hand (the number did not load on the page) [site].
How it works	Pulls invoices from the tax authority (DIAN in Colombia, CFDI in Mexico) and email; classifies each to accounts, taxes and cost centres; shows only exceptions; posts to the customer's ERP after sign-off; builds reports [site]. Uses agreement between several AI models to cut errors [press].
The output	Booked entries in the client's own accounting software, plus live reports and "leadership signals" [site]. YC copy adds invoice payment, reconciliation and tax filings [yc]. Bank reconciliation is a module they said they would build [press].
Proof style	Counts of invoices, money processed, accuracy and seconds per invoice. No money saved for the customer. [site] [press]
Who sells it	Own sales team. Half of the team is sales [press]. The one open role is a Founding Sales Development Representative in Bogotá, $1.5K to $3.5K a month plus 0.03% to 0.05% equity [jobs]. Demo-only call to action [site].
Price model	Subscription by volume and integrations [press]. Not on the site [site].
The story	Enemy: manual transcription. Belief: "AI will not eliminate the accountant" over the next decade (their Spanish: "la IA no va a eliminar al contador") [press]. Build infrastructure around the best models rather than their own model [press].

3. Positioning

Category they claim: "Agentic engine for accounting" [site]. Earlier: "accounting copilot" [blog].
Category a YC partner would put them in: vertical AI for accounting firms in Latin America, an AI-native accounting back office (inference).
What they want to be known for: the AI accountant that connects straight to the tax authority and ERP, so no one types invoices.
Where they sit on the map: L1/L2/L3 = 1/2/2, buyer = sme, business. Sells to accounting firms in Colombia and Mexico.

4. Three layers vs ours

Layer	Them (score + evidence)	Us (Marton's logs show)
L1 Company brain	1. They claim each booked invoice improves future accuracy and that the system learns each client's accounting patterns [site] [press]. That is a memory of accounting rules per client, not a brain of the whole firm.	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning.
L2 Agent orchestration	2. Several steps in sequence (intake, classify, approve, post, report) with human approval and several models checking each other [site] [press]. No sign of a visible team of agents with handovers (inference).	About 5,600 jobs run by agents from three makers, each checked by a different AI before it can merge.
L3 Execution and learning	2. Does the work, a human signs off, and they report accuracy [site] [press]. The learning loop is a claim without a public log or result (UNVERIFIED).	Every failure gets a record and an automatic check (about 160). Offer feedback lines fell from 44 to 18 as 22 rules were added. Only about 15% of jobs reached the main code, and merges were unreliable in October.

5. Same and different

Topic	Same or different	Them	Us
Belief	Same	Agents do the work, people approve the exceptions: "Nothing posts without human sign-off." [site]	Marton's logs show agent work is checked by a different AI, then by tests, before it merges.
Buyer	Same	Small accounting firms, 3 to 20 clients each [press].	W5 targets one type of small firm, accountancy as the example (brief).
Place	Different	Colombia and Mexico, tied to DIAN and CFDI [site].	UK mid-sized firms is the target (brief). No customer yet.
Owner of the result	Same	They post entries; the firm and client own the books [site].	Not yet tested in accounting.
How they capture	Different	Direct feed from the tax authority and email [site].	Calls and notes filed into one vault (Marton's logs show). No accounting feed.
Kind of work	Same	Repeating, rule-heavy back-office work.	Marton's logs show repeating build and content work on his own business.
Size	Different	17 people, $1M seed, about $1M ARR (they claim) [yc] [press].	Two founders, no outside money, no customer.
Delivery	Different	Software plus a large sales team [press].	The system plus the two founders.
Proof	Different	Invoices processed, accuracy, Sean Ellis score (they claim) [press].	No customer number yet. Own-business numbers only.
Price	Different	Subscription by invoice volume; no public plans [press].	Not set.

6. Wedge overlap

Wedge	Fit	Why
W5 One layer for one type of small firm	direct	Same idea, same buyer type: a back-office layer for accounting firms, built once and sold to many [yc] [press]. The brief's own W5 example is an accountancy layer above the accounting software. They have done the Latin American version first.
W7 Find and fix margin leaks for commercial or finance leaders	adjacent	Same finance world, but they book and reconcile; they do not hunt margin leaks. Their blog does name margin as a firm pain [blog].

7. Whitespace (where they stop)

Buyer gap: no UK or European customers, offices or tax links on the site; only Colombia and Mexico [site]. Their edge is local tax rules, which do not carry over (inference). To own it we need UK accountancy firms (HMRC, Xero or similar) saying they have the pain.
Number gap: they count invoices and accuracy, not money or margin won back for the firm [site] [press]. To own it we need one before/after number a firm partner would sign, such as hours per client per month or filings missed.
Price gap: the open sales role pays $1.5K to $3.5K a month [jobs], which suggests a low-price volume model (inference). A UK firm price would be set very differently. To own it we need a price test with 3 to 5 firms.
Layer gap: weak. They already do and check the work (L3 2). The one real gap is a firm-wide memory (L1 is 1): their learning is per invoice, not per firm. To own it we need a visible rule log tied to checked runs and falling rework.
Speed gap: not found. Demo-only entry, no stated time to first value [site].

8. For the YC application

Proven today: Marton's logs show his system already runs agents that do work and have it checked by a different AI: about 5,600 jobs, about 160 failure records that each added an automatic check. We have not run any accounting work. Cifrato claims about $1M ARR and 270+ direct clients in Colombia and Mexico [press].
Draft answer to "how are you different from Cifrato?": DRAFT. "Cifrato does invoice booking for accounting firms in Latin America, and they are ahead of us there. We are not in accounting yet. Our system already runs agents that do work and check it, on our own business, and we want to bring that to one kind of UK firm." Founders say it their own way.
Future claim (needs proof): "For UK accountancy firms we own one number, hours or errors per client, and the layer learns from every checked run." Needs one UK firm's workflow run for 4+ weeks with a before/after number, and a rule log tied to checked runs.
What we understand that they don't: a firm-wide memory that learns from checked results, not only per-invoice accuracy. needs proof; Marton's logs show it only on his own work (offer feedback lines fell from 44 to 18 as rules were added).
Copyability (M3): medium. They could add a firm-wide memory and agent handovers, since they already learn per client and use several models [press] (inference). The UK is a different tax system, ERP set and sales motion, and they have just raised to deepen Colombia (use of funds: Colombian operations, ERP and POS links, bank reconciliation) [press], so a UK move is unlikely soon (inference). Not hard: M3 can not score above 3 against Cifrato for W5.
Why now (M4): None found. Do not use "agents can now do the work": that is general AI progress.
Do not claim: that nobody sells an AI layer to accounting firms; that we do accounting work today; any figure for UK demand without a named firm; "company brain" as the headline.

9. Ideas worth taking

Free tools as a front door: their blog offers ten free tools (account-code search, tax-ID lookup, tax calendar, holiday-pay calculator) [blog]. Cheap leads from a firm's own daily questions.
The Sean Ellis "very disappointed" score as proof, reported with a share [press]. Cheap to run on our first pilots.
A trust line near the top: nothing posts without human sign-off [site]. It answers the first worry of a finance buyer.

10. Unknowns

Real price per firm and the contract length. Ask a firm owner who has seen a quote.
Whether the ARR and client numbers hold: Forbes figures are theirs, the site's metric values did not load, and the site says "3.5k+ companies" while Forbes says 270 direct clients.
Whether tax filing is live or planned: the YC text lists it, the site stresses compliance, the March 2026 use of funds says bank reconciliation is still to build.
Integration list: a search summary named five ERP and POS tools, but I could not open its source (UNVERIFIED).
Any plan to enter Spain or the UK. Check their careers page and later press.

Sources

cifrato.ai (home) [site]
Cifrato blog index [blog]
Blog: Más clientes, menos margen, 20 Apr 2026 [blog]
Blog: copiloto contable para cerrar el mes, 18 Nov 2025 [blog]
Book a demo page [site]
YC company page [yc]
YC jobs page [jobs]
Forbes Colombia: Cifrato, 25 Mar 2026 [press]
LatamList: Cifrato raises over $1M, 26 Mar 2026 [press]
HuntScreens listing [press]

Cifrato

What they do

How close to us

Wedges

Who buys

Full teardown