StatementKit
As described:StopNarrowed:ContinueNo as a generic PDF-to-Excel converter. Go as a reconstruction tool for litigation, tax-resolution and catch-up work.
forensic & catch-up accountants · ~$73 a month · last reviewed (next review by 29 Dec 2026) · 33 cited sources
Start this week
- Tell me your hours per week available, your country (for US company and payment setup), and whether you know any forensic accountant, divorce financial analyst, EA or CPA doing tax resolution.
- Run the name check for "StatementKit" (USPTO, domain, existing products), and pick a fallback.
- Book 10 calls from that list and from LinkedIn searches. Ask about their last 3 reconstruction jobs: how many PDFs, how many hours, which tool, what went wrong.
- Build a 50-statement mini corpus from your own and friends' statements (with consent), and measure the balance-check pass rate of a plain LLM pipeline. That number is your baseline to beat.
Checkpoints and kill gates
Decide in advance what counts as on track. Missing a checkpoint means changing the buyer or the offer, not adding features.
Day 30
~15 hrs/week
3 paid concierge cases ($99+ each)
Full-time
5 paid concierge cases
If you miss it
Switch buyer once: from litigation to tax resolution, or the other way. Missing twice means stop.
Day 90
~15 hrs/week
10 paying subscriptions. ≥97% first-run balance pass on the corpus.
Full-time
20 paying subscriptions. ≥97% pass.
If you miss it
Cut down to one buyer and one output (the workbook). Drop brokerage and accounting imports until it passes.
Month 6
~15 hrs/week
$1k MRR, ≥70% month-3 retention, 150 search pages live
Full-time
$2.5k MRR, ≥70% month-3 retention
If you miss it
Pivot A: catch-up bookkeepers only (QuickBooks/Xero import focus). Pivot B: non-English statements for one Spanish-speaking market.
Month 12
~15 hrs/week
$3k MRR, churn ≤5% a month
Full-time
$7k MRR, churn ≤5%, 25+ Firm accounts
If you miss it
Stop adding features. Fix churn (annual plans, onboarding) and double down on the best search cluster.
Month 18
~15 hrs/week
$5k MRR
Full-time
$13k MRR
If you miss it
Put everything into the single best channel (search or community).
Month 24
~15 hrs/week
$8k MRR
Full-time
$20k MRR, profitable: target hit
If you miss it
Above $5k: keep running it profitably. It's a durable, low-operations business. Below $2k: stop or pivot.
Do
1.Build for reconstruction jobs, not "convert my statement". The unit is a case: 20–500 PDFs across several accounts and months. The case gets sorted, checked, merged and exported.
Target: Every user can say it in one line: "I drop in a box of statements and get one workbook I can defend."
Why: One statement is now a free job for QuickBooks (English only), ChatGPT and dozens of converters. A 300-PDF case with provable totals isn't.
2.Sell where bank feeds can't reach: opposing-party statements in divorce and litigation, deceased account holders, closed accounts, years older than the feed window, and clients who won't share bank logins
Target: ≥70% of paying accounts cite one of these jobs at signup
Why: Plaid gives up to 24 months of history, and only with the account holder's consent (verified). Open banking can't touch the opposing party's discovery PDFs.
3.Prove every statement balances before export. Opening balance + rows = closing balance, running-balance check per row, row count per page, and missing-month detection. Show a green or red badge per statement.
Target: ≥97% of statements pass on the first run by day 90, and ≥99% by month 12
Why: "A twelve-page statement can be 99% accurate and still be useless" if one row is missed (research/10). Proof, not an accuracy claim, is what an LLM chat can't give you.
4.Link every row to its source: file, page and position, clickable from the workbook
Target: 100% of exported rows carry a citation
Why: Forensic and litigation users have to defend every number. That's your sales pitch in one feature.
5.Use a hybrid engine: deterministic parsers for the top 30–50 layouts, and an LLM for the long tail with the balance check as a gate. Every fixed failure becomes a saved layout.
Target: Top 50 layouts parsed at ~$0 model cost. Model cost ≤5% of revenue.
Why: The comparable chose bank-specific parsing over one universal model, and spent 503 commits on accuracy in 7 months. Saved layouts compound. Prompts don't.
6.Handle brokerage statements properly: holdings, activity, dividends and realized gains as separate sheets
Target: Top 10 US brokerages parsed by month 9
Why: This is the document type generic converters handle worst (multi-section, not one table). Estate and divorce work needs it.
