Result

# Mega Plan Review: B2B Accounting SaaS for Solo Founders

Hold up. Before I run the full 11-section plan review, I need to flag something. What you've given me isn't a plan. It's a one-line product pitch. A plan review machine pointed at a pitch produces vapor.

So let me do the thing that's actually useful here: pressure-test the premise, map the landscape, and tell you what a real plan would need to look like before any of the 11 sections can bite into it.

---

## Step 0A: Premise Challenge

**Is this the right problem to solve?**

Solo founders doing their own books. The pitch assumes:
1. Solo founders are doing manual bookkeeping
2. They hate it enough to pay for software
3. AI OCR + auto-categorization is the wedge

Let me push on each.

**(1) Are solo founders actually doing manual bookkeeping?** Most aren't. They're either:
- Using QuickBooks/Xero with bank feeds (auto-categorization already exists, has for a decade)
- Paying Bench, Pilot, or a local CPA $200-500/mo to do it for them
- Ignoring it until tax time, then dumping receipts on their accountant
- Using Wave (free) and tolerating it

The "manual bookkeeping" pain point is mostly solved at the data-entry layer. Bank feeds + Plaid pull 95% of transactions automatically. The remaining pain is **categorization decisions and tax-readiness**, not OCR.

**(2) What's the actual job-to-be-done?** When a solo founder hires Bench, they're not buying bookkeeping. They're buying "I don't have to think about this." OCR doesn't deliver that. A human (or an AI agent that operates like one) does.

**(3) "AI-powered OCR and auto-categorization" is table stakes, not a wedge.** QuickBooks has had receipt OCR since 2017. Expensify since forever. Ramp and Brex do it natively for free as part of their card products. Dext (formerly Receipt Bank) built a whole company on this and got acquired. The technology isn't differentiated. The technology is commoditized.

So the premise as stated is: "We're going to build a feature that already exists, for a customer who mostly doesn't have the problem you think they have, and charge them for it."

That's not a business. That's a graveyard.

---

## Landscape Check

**Layer 1 (tried and true):** QuickBooks Online ($30-200/mo, 7M+ customers), Xero ($15-80/mo), FreshBooks (solo-focused, $19-60/mo), Wave (free). All have OCR, all have categorization, all have bank feeds. Bench and Pilot occupy the "done-for-you" tier at $200-500/mo. This space is saturated and the incumbents are good enough for most users.

**Layer 2 (new and popular):** Digits ($150M+ raised, AI-native books for SMBs), Puzzle ($30M+ raised, real-time accounting for startups), Pry (acquired by Brex), Finally (AI bookkeeping), Campfire (AI-native ERP). All raising on the "AI replaces your bookkeeper" thesis. Some are 2-3 years ahead with real ML pipelines, real customer bases, real enterprise integrations.

**Layer 3 (first principles):** What does a solo founder actually need at month-end?
1. A P&L they trust enough to make decisions
2. Books clean enough that taxes don't blow up
3. Cash flow visibility for the next 90 days
4. Receipts/documentation organized for an audit they pray never happens

OCR is upstream of all four. It is plumbing. Nobody pays a premium for plumbing when the existing plumbing is free.

**Where the conventional wisdom might be wrong:** The interesting first-principles move isn't "better OCR." It's **"what if the books reconciled themselves and surfaced decisions, not data entry?"** That's an agent product, not an OCR product. Different architecture, different team, different price point, different moat.

---

## What the actual opportunities look like

If I squint hard at "B2B accounting SaaS for solo founders," there are real wedges. Not the one you proposed. These:

**1. The "fire your bookkeeper" agent.** A real autonomous agent that ingests your bank feed, card transactions, and emailed receipts, does the categorization, drafts the journal entries, flags anomalies, asks questions in Slack like a human bookkeeper would, and produces a clean trial balance monthly. Price: $99-199/mo. Customer: solo founders currently paying Bench $300/mo. Wedge: "Bench, but built by people who've shipped software in the last decade." Hard but doable. This is what Digits and Puzzle are chasing.

**2. The "tax-ready always" play.** Books are a means; taxes are the end. Build the product that promises: "On April 1, you push a button and your Schedule C / 1120-S is ready." This is what Keeper Tax did for freelancers ($20/mo, growing fast). Solo founders with an LLC have the same problem, harder. Underserved.

**3. The vertical play.** "QuickBooks for [specific industry]." E.g., bookkeeping for Shopify sellers, for SaaS founders with Stripe + AWS, for content creators with multiple platform payouts. The categorization problem becomes 10x easier when you know the vertical, and the unit economics improve because integrations are deeper.

**4. The "embedded in your bank" play.** Mercury, Relay, Rho all want to own this layer. If you're not a bank, you compete with banks giving accounting away free as a retention play. This is structurally hard unless you partner.

None of these are "B2B accounting SaaS with AI OCR for solo founders." That description fits about 40 companies, most of which will not survive the next 24 months.

---

## What a real plan would need before I can review it

I can run the 11-section plan review (architecture, error map, security, data flows, tests, performance, observability, deployment, trajectory, design, plus scope cuts) but I need an actual plan. At minimum:

1. **A specific user.** Not "solo founders." A specific persona with a specific accounting setup today, a specific pain, a specific willingness to pay. Names, ideally. Five customers you've talked to.

2. **A specific wedge.** Not "AI OCR." A specific job-to-be-done the incumbents do poorly, that you can do 10x better, that compounds into a moat.

3. **A specific architecture sketch.** What's the data model? Where does transaction data live? How do you handle the chart of accounts? What's the integration surface (Plaid? direct bank? card networks? Stripe?)? What does the ML pipeline look like? Where's the human-in-the-loop?

4. **A specific go-to-market.** How does the first customer find you? What's CAC? What's the expansion path?

5. **A specific competitive answer.** Why doesn't QuickBooks just ship this in their next release? Why doesn't Digits, with their head start and capital, beat you to it?

Without these, the plan review machine is reviewing fog.

---

## Recommendation

**Before any plan review, run `/office-hours`.** That skill is built for exactly this stage — premise pressure-testing, user definition, wedge identification, competitive landscape. It will produce a design doc that I can then review at full depth.

You'll come out the other side with one of three outcomes:
- A sharp, defensible wedge worth building (then we plan-review the hell out of it)
- A pivot to a different problem in the same space (likely outcomes 1, 2, or 3 above)
- A decision to not build this (which is also a win — bad work is worse than no work)

The honest read: as currently scoped, this idea is in the part of the market where good engineering won't save you. Distribution, vertical depth, or radically better UX is what wins. AI OCR is not.

What's your actual edge here? Why you, why now, why not QuickBooks? Tell me that and we can build something real.