Soraban (YC Winter 2021): teardown vs XMarts

In one line: Soraban sells US accounting firms a tool that collects client tax documents, builds the workpapers, types the data into their tax software and sends the signed return back with a payment link [site]. Threat: 2 because it is the clearest proof that "one layer for one type of firm" sells (W5), and it is funded and live, but it only covers US individual tax season, not the UK work in W5 [site] [press].

1. Fact file

Item	Finding	Source
What they say	"The work behind the return. Handled." Tagline: "Tax Workflow Execution Layer for accounting firms."	[site]
Founded, team, funding	Founded 2021 [site] [yc]. Team 50 [yc]. YC says it raised $18M [yc]. Series A led by Altos Ventures, amount not stated; own post dated 16 Apr 2025, a local news story dated 12 Aug 2025 (dates differ, not resolved) [blog] [press]. Seed reported in 2023 (investors named by a profile site, amount an estimate) [press]. An old job ad says "over $4.5M" raised [jobs].	[yc] [blog] [press] [jobs]
Founders	Enoch Ko, CEO, sole founder listed. A software engineer who first tried an "AI-first accounting firm", found it brought the same long hours, then built the product [site]. YC likely backed a technical founder who had run the pain himself (inference). Names a Tempe, AZ base [yc].	[site] [yc]
Customers	They claim 350+ firms [site]. Series A post said 300+ [blog]. Named testimonials and stories: Green Isle Partners, Schmerer Corbridge CPAs, Aiola CPA, Necelis CPA, DHA CPAs, The Business Tax Group, HighPoint CPAs, Zuazo & Associates, Lee & Crowley [site]. Firm stories show 500 to 2,000 returns a year each [site].	[site] [blog]
Price	Not on the site (no Pricing link) [site]. Third parties give $25 a return for Collect, $20 for Deliver, $40 for both, 50-return minimum (one review is by a rival, Taxhance) [press]. A search summary of newer pages says credits per client per season, packages from 150 to 250 returns and a one-time setup fee of $1,000 to $4,500 (UNVERIFIED: I could not open their pricing page, it returned 404).	[site] [press]
Path	Started as client data intake for accountants (YC page) [yc]. Series A post (Apr 2025) calls it an "Intelligent Admin Copilot" [blog]. Rebrand 3 Dec 2025: "Workflow is table stakes. Capacity is the future." [blog]. Jun 2026: four products named Collect, Prepare, Connect, Deliver, with Prepare in limited beta that summer [blog].	[yc] [blog]

2. Their proposition, reverse engineered

Part	What they do
Who buys	Small and mid-size CPA firms in the US. Named firms do roughly 500 to 2,000 returns a year [site]. The partner or firm owner signs (inference: testimonials and stories are from firm owners).
The pain	Firms "can't bill" 40 to 60% of the workweek, they claim. The CEO's own words: "The industry is hitting a wall." [blog] The story: chasing documents, renaming files, following up [site].
The promise	More returns with the same headcount: "without adding headcount" [site].
How it works	Four steps. Collect: client intake and auto-sort. Prepare: leadsheets and workpapers linked to source documents, with an AI review. Connect: pushes fields into UltraTax, Lacerte, Drake and CCH Axcess. Deliver: return packet, IRS 8879 e-signature and payment in one link [site].
The output	A review-ready file, data already in the tax software, a signed and paid return [site].
Proof style	They count minutes and adoption: "30 min → 3 min" per return delivered, 89% client adoption in year 1, 99.9 to 100% extraction accuracy, +30% returns per season at one firm [site]. The capacity counters on the home page showed "0" when I read it [site].
Who sells it	Demo and sales-led. "Request a Demo" and a weekly live demo; also sells through webinars and gated guides [site] [blog]. Integrations with Karbon and Zapier [site].
Price model	Per return (older, third-party figures) [press]. Newer model unverified, see Fact file.
The story	Enemy: workflow tools that only track work. Belief: "Other systems track work, maybe move it around for you. Soraban does it." [blog] The AI keeps the accountant in charge: "The AI handles the prep work, the accountant makes the decisions." [blog]

3. Positioning

Category they claim: "Tax Workflow Execution Layer" and "capacity-first" tax workflow [site] [blog]. YC line: "AI tax workflow platform for accounting firms" [yc].
Category a YC partner would put them in: vertical AI for accounting firms, tax-season back office (inference).
What they want to be known for: doing the admin work between the client and the tax software, so a firm can take more returns without hiring.
Where they sit on the map: layers L1/L2/L3 = 1/2/2, buyer = sme, business.

4. Three layers vs ours

Layer	Them (score + evidence)	Us (Marton's logs show)
L1 Company brain	1. Says the platform "learns your firm" and carries context between handoffs [site]. No public sign of a lasting memory that other work reads. Job ad lists embeddings and Pinecone work, old and undated [jobs] (inference: planned or early).	A vault of about 1,400 pages, 112 call transcripts and about 8,700 customer quotes, searchable by meaning.
L2 Agent orchestration	2. A four-stage pipeline with human approval and a handover to the tax software and then to the client [site]. Called "AI data entry agent" and "AI intake organizer" in their posts [blog]. No claim of a team of routed agents.	About 5,600 jobs run by agents from three makers, each checked by a different AI, then automatic tests.
L3 Execution and learning	2. Does the work and checks it: an AI review flags differences and missing items [site]. Measures time per return and extraction accuracy [site]. No public claim that results change the next run. A rival review says Connect had a 3% error rate in beta (rival source, treat with care) [press].	About 160 failure records, each adding an automatic check. Offer feedback lines fell from 44 to 18 as rules were added. Only about 15% of jobs reached the main code; merges were unreliable in October.