7.Price by value, not by page. Free: preview the first 3 pages of any statement, with the balance check shown. Solo $49/mo or $490/yr (1,500 pages). Firm $149/mo or $1,490/yr (6,000 pages, 5 seats, transfer matching, API). Case pass $99 (800 pages, 60 days).
Target: ≥50% of new Firm accounts on annual plans. Blended ARPU ≥$70.
Why: Per-page prices are collapsing: LLM cost is well under 1¢ a page. Firms bill case hours at $150–400 an hour, and a workbook that saves 6 hours is worth $49.
8.Run a concierge service first. Convert the first 10 cases with your own scripts plus manual checking, charged at $99–299 a case.
Target: 5 paying cases by day 30. 10 by day 60.
Why: It proves they'll pay, builds the test corpus, and shows you the workbook format they actually want.
9.Win long-tail search with real proof. Build bank-specific pages ("Navy Federal statement to Excel") and use-case pages ("reconstruct bank statements for a divorce case", "bank deposits analysis for unfiled returns"), each with a redacted sample output and its balance check.
Target: 150 pages by month 6 and 400 by month 18. ≥50% of signups from organic search by month 9.
Why: This was the comparable's main channel after paid ads failed. Use-case pages are less contested than "PDF to Excel".
10.Get into professional communities, not ad auctions. Speak at or write for forensic-accounting and divorce-financial-analyst groups, and tax-resolution associations.
Target: 3 talks or guest articles by month 9
Why: Small, tight communities with high trust. One respected practitioner's mention beats a month of ads.
11.Build a security posture firms can sign off on. Auto-delete files after 30 days (configurable), no model training on customer files, zero-retention model terms, a written security policy, MFA, and an audit log.
Target: A one-page security summary live by day 60. Zero incidents.
Why: Tax preparers must oversee service providers under US safeguards rules (not verified here; see analysis/09). Litigation users ask about chain of custody.
12.Keep operations near zero. Self-serve signup, in-app "report a bad row" button that files the PDF into the test corpus (with consent), and a 48-hour layout-fix promise for Firm plans only.
Target: Support ≤3 hrs a week at $5k MRR
Why: It has to fit your hours per week. The comparable runs on ~$500 a month of costs with no staff.
Avoid
1.Positioning as "PDF bank statement to Excel"
Why: It's the most contested phrase in the category, QuickBooks now does it natively in English, and ChatGPT does it for one file. You'd be a late clone.
2.Invoices and receipts
Why: Dext, Hubdoc, QuickBooks, Xero and every AP-automation tool own this. A different buyer, a different job and a price war.
3.Payslips as a launch document type
Why: Income verification is sold to lenders through enterprise deals. Add payslips only as a P2 for divorce income schedules.
4.India bank statements into Tally as the wedge
Why: Vyapar TaxOne (formerly Suvit) sells unlimited bank statement uploads, including scanned PDFs, into Tally for ₹10,000 a year, and claims 10,000+ practising firms (verified). That sets a ceiling of about $10 a month.
5.Competing on price per page
Why: The model cost is ~$0.003–0.007 a page. Someone will always be cheaper. Compete on proof, bulk handling and the case workflow.
6.Promising "99.9% accuracy"
Why: DocuClipper already claims it. It's unprovable per customer. Show the per-statement balance check instead.
7.A pure LLM pipeline with no balance gate
Why: One missed row in an expert report or a court filing ends your reputation in a small professional community.
8.Lender and underwriting APIs
Why: Enterprise sales, security questionnaires, and competitors like Ocrolus with large teams. Not reachable at your hours per week.
9.Litigation analytics beyond the ledger: fund tracing graphs, lifestyle analysis, expert-report writing
Why: Valid8 serves that end with a demo-led enterprise sale. You'd be building to avoid selling. Export clean data, and let the expert do the analysis.
10.Paid search ads before month 9
Why: The comparable spent about $3,900 for about $990 of revenue before stopping (verified). High-intent terms are bid up by funded competitors.
11.More than one country before $5k MRR
Why: Each country means new layouts, new accounting formats and new search pages. US English first.
12.Holding files forever, or training models on customer statements
Why: These are third parties' financial records. Retention is liability, not an asset.
13.Calling any output "audit-ready", "court-ready" or "tax advice"
Why: Say "balance-verified and source-linked". The professional signs off, not you.
14.Adding document types when a checkpoint fails
Why: A miss means the wrong buyer or the wrong channel. Change that, not the feature list.