How FynHelp's agents read documents, reconcile bank and books, apply Indian GST, TDS and payroll rules, and hand everything uncertain to a person. Every threshold here is the one the product runs on.
Document intake and extraction
Agentic AI
Data ingestionFormat detectionParser / AI readerConfidence gate
Every file that arrives (upload, Gmail or WhatsApp) is identified, read into structured lines and scored. Only lines the agent is unsure of go to a person.
LogicHuman-in-the-loop
Trust gateLines below 75% confidence go to Review
TimeSeconds per file; AI reads stop by 35 s
Bank ↔ ledger reconciliation
Reconciliation
Bank + booksExact & fuzzy matchException checkpointReconciled ledger
Matches every bank line against the books in five stages, from exact to part payments. Anything it cannot settle with certainty becomes an exception with a reason code.
LogicAutonomous agent
Trust gateFuzzy match ≥ 0.82 and clear of the runner-up by 0.03
People decide what the agent could not. Every decision needs a note, can be undone, and teaches the Recon agent to propose the same decision next month.
LogicHuman-in-the-loop
Trust gateNothing learned is applied without a person confirming
TimeOne click from any unmatched line
Tally day book and ledger import
ERP & Tally Integration
Tally exportVoucher detectionDirection from Vch TypeBooks side
Tally XML day books and CSV/Excel ledger exports are read as the books side, with Tally's debit/credit convention handled correctly whatever the file is called.
GST registers are summarised with their totals and checked under GST rules in code. They are kept out of bank reconciliation so nothing is counted twice.
LogicRule engine
Trust gateChecks run in code; failures are flagged, never auto-corrected
TimeUnder a second per file
TDS register and section rates
CA & Tax Compliance
TDS registerSection groupingRate checkSummary + due date
TDS registers are totalled by section from the TDS deducted column and every entry is checked against the rate for its section.
LogicRule engine
Trust gateRate checked per entry; 20% no-PAN rate accepted
TimeUnder a second per file
Payroll register: PF and ESI
CA & Tax Compliance
Payroll sheetTotalsPF / ESI checksSummary + due date
Payroll sheets are summarised (gross, PF, ESI, PT, TDS, net) and each employee's PF and ESI is checked. Salary payments themselves are reconciled from the bank.
Gateways pay the bank net of fees. With the settlement report uploaded, each payout is grossed up in the MIS: revenue at what customers paid, fees and GST on fees as expenses.
LogicRule engine
Trust gatePayout must match net within ₹1 and 5 days
TimePart of MIS generation
MIS generation and partner sign-off
Agentic AI
Matched linesFigures in codeCited AI insightsPartner sign-off
The Narrate agent writes the monthly MIS from matched transactions only. Figures are computed in code; AI insights must cite real transaction ids; a partner signs off.
LogicAgent + partner sign-off
Trust gateEvery insight must cite transaction ids and quote given figures
Time~14 s per MIS (measured)
Document chaser
Agentic AI
Missing sideChase createdFollow-upsAuto-close on arrival
When a month is missing a statement or the books, the Chaser follows up with the client by email or WhatsApp and closes itself when the document arrives.
LogicAutonomous agent
Trust gateCloses only when a matching document arrives
TimeFollow-ups on the firm's chase day
Timings marked measured come from FynHelp's own retest packs. Rules follow Indian GST, TDS (Income-tax Act, 2025, read with the familiar 1961 section codes) and EPF/ESI practice; checks flag issues for a person and never change your data.