Product Brief — Bank Statement Analyzer
First product under the product-led strategy (
../strategy.md). Status: BUILT and LIVE (https://obsrv.in ) since 2026-06-12. Geography: India-first (AA rails, NBFC/lending boom, CA firms), globally applicable. Backend system design:../server/ARCHITECTURE.md.
The one-liner
AI-native bank statement analyzer: upload (or pull via Account Aggregator) a bank statement and get clean transactions, categorization, decision-ready analysis, and an eligibility + approve / refer / decline recommendation (with an optional human-approval workflow) in seconds — self-serve, pay-per-statement.
Why now (the 2026 timing edge)
- LLM/vision models parse messy, multi-format statements far better and cheaper than the old OCR + rigid-template approach the incumbents were built on.
- Incumbents (Perfios, FinBox, Ocrolus) own enterprise lending — expensive, sales-led, template-based. They underserve smaller lenders/DSAs and accountants.
- Solo wedge = AI-native + self-serve for those underserved buyers.
Buyer — LOCKED: Small/mid lenders & DSAs (NBFCs, lending brokers/DSAs)
Primary buyer chosen 2026-06-08. Build the MVP, waitlist page, and messaging around them. Value = faster, better underwriting decisions. Pricing = pay-per-statement. Accountants/SMB remain documented as later expansion, not the first focus.
| Buyer | What they need | Willingness to pay | Notes |
|---|---|---|---|
| Small/mid NBFCs, lending DSAs/brokers (PRIMARY) | Fast underwriting signals: income, cash flow, bounces, obligations, risk summary | High (it speeds a money decision) | Clear ROI; per-statement pricing fits |
| CA / accounting / bookkeeping firms | Categorize, reconcile, export to Tally/Zoho/QB | Medium, but high volume | Strong alternate; less risk-analysis depth |
| SMB finance teams | Cash flow + spend categorization | Lower | Broader, fuzzier value |
| Personal finance (B2C) | — | Low | Avoid: crowded, low ARPU |
Feature menu (build a slice, not all of it)
Ingest: PDF (digital + scanned), CSV/Excel, password-protected PDFs, multi-bank formats, multi-month/multi-account aggregation, (India) Account Aggregator pull with consent. Extract: transactions (date/narration/debit/credit/balance), header data (holder, a/c, IFSC, bank, period, opening/closing), multi-page tables, file tamper/fraud detection. Categorize: auto-categories, counterparty ID, recurring detection, income vs expense. Analyze (lending): avg monthly balance, income detection & verification, net cash flow, bounced/returned payments (NACH/ECS), existing EMIs/obligations (FOIR), negative-balance days, circular/suspicious patterns, risk summary/score, auto underwriting memo. Analyze (accounting): P&L-style view, reconciliation helpers, Tally/Zoho/QB export, GST flags. AI-native (the differentiator): natural-language Q&A over statements, plain-English anomaly explanations, auto-written summaries. Deliver: dashboard, downloadable report (PDF/Excel/JSON), API, white-label. Trust/compliance: confidence scores + human-review flags, audit trail, strong data security, minimal/no raw-data retention. (Sensitive financial PII — treat as first-class.)
MVP (smallest sellable version — build this, defer the rest)
The core loop that delivers value end-to-end:
- Upload a PDF bank statement (start with the top 3–5 Indian banks’ formats + scanned support).
- Extract clean transactions + header data reliably, with confidence flags.
- Auto-categorize + detect recurring income/EMIs + bounces.
- One decision-ready output: a clean dashboard + downloadable report with the key lending signals (income, avg balance, cash flow, bounces, obligations) + a natural-language summary.
- Pay-per-statement (credits) or a small monthly plan.
Defer: Account Aggregator integration, API/white-label, accounting exports, every bank format, deep fraud forensics. Add based on what early users pull for.
The moat (since “AI analyzer” alone isn’t defensible)
- Breadth + reliability of bank-format coverage (compounding data asset).
- Trust: accuracy, confidence scoring, tamper detection — the things finance buyers actually gate on.
- Distribution: product-led self-serve + direct outreach to lenders/DSAs + SEO/AEO (see go-to-market/gtm-plan.md).
Risks / watch-items
- Accuracy is existential in finance — wrong numbers kill trust instantly. Confidence flags + human-review UX from day one.
- Data security / PII — minimize retention, encrypt, be explicit. A breach ends it.
- Employer conflict-of-interest — sanity-check that a bank statement analyzer (lending) is clear of Kapittx (AR/collections) obligations before going public.
- Incumbents — stay in the self-serve/AI-native lane they won’t chase.
Validation before heavy build
- Landing page + self-serve signup; drive lenders/DSAs and CA firms to it.
- 5–10 problem interviews with target buyers (small lenders / CAs): how do they analyze statements today, what’s painful, what would they pay per statement?
- A scrappy demo: analyze a real (anonymized) statement live → instant “wow” / proof.
Distribution
Product-led self-serve + direct outreach to lenders/DSAs + SEO/AEO + partnerships. Trust is
institutional (company, India data residency, the verification engineering), not a personal
brand. Full plan: ../go-to-market/gtm-plan.md.