5. Same and different

Topic	Same or different	Them	Us
Core belief	Same	The tool should do the work, not track it [blog].	Same (brief: we do the work and own the number).
Buyer	Same type, different country	US accounting firms, 500 to 2,000 returns [site].	W5 targets one type of firm, for example accountancy. UK is the example in the brief. No customer yet.
Kind of work	Different	US individual tax returns (1040) in a short season. A rival says no business returns (rival source) [press].	Payroll, VAT returns and forecasts above Xero (brief example). Not built.
Layer	Different	One workflow, intake to signed return [site].	Company memory, agent team and learning on one workflow (brief).
Proof	Different	Minutes saved, adoption, accuracy [site].	No customer number yet.
Delivery	Different	Software with a setup fee (unverified) and demos [site].	The system plus the two founders.
Human in the loop	Same	Preparer keeps control, can dismiss AI items [blog].	Marton's logs show approvals and review by a second AI.
Focus	Different	"100% focused on accounting" [site].	Not set: seven open wedges (brief).
Size of company	Same direction	Seed to Series A, $18M [yc].	Pre-seed, applying to YC.

6. Wedge overlap

Wedge	Fit	Why
W5 One layer for one type of small firm	direct	This is the same shape: one workflow layer, sold to many firms of one type, accountancy [yc] [site]. The buyer type matches. The country and the work do not (US tax vs UK payroll and VAT in the brief).
W4 Fit call to proposal in about 10 minutes	adjacent	Their Deliver step packages the return, signature and payment in one link [site]. Same "finish the paperwork fast" idea, for a different paper.

7. Whitespace (where they stop)

Layer gap: no public sign of a lasting company memory or of results changing the next run (L1 1, L3 not at 3) [site] [blog]. To own it we need a rule-change log tied to checked runs, with rework falling on a real workflow.
Buyer gap: US individual tax only; the site and integrations are US (IRS forms, US tax software) [site]. A rival review says no business returns (rival source) [press]. To own it we need UK firms, which means HMRC and Xero rather than IRS and Lacerte, and named UK people with the pain.
Number gap: they prove minutes and adoption, not money or margin; the capacity counters on the home page showed "0" [site]. To own it we need one before/after number a firm owner would sign, such as hours of unbillable work removed.
Price gap: not true as far as I can see. Per-return pricing with a small minimum (third party) is reachable for small firms [press].
Speed gap: not true. They claim onboarding in about a week [blog].

8. For the YC application

Proven today: Marton's logs show his system already runs agents that do and check work (about 5,600 jobs, each checked by a different AI) and turns every failure into a check (about 160 records). Soraban's public results are minutes saved and client adoption on one US workflow [site]. We have no accounting-firm customer or workflow run.
Draft answer to "how are you different from Soraban?": DRAFT. "Soraban does US tax-season admin for accounting firms. We would build the working layer for UK firms on payroll, VAT and forecasts, with a memory that learns from checked results." Founders say it their own way.
Future claim (needs proof): "One UK accountancy workflow, run by our agents for 4+ weeks, with a before/after number." Needs 3 to 5 UK firms who confirm the pain and one live workflow with a measured result.
What we understand that they don't: a layer that learns from each checked run gets better for every firm of the same type. 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 are 100% accounting [site], so they would want more workflows, and the Series A money is for "more workflows" [blog]. But UK payroll, VAT and Xero need new integrations and new tax rules, and a US-only product has no UK sales motion (inference). Not hard: no data or loop barrier stops them, so M3 can not score above 3 against Soraban alone.
Why now (M4): None found in this teardown. Their Series A and rebrand are their changes, not ones that help us more than them.
Do not claim: that nobody has built a layer for accounting firms (they have, 350+ firms they claim). That our system learns for firms (no customer run). That UK firms want it (the brief names no UK accountancy firm as a lead for W5). Any money result.

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

9. Ideas worth taking

Sell one product first and let firms add more: "start with all products or just one" [site]. A low-risk first step for a small buyer.
Keep the human in charge: AI items can be dismissed and edits are never overwritten [blog]. It is how they got accountants to trust it.
Publish proof as a small count per unit: minutes per return, adoption rate [site]. Use the same shape for our first number.

10. Unknowns

Real current prices and whether the setup fee exists. Look for a buyer's quote or their pricing page (it returned 404 on 2026-10-19).
Exact customer count (350+ on the site; 300+ in 2025 posts; one search summary said 500+, not confirmed).
Series A size and date (own post says 16 Apr 2025, a news story 12 Aug 2025).
Whether the AI learns from reviewer corrections (Prepare page does not say).
Whether they plan a UK or business-return product. Watch the blog and careers page after the next tax season.
Founder count: only one founder is named; check the YC page again for others.

Sources

soraban.com [site]
About Soraban [site]
Firm stories [site]
YC company page [yc]
YC jobs page [jobs]
Older AI Engineer/Research job ad, undated, "6+ months ago" [jobs]
We closed Series A, 16 Apr 2025 [blog]
Announcing Our Rebrand, 3 Dec 2025 [blog]
Prepare, the AI Workpaper Management System, Jun 2026 [blog]
Soraban blog index [blog]
Greater Phoenix In Business: Series A, 12 Aug 2025 [press]
Taxhance review of Soraban, 2026, a rival's page [press]

Soraban

What they do

How close to us

Wedges

Who buys

Full teardown