Use case · Documents and finance
Bank statements downloaded and reconciled against the ERP before anyone logs in
Where the bank has an API or a CSV export, the workflow fetches transactions directly. Where it only has a portal, a bot logs in (a person approves the MFA prompt from their phone), downloads the day's statement and hands it to IntelliDocs. An agent then proposes a match for every transaction — invoice, customer, amount, tolerance — and writes the confident ones to the ERP as matched; the rest arrive as a reviewed list with the agent's reasoning.
- Who it is for
- Finance teams and accounting firms reconciling several bank accounts a day against open items in an ERP or accounting system — especially in Latin America, where many bank portals have no API and statements arrive as PDF.
- Systems involved
- Bank portals (Studio bot)IntelliDocs / CSV parsingERP or accounting API (NetSuite, SAP B1, Siigo, Alegra, Omie)Teams / SlackExcel 365
How the run goes
5 steps · logged per run
- Trigger01
Every morning 06:00
Per bank account
- Bot02
Download statement
Portal login; MFA approved by a person
- Document03
Statement to transactions
PDF via IntelliDocs, or CSV/API
- Agent04
Propose matches
Open items, tolerances, references
- Person05
Confirm exceptions
Accountant reviews the unmatched
How it runs, step by step
- 1
A schedule starts one run per bank account. For banks with an API or Open Banking access, a workflow step fetches the transactions; for the rest, a Studio bot logs into the portal on a Bot Runner. When the portal asks for a one-time code, the run pauses and sends an approval to the account owner's phone; the code is entered and the run continues.
- 2
PDF statements go through IntelliDocs and come back as a list of transactions with date, description, reference, debit and credit. CSV and API results are normalized to the same shape.
- 3
An agent step reads the ERP's open items (receivables and payables) and proposes a match for each transaction, with a reason: exact reference, amount within tolerance, customer name in the description, partial payment. Confident matches are written to the ERP through its API as cleared; anything below the threshold is left open.
- 4
The accountant gets one list: the unmatched transactions with the agent's best guess and the reasoning. They confirm, correct or leave open, and the answers are written back.
- 5
At month-end the run log is the reconciliation audit trail: who matched what, when, and why.
Limits — what this does not do
- Bank portal terms of service vary; we check them before building a portal bot and prefer APIs where they exist.
- MFA needs a person; the run pauses rather than bypasses it.
- Match proposals are suggestions with reasons, not accounting decisions; the confidence threshold is yours to set.
Questions about this process
No. The run pauses and asks the account owner for the code through an approval; this is the human-in-the-loop step, by design.
Any with an API for clearing or matching documents — NetSuite, SAP Business One (Service Layer), Siigo, Alegra, Omie, Xero and QuickBooks Online among them. Desktop systems use a bot for the write-back.
The deterministic bot stops at the changed screen and reports; the fix is a selector change in Studio, usually the same day.
Related use cases
Build this on your systems in a guided sandbox.
Tell us the process; after a 30-minute call the sandbox is ready from 24 hours, and we build the first run with your team.