AI · ACCOUNTING AI
Mid-market trade finance teams spend most of their working week on data plumbing. Accounting AI does the plumbing — categorizing transactions across your chart of accounts, reconciling bank statements to invoices as payments land, surfacing anomalies before month-end, answering natural-language queries over your live books. Built on the operational graph — it knows which payment came in for which invoice for which order with which client. Operational context is what turns generic accounting AI into one that actually understands trade.
LIVE CLOSE IN PROGRESS · OPERATIONAL GRAPH ATTACHED
Day 2 of an April close. Categorization is running. Bank wires are being reconciled to invoices as they land. Anomalies are flagged the moment a pattern breaks. The CFO types a question in plain English and gets the answer from live data — not a custom-report build, not a bookkeeper's reply tomorrow.
Same screen for daily reconciliation, quarter-end, year-end, and an FX revaluation run. Categorization, reconciliation, anomaly detection and NL queries all live on the same operational graph.
THE PROBLEM ACCOUNTING AI SOLVES
New transaction lands. The bookkeeper looks up the supplier, matches it to a PO, finds the chart-of-accounts entry, types the memo. 1,200 of these a month. 80 hours of skilled labor on work that nobody promotes and nobody insources — just somebody has to do it before close.
Bank statement arrives the 3rd of the next month. Finance spends week 1 reconciling. Issues from week 4 of the prior month surface too late — the supplier moved on, the customer disputes are stale, the trail is cold. The lag is the bug.
Want AR over $100K from EU customers with terms over Net 45? Build the report in QB / Xero (3 hours of filter chains) or ask the bookkeeper (1 day later). Neither is real-time. Neither is something a CFO can do mid-board-meeting.
FIVE NAMED FUNCTIONS
Accounting AI is a tool, not a bookkeeper. It handles the routine end-to-end, surfaces the unusual for an operator to look at, and posts back to your existing GL with a full audit chain attached.
Map every transaction to your chart of accounts using operational context — this payment came in against this invoice on this order, with this vendor, in this currency. Accuracy improves as operators correct edge cases.
Match bank statement entries to invoices, marketplace payouts to channel sales, supplier payments to POs. Cross-references bank API + Xero / QB + invoice records + Stripe webhooks + WhatsApp "PAID OK" confirmations.
Duplicate invoices, unusual FX gains, payments from new bank accounts, irregular vendor patterns, sudden DSO shifts. Flagged in real time, not at month-end. Internal use at EDMA Group has caught $50k–$200k a year in errors that would otherwise close as "reconciling differences".
"Margin trend by product category Q1–Q4." "Unpaid invoices > $100K from EU customers." "FX exposure on open orders." Real answers from live data, with the chart, the rows, the JE references, instantly — not a custom-report build, not a bookkeeper-reply tomorrow.
Month-end, quarter-end, year-end. Auto-generated closing entries with operator review. JE templates per scenario: FX revaluation, prepayment recognition, accrual reversals, deferred revenue. Close in 2–3 days, not week 1 of the next month.
WHY THIS IS DIFFERENT
Generic accounting AI sees the wire. Accounting AI sees the wire and the invoice and the order and the shipment and the WhatsApp confirmation — as one graph. Which is why it categorizes and reconciles with operational context instead of guesswork.
A wire is just an amount and a sender. The graph is what tells Accounting AI which invoice for which order with which client.
INTEGRATIONS
EDMA's Accounting AI runs on the operational graph; your existing accounting system stays the system of record for the GL. Best of both — nobody asks finance to abandon a ledger they trust.
Bi-directional sync. Transactions flow from EDMA to Xero with categorization complete and the chart-of-accounts entry already selected. Reconciliation status, payment matches, and JE entries sync back. Your bookkeeper sees the same numbers in either UI.
Bi-directional sync. Categorized transactions, reconciliation status, and journal entries posted directly. Multi-entity supported — EDMA pushes categorized data into each QB tenant per legal entity, with intercompany flagged separately.
API integration for enterprise customers. Custom mapping per tenant. Subsidiary structures, segment dimensions, and approval workflows preserved. Onboarded in a single engagement — a 4-week implementation, not a year.
Accounting AI is not asking you to switch GLs. It is asking you to stop typing transaction memos.
RECONCILIATION PATTERNS
Per-pattern auto-rate is shown alongside each match, not hidden. The unusual gets routed to operator review — the rest comes back with the suggested match and a confidence score the operator confirms.
Bank wire → invoice
Customer pays invoice in full. Match by amount + reference + customer graph.
autoBank wire → partial invoice
Customer pays 60% installment. Schedule resolved by milestone graph.
autoMarketplace payout → multi-invoice
Amazon settlement covers 12 orders. Fees + returns + chargebacks unbundled.
autoSupplier payment → PO
Operator pays supplier per terms. Match via PO graph + payment-schedule.
autoFX revaluation
EUR receivable, USD books, monthly mark. Rate source: ECB / XE / OANDA / custom.
rule-basedBank fees → ledger
SWIFT / ACH / wire fees. Auto-split by route. GL 6810 or per-tenant mapping.
autoRefunds → original invoice
Returns refunded to channel. Match via order graph + Stripe webhook chain.
autoChargebacks → invoice
Amazon FBA late-shipment penalty. Reconciled to shipment record + dispute path.
autoCustom patterns onboarded in one engagement. Operators publish auto-rates by pattern; nothing hidden.
ANOMALY DETECTION · THE UNSUNG HERO
Errors don't close as "reconciling differences" at year-end if they're caught in week 1. Anomaly detection runs on every transaction against the operational graph — supplier history, FX baselines, channel patterns, DSO drift — and flags before the trail goes cold.
Same milestone billed twice. Same PO referenced on two different invoices. Caught at $40K average per occurrence internally — the kind of error that closes silently if nobody is looking at the graph.
caught · $ 40K avg / occurrenceFX gain or loss outside the L12M baseline distribution. Caught a $180K revaluation error internally where the rate source had been silently re-pointed. Surfaced before the month was closed.
caught · $ 180K revaluationPayment from a bank account that has never been seen for a counterparty before. Caught a $200K wire mistakenly addressed to the wrong account — the operator's bookkeeper would not have spotted it without the graph.
caught · $ 200K mis-addressed wireSudden DSO shift on a top-20 customer. Irregular vendor patterns — a single supplier's payment frequency doubling, terms quietly relaxing. Surfaced in real time, not at month-end.
live signal · not month-endVS. ALTERNATIVES
Each row is something a mid-market trade finance team actually needs. Honest answer for each tool — including ours.
| Capability | Accounting AIEDMA · trade-native | Vic.aicategorization AI | Numericclose automation | Trullionrevenue recognition AI | Xero / QB native AIin-product assistants |
|---|---|---|---|---|---|
| Operational graph as context (orders, shipments, payments, communications) | ✓ | — | — | — | — |
| Trade-specific patterns (marketplace payouts, FX, LC, supplier payments) | ✓ | — | close only | rev-rec only | generic rules |
| Bi-directional Xero / QB / NetSuite integration | ✓ | push-only | read + JE | read + JE | ✓ |
| NL queries over live books | ✓ | — | close metrics | — | scoped Q&A |
| Anomaly detection across operational + financial chain | ✓ | categorization only | flux only | — | — |
| Categorization at scale | 98% · day 90 | ~ 90% claimed | — | — | rules + ML hints |
| Cross-system reconciliation (bank + GL + invoice + Stripe + comms) | ✓ | — | bank + GL | — | bank + GL |
| ZDR on every commercial AI provider | ✓ | enterprise tier | enterprise tier | enterