2. Market Reality Check
Prices were checked on 30 Sept 2026 where marked "verified" (see 10). Everything else is an analyst estimate. Re-check prices before quoting them externally.
2.1 Who exactly is the buyer?
"Accountants and bookkeepers" isn't one buyer. They split by why they have a PDF instead of a bank feed, and that decides whether they'll pay.
| Segment | Why they have PDFs | Volume per job | Willingness to pay | Pain (1–10) | Verdict |
|---|---|---|---|---|---|
| Ongoing monthly bookkeeping (current clients, current periods) | Rarely. Bank feeds cover them. | 1–5 statements a month | Low. QuickBooks now does it natively in English. | 3 | Avoid as the core buyer. The feeds and the platforms own this. |
| Catch-up / cleanup bookkeeping (new client, 1–3 years behind) | Periods older than the feed window, closed accounts, client won't share logins | 30–150 statements | $30–100 a month, or per job | 6 | Secondary buyer. It's the "between cases" use that keeps accounts active. |
| Tax resolution (unfiled returns, IRS or state audits, bank deposits analysis) | Years-old records; the IRS itself works from statements | 50–300 statements | $50–150 a month. The fees are billed to the client. | 7–8 | Core buyer. |
| Forensic accounting and litigation support (fraud, disputes, bankruptcy) | Discovery and subpoena productions arrive as PDFs, often scanned | 100–2,000+ statements | $100–500+ a month, or per case | 8–9 | Core buyer, but check who else already sells to them (Valid8, see 2.2). |
| Divorce financial analysts and family-law support | The other spouse's accounts come through discovery. No feed is possible. | 50–500 statements | $50–150 a month, or per case | 8 | Core buyer. Episodic, so annual plans and case passes are essential. |
| Estate and trust accountants (fiduciary accountings) | The account holder has died, so there's no consent and no feed | 50–300 statements, including brokerage | $50–150 a month | 7 | Core buyer once brokerage parsing works. |
| Lenders and underwriters (income and cash-flow verification) | Applicants upload PDFs | Thousands a month | High, but enterprise | 7 | Avoid. Enterprise sales and funded incumbents. |
| Individuals ("my bank only gives PDFs") | Personal budgeting | 1–12 statements | ~$0. They'll use ChatGPT. | 3 | Free funnel only. |
The pattern: willingness to pay rises with (a) the number of statements per job and (b) how much each number has to be defended. Those are exactly where a chat-based LLM and a single-file converter are weakest.
Pain score for the product as described: 5/10. For reconstruction work it's 8/10. A tax-resolution or divorce job with 200 statements is 10–30 hours of manual keying or checking, at billing rates of $150–400 an hour (analyst estimate).
2.2 Competitors and substitutes
Free and native substitutes (your biggest threat)
| Substitute | What it covers | Why a customer would or wouldn't use you instead |
|---|---|---|
| QuickBooks Online native PDF import (verified, page updated 24 Aug 2026) | Upload a PDF or image statement, "our AI extracts the info", review, then create transactions. English-language statements only. Up to 1,000 lines per upload. | Won't switch for one statement going into QuickBooks. Will switch for 200 statements across 6 accounts, non-QuickBooks work, litigation (no ledger at all), or when they need proof and citations. |
| General LLM chat (ChatGPT, Claude, Gemini) | Reads a statement and returns a table. Gemini counts a PDF page as 258 tokens and takes up to 1,000 pages (verified). | Fine for 1–5 statements. Weak at bulk, no balance proof, silent row drops, and confidentiality worries with client data. |
| Bank feeds / aggregators (Plaid, QuickBooks/Xero feeds, UK open banking) | Current and recent transactions, with the account holder's consent. Plaid returns up to 24 months (verified). | Can't reach opposing parties, deceased holders, closed accounts, or anything older than the feed window. See 2.3. |
| Xero | Imports CSV/OFX statement files. Native AI PDF import couldn't be verified in this run. | If Xero ships PDF import like QuickBooks did, the catch-up segment shrinks further. |
Specialist converters and document AI
| Product | Price (checked) | Positioning | Threat to the wedge |
|---|---|---|---|
| Bank Statement Converter (solo founder) | $30 a month for 400 pages, and higher tiers (as reported Feb 2024; the pricing page returned 404 on 30 Sept 2026) | "1000s of banks world wide", thousands of financial, accounting and legal firms, API available | High on search for bank-specific terms. It owns the "[bank] statement to Excel" long tail. Weak on case workflow and proof (as far as its public pages show). |
| DocuClipper (verified) | Starter $20/mo (60 pages), Business $111/mo (640 pages, cash flow analysis, fraud detection), Enterprise $360/mo (2,000 pages). 30% off annually. | "10,000+ finance teams", "99.9% field-level accuracy", SOC 2 Type II, exports to Excel, CSV, QBO, IIF, OFX, QFX and QIF, covers bank, credit card and brokerage statements | The most direct competitor. Already markets to accounting and lending firms and claims brokerage support. Your edge has to be proof, case workflow and price per page for bulk. |
| Dext (snippet-verified) | Business from $31.50/mo. Bank statement extraction add-on about $0.32 per document. | Receipts and invoices first, with bank statements as an add-on | Low for the wedge. It's a bookkeeping pre-processing tool. |
| ProperSoft / ProperConvert (snippet-verified) | From $19.99/mo. MoneyThumb 2qbo Convert Pro is $599.95–699.95 one-time. | Desktop and web converters into QBO, IIF and QBJ | Low. Aimed at QuickBooks users, and QuickBooks now does it natively. |
| Nanonets (third-party figures) | Credit-based, $0.02–0.30 per workflow block run. Third parties quote ~$499/mo for about 5,000 pages. | Enterprise document-AI platform | Low. It's a build-your-own workflow platform, not a finished workbook. |
| Parseur, Docparser, Airparser, Affinda, ImageToTable (third-party list, Sept 2026) | ~$33–67 a month, some with free tiers | Generic extraction: templates, email-to-sheet, GPT-based schemas | Low to medium. Generic, with no case workflow. |
| Valid8 (verified: customers and features; no public price) | Demo-led, pricing not published | Forensic: extract from 20,000+ institutions, verify every entry "against source evidence", reconcile, trace funds across accounts. Customers include accounting firms, law firms, government and Medicaid agencies. | High at the top of the forensic market. It already does verification and tracing. Your space is below it: self-serve, priced per month, for solo practitioners and small firms who won't book an enterprise demo. |
| Vyapar TaxOne (formerly Suvit, India; verified) | ₹10,000 a year (list ₹20,000), unlimited bank statement uploads including scanned PDF, into Tally | 10,000+ practising firms, 30,000+ accountants | Kills the India-into-Tally wedge on price. |
| Lender platforms (Ocrolus and similar; not checked) | Enterprise | Income and cash-flow verification for lenders | Not your buyer. |
Why a reconstruction buyer would switch to you:
- one case folder in, one balance-proven, page-cited workbook out;
- priced for hundreds of pages, not 60;
- no enterprise demo needed.
Why they wouldn't:
- DocuClipper is already good enough for them;
- Valid8 is already in their firm;
- they don't trust a new vendor with discovery documents.
The honest read: the gap is narrower than it looks, and it's won on execution, not on a feature nobody else has.
2.3 Do bank feeds and open banking reduce demand?
Yes, for current-period bookkeeping. No, for reconstruction.
| Force | Evidence | Effect on demand |
|---|---|---|
| Aggregator history limits | Plaid returns up to 730 days (about 24 months) on the first pull. History grows only while the connection stays live (verified). | Anything older than ~2 years, or from before the connection, still needs PDFs. |
| Consent | Feeds need the account holder to connect. | Opposing parties in litigation, deceased holders and uncooperative clients never connect. That's the wedge. |
| UK open banking scale | 1 billion payments and 100 billion API calls by 30 July 2026. 88 account providers and 238 third-party providers (verified). | Mature feeds shrink the UK catch-up market. It's one more reason not to lead with the UK. |
| US open-banking rule (CFPB 1033) | Final rule 22 Oct 2024. Reconsideration started with an advance notice on 22 Aug 2025 (verified). The 2026 status wasn't verified. | A mandatory US data-access regime is delayed and uncertain, so PDFs stay common in the US for longer. |
| The comparable's revenue through 2025 | ~$30k a month in late 2024, rising to $42,175 in Aug 2025 (snippet-verified). | Demand for PDF conversion was still growing in 2025, despite open banking and LLMs. There's no data after Aug 2025. |
| QuickBooks native PDF import (2026) | English statements, 1,000 lines per upload (verified) | Cuts the bottom of the market: single-statement QuickBooks catch-up. That was probably a large share of the generic converters' volume. |
2.4 LLM commoditisation: how a solo founder differentiates in 2026
The threat is real. A page costs Gemini 258 input tokens. At $0.25 per 1M input tokens and $1.50 per 1M output tokens (Gemini 3.1 Flash-Lite, verified), a page costs about $0.002–0.004 to extract, before retries. The market's list prices are $0.03–0.33 a page (DocuClipper Starter: $20 for 60 pages; Dext: ~$0.32 a document). That spread won't last. Anyone selling extraction by the page is selling a falling price.
What an LLM (or a chat window) doesn't do, and what you sell instead:
| Moat | What it means in the product | Why a general model or a clone can't copy it cheaply |
|---|---|---|
| 1. Proof | Opening balance + rows = closing balance, running-balance check per row, row count per page, missing-month detection. Every statement gets a pass or fail badge. | Models don't check themselves. As one 2026 review put it, "a twelve-page statement can be 99% accurate and still be useless". Proof needs deterministic code around the model. |
| 2. Provenance | Every row links to file, page and position | Needed in litigation and tax resolution. Chat tools don't give it. |
| 3. Bulk case workflow | Drop 300 PDFs, auto-sort by account and period, detect duplicates and gaps, match transfers between accounts, merge into one workbook | Chat windows are one conversation, one file. Converters are one file, one table. |
| 4. Layout memory | Deterministic parsers for the top layouts (near-zero cost, exact), LLM for the long tail, and every fixed failure saved as a layout | The comparable's playbook: bank-specific parsing and 503 accuracy commits in 7 months. It compounds with every customer. |
| 5. Search position on use-case terms | Pages for "[bank] statement to Excel" and "reconstruct statements for divorce / unfiled returns / estate accounting" | The generic terms are taken. The use-case terms are smaller and less contested (volumes not verified). |
| 6. Trust posture | Auto-delete, no training on files, zero-retention model terms, an audit log | Professionals handling third parties' financial records can't paste them into a consumer chat app. |
What won't differentiate: accuracy percentages, "AI-powered", more output formats, a lower price per page.
2.5 Vitamin or painkiller?
The generic converter is a cheap painkiller that has become a free one. The reconstruction tool is still a painkiller:
- "I have 240 PDF statements from the other spouse's 5 accounts, received in discovery, due in 3 weeks. I bill $250 an hour, and my client doesn't want to pay 25 hours of data entry."
- "New tax-resolution client, 4 unfiled years. I need a bank deposits analysis by Friday."
Test: if StatementKit disappeared tomorrow, what would the user lose? For the generic converter, nothing: they'd use QuickBooks or ChatGPT. For the reconstruction tool, 10–30 billable hours per case and a workbook they can defend.
2.6 Market size: venture-scale, lifestyle, or a feature?
- The generic job is a feature. QuickBooks has already absorbed it.
- The reconstruction job is a lifestyle-to-small business. A loose bottom-up estimate (not verified):
- US firms and practitioners who do reconstruction work at least a few times a year (tax resolution, forensic and litigation support, divorce financial analysis, estate accounting, catch-up bookkeeping): tens of thousands, perhaps 30,000–60,000;
- realistic spend with one tool: $600–1,800 a year;
- SAM ≈ $20–100M a year. 275 accounts ($20k MRR) is 0.5–1% of the addressable firms. That's achievable in principle, but only with a strong search position.
- The proven ceiling for a solo operator is the comparable: ~$320k in 2024, and roughly $40k a month in mid-2025, from a broader, earlier position.
- Venture-scale? No. The venture-scale version is an underwriting or forensic platform with enterprise sales (Ocrolus, Valid8). It's a different business.
2.7 Geography
- US first. It has the most PDF-dependent reconstruction work, an uncertain open-banking mandate, USD pricing, and thousands of banks and credit unions with their own layouts (the long tail that general tools get wrong).
- Canada, Australia and South Africa second. They're English-speaking, but the QuickBooks native import also covers them, so only for the reconstruction buyer.
- UK later. Open banking is mature there, so the catch-up market is smaller.
- India: no. It's priced out by Vyapar TaxOne.
- Non-English markets: only as Pivot B, after a verified competitor check.