Double (YC Summer 2021): teardown vs XMarts

In one line: Double sells month-end close software to bookkeeping and accounting firms, with AI that codes bank feeds, drafts journal entries and fixes reconciliations inside the client's ledger, priced per client [site] [docs]. Threat: 3, because it is the live, funded, named-customer version of our W5 idea (one layer for one type of small firm) and it already ships agent work with approvals [site] [press].

1. Fact file

Item	Finding	Source
What they say	Headline: "Put AI to work on your month-end close." Sub-line: "Powered by a two-way sync with the ledger." YC line: "We make software for bookkeepers."	[site] [yc]
Founded, team, funding	Founded 2020, YC S21, 80 people. $6.5M Series A reported 13 Dec 2025, led by Album Ventures with Jack Altman and YC. Total raised "more than $12.5M" (they claim).	[yc] [press]
Founders	Ben Stein (CEO; ex-CFO at Mira, ex-SpaceX operations, ex-Amazon analytics) and Kenny Song (CTO, college friend). What YC likely saw: a founder who lived the month-end pain as a CFO, plus a co-founder who builds (inference).	[yc] [site]
Customers	They claim 4,000+ accounting and bookkeeping firms in North America. Site logos include Springstead Solutions ("10 days" faster close), Attivo Partners ("1,000 hours" saved a month), MCO Bookkeeping ("95%" portal adoption), all self-reported. Net dollar retention of 150% (they claim).	[site] [press]
Price	Public, per connected client per month: Core $10, Plus $25, Scale $50 (Sept 2025 tiers). Tax Suite add-on included with a $200 a month annual commitment. Internal finance teams: flat annual plan, price on request.	[blog] [site]
Path	Keeper (2020) → renamed Double on 23 Oct 2025. Reason given by the CEO: "we got sued". Moved from practice and close tools to "AI on the close" by 2026. First customer 8 Feb 2021.	[blog] [site]

2. Their proposition, reverse engineered

Part	What they do
Who buys	Owners and managers of bookkeeping and accounting firms; also internal finance teams (separate annual plan) [site].
The pain	Month-end close is slow and scattered: "back-and-forth emails, clunky spreadsheets", about 79 open browser tabs (founder story, their words, paraphrased) [site].
The promise	Close faster and with fewer errors, keep a human in control, grow work without growing headcount [site] [blog].
How it works	Two-way sync with QuickBooks Online, Xero, Sage and NetSuite. A close page tracks tasks per client. AI sorts bank feeds into matches, rules and "needs review". A chat agent builds journal entries from plain words or uploaded files. AI proposes reconciliation fixes and writes client summaries [docs].
The output	Posted ledger entries, a checked close, client-ready commentary, receipts, 1099s, reports, a client portal [site] [docs].
Proof style	Hours saved and days cut, from customer quotes (1,000 hours a month; 10 days faster close). Firm count and retention in press [site] [press]. These are claims.
Who sells it	Own sales team: SDR, implementation specialists, customer success. Self-serve "Get started" also on the site [jobs] [site]. Partner ecosystem is a stated use of funds [press].
Price model	Per connected client, mix and match. Team members and unconnected clients free. 20% launch promo on Scale for 10 clients on an annual contract [blog] [site].
The story	Accountants should do judgment work, not drudgery. "Double" nods to double-entry and doubling a firm's capacity. A human approves every close [docs] [blog].

3. Positioning

Category they claim: AI close management and practice management for accountants [site].
Category a YC partner would put them in: vertical software with an AI layer for accounting firms (a "Pilot, but sold as software" play). The YC text says this almost word for word [yc].
What they want to be known for: the AI that works inside the ledger with a two-way sync, so nothing is copied out and back [site] [press].
Where they sit on the map: layers L1/L2/L3 = 2/2/2, buyer = sme, business.

4. Three layers vs ours

Layer	Them (score + evidence)	Us (from the system inventory, "Marton's logs show")
L1 Company brain	2. Recommendations use the client's ledger history. Client preferences are stored automatically. A finished entry can be saved as a reusable skill [docs]. Memory is per client inside accounting; no firm-wide knowledge product is claimed (inference from docs).	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning. Learns from calls and checked work.
L2 Agent orchestration	2. Several AI steps (bank feeds, journal entries, reconciliations, summaries) with human approval on every close. Task assignment on the close page. An MCP server lets Claude, ChatGPT or Gemini act on clients, tasks and close periods [docs].	About 5,600 jobs run by agents from three makers, each checked by a different AI, then automatic tests.
L3 Execution and learning	2. Posts entries to the ledger, flags miscoded items against history, proposes reconciliation fixes, saves analysis as a task comment [docs] [site]. No public claim that measured results change the next run beyond saved skills and preferences (inference).	Every failure gets a record and an automatic check (about 160). Offer feedback lines fell from 44 to 18 as rules were added. Only about 15% of jobs reached the main code, and merges were unreliable in October.

Note: these scores come from marketing pages and docs only. The depth of each claim (how much is real automation versus suggestions) is UNVERIFIED without a demo.

5. Same and different

Topic	Same or different	Them	Us
Core belief	Same	Agents plus human approval, accountants keep judgment [docs].	Same (brief: approvals and handovers).
Buyer	Same shape	Small and mid-size accounting firms, one type of firm [site].	W5 example in the brief is accountancy firms; target is SMEs and scale-ups. No customer yet.
Geography	Different	North America first: US tax forms, USD prices, Austin office [site] [yc]. UK use not shown (UNVERIFIED).	UK-based founders; UK mid-sized firms are the stated target (brief).
Kind of work	Different	Month-end close and client admin, inside the ledger [site].	A layer across work types (payroll, VAT, forecasts in the brief example). Not built for this buyer.
Owner of the result	Different	The firm owns the result; Double sells software seats per client [blog].	Brief aims to own one number on one workflow (needs proof).
How they capture	Same	Reusable skills and stored preferences after each task [docs].	Marton's logs show rules added from failures and feedback.
Proof	Different	Hours saved, days faster, retention, firm count (they claim) [site] [press].	No customer number. Our proof is our own system.
Delivery	Different	Self-serve plus onboarding specialists [jobs].	Two founders and the system.
Price	Different	$10 to $50 per client per month, public [blog].	Not set.

6. Wedge overlap

Wedge	Fit	Why
W5 One layer for one type of small firm	direct	The brief's own example is accountancy firms. Double is exactly that: one layer, one type of firm, sold to thousands [site] [press].
W7 Find and fix margin leaks for finance leaders	adjacent	Same finance buyer type, different job. Double flags miscoded items and writes variance commentary; it does not claim to find margin leaks [docs]. Its internal finance teams plan is a possible way in [site].
W8 The other 90% of an AI skill	adjacent	They save a finished entry as a reusable skill and store client preferences [docs]. No public claim of skill upkeep, retirement or drift checks (UNVERIFIED that none exists).

7. Whitespace (where they stop)

Layer gap: none we can honestly claim. They cover all three layers at level 2 inside accounting close [docs]. To own a layer gap we would need a loop Double lacks: measured results that rewrite the next run, shown with a rule-change log.
Buyer gap: no UK or European customers shown on the site; Austin office only; US-specific products (1099s) [site] [yc]. Xero is supported, which UK firms use [site]. To own it we would need UK firms who say they have the pain and a UK-fit workflow (VAT, payroll).
Number gap: proof is hours saved and days faster from customer quotes, not money or margin [site]. To own it we would need one before/after number a finance person would sign.
Price gap: per-client pricing is cheap ($10 to $50) [blog]. A small firm can already pay it. We do not own a price gap. A result-based price would be a different model, not a cheaper one (inference).
Speed gap: self-serve start exists [site]; first-value time is not stated (UNVERIFIED).

8. For the YC application

Proven today: Marton's logs show his system already runs agents from three makers that do and check work, about 5,600 jobs with each one reviewed by a different AI, and about 160 failure records that each added an automatic check. Double's public proof is hours saved by firms, in North America [site].
Draft answer to "how are you different from Double?": DRAFT. "Double is a good product for US bookkeepers and works inside the ledger. We are a different buyer and a different job: we are not selling a close tool, and we will say what we do for UK firms once we have run it." Founders say it their own way.
Future claim (needs proof): "On one UK firm's workflow our agents do the work and the number moves." Needs one customer workflow run for 4+ weeks with a before/after number.
What we understand that they don't: needs proof. Nothing proven yet separates us from Double on the full loop; they already ship agent work with approvals and reusable skills [docs]. Our edge on W5 is only the buyer (UK) and the number, and both are unproven.
Copyability (M3): easy. Any move we make inside accounting is a feature for them: they already have the ledger sync, 4,000+ firms (they claim) and a weekly release pace [site] [press]. They would want it because it fits their buyer. Chasing UK-first is the only thing that pulls them off focus, and that is slow, not hard (inference). Not hard, so M3 cannot score above 3 against Double.
Why now (M4): None found.
Do not claim: "no one does the full loop" (Double scores 2 on all three layers in one vertical). "Nobody serves accounting firms with agents." "UK firms are ignored" (UNVERIFIED). Any claim about Double's real automation depth without a demo.

M2 (what YC funds now) is NOT judged here.

9. Ideas worth taking

Per-client, mix-and-match pricing that mirrors how a firm packages its own services [docs]. Easy for a small buyer to understand.
An MCP server so the buyer's own AI (Claude, ChatGPT, Gemini) can work inside the product [docs]. Lets customers bring the agent they already pay for.
Save the instructions as a reusable skill the moment a task is done [docs]. Same shape as our own habit of turning feedback into a rule.

10. Unknowns

Whether they have UK or European customers, a UK data store or UK tax features. Look at their Xero UK marketplace page and customer list.
How much of the AI is real automation versus suggestions a human redoes. Needs a demo or a customer who uses it.
Revenue or ARR. Not public; the only numbers are firm count and retention (claims).
Anything about the lawsuit that forced the rename. The CEO gives no details [blog].
Whether "Ask Double" and the voice commands are live for all plans (seen only in a promo page via search).

Sources

Double home page [site]
Double pricing [site]
Double AI Workflows [docs]
Double about us [site]
YC company page: Double [yc] [jobs]
Keeper is now Double, 23 Oct 2025 [blog]
Introducing Double Scale, Sep 2025 [blog]
Double blog index [blog]
Double: $6.5 Million Series A (Pulse 2.0), 13 Dec 2025 [press]
Keeper Has Rebranded to Double (CPA Practice Advisor), 24 Oct 2025 [press] (seen in search results only)
Double on the Xero UK app store [site] (seen in search results only)

Double

What they do

How close to us

Wedges

Who buys

Full teardown