8. If I Were You: The Next 90 Days
The single most important thing
Get 10 accountants paying for reconstruction cases before you build the full product. At ~15 hrs/week that's the day-90 target. Full-time, aim for 20.
- Buyer: US practitioners who rebuild history from PDFs: tax resolution (unfiled years), forensic and litigation support, divorce financial analysis, estate accounting, and catch-up bookkeepers.
- Job: "Turn this case's statements into one balance-verified workbook, with every row traced to its page."
- How: start with a concierge service. You run your own scripts plus a model, check every statement's balance by hand, and deliver the workbook. Charge from the first case ($99–299). Automate only what you've done by hand at least 10 times.
Hours: the plan below assumes ~15 hrs/week. Full-time, compress weeks 1–12 into about 8 weeks and double the call and page targets.
Week-by-week
| Weeks | Goal | Hours split (~15/wk) | Exit criteria |
|---|---|---|---|
| 1–2 | Discovery + name check. 15 calls with target practitioners. Ask about their last 3 reconstruction jobs: how many PDFs, how many hours, which tool (DocuClipper? Valid8? manual? ChatGPT?), what went wrong, what they billed. Run the trademark and domain check for "StatementKit". | 10 hrs calls, 5 hrs research and name | The top 3 pains written in their words, with hours and dollars attached. One design partner agreed (paid $300–500 a month or advisor equity). |
| 3–4 | Corpus + concierge. Collect 100+ real statements (redacted, with consent) across 25+ institutions. Build a rough pipeline: text-layer parser, model fallback, balance check. Deliver 3–5 paid concierge cases. | 8 hrs pipeline, 7 hrs cases | 3+ paid cases (5 full-time). If fewer than 2 will pay, change the buyer once (litigation ↔ tax resolution), not the pitch. |
| 5–8 | Build the thin v1 (web only): case upload (ZIP or folder), auto-sort by account and period, extraction (top 30 layouts deterministic, model fallback), per-statement balance proof, gap and duplicate detection, row-to-page citations, Excel workbook export, free 3-page preview, Stripe billing (Solo, Firm, case pass), auto-delete. | 12 hrs build, 3 hrs cases | Concierge users move to self-serve. ≥95% first-run balance pass on the corpus (300 statements by week 8). |
| 9–10 | Search engine start. Publish 30 pages: 15 use-case pages (divorce discovery, bank deposits analysis, unfiled returns, estate accounting, catch-up cleanup) and 15 bank or credit-union pages, each with a redacted sample workbook and its balance check. Publish the security summary. | 8 hrs pages (templated from the corpus), 7 hrs fixes | 30 pages indexed. Security page live. First organic signups. |
| 11–12 | Measure and decide. Onboard every paying account personally. Survey hours saved per case. Fix the top 10 failed layouts. | 6 hrs onboarding, 9 hrs fixes and pages | 10+ paying subscriptions (20 full-time), ≥97% first-run balance pass, ≥60% of payers have run 2+ cases, median hours saved per case ≥5. |
Kill criteria (write these down now)
At day 90, the wedge is wrong if any of these is true:
- fewer than 5 paying subscriptions;
- paid month-1 retention below 50%;
- users say DocuClipper or QuickBooks is "good enough" in more than half of the calls.
Don't respond by adding document types. Respond by changing the buyer: Pivot A is catch-up bookkeepers only, and Pivot B is non-English statements for one market, after a competitor check.
The one thing to stop doing
Stop thinking of this as "PDF to Excel, but better".
That job belongs to QuickBooks (free, English, verified), to every LLM chat app, and to a founder with a 5-year search head start. "More accurate" isn't something a buyer can check, so it isn't something they'll pay for. Proof is something they can check. Every week spent on generic conversion quality, instead of the case workflow and the balance proof, is a week spent building the commodity.
A close second: don't add invoices, payslips or more countries in the first 90 days. Each one feels like a bigger market and is really a new set of competitors.
What stays on the long-term roadmap (only after proof)
- Brokerage statements parsed into sections (month 4–9).
- Cross-account transfer matching and a deposits summary (Firm plan).
- QuickBooks and Xero import files for the catch-up buyer.
- API for firms (the comparable's first big jump came from one API customer).
- A public "does this statement balance?" checker.
- Then, only if Pivot B tests well, one non-English market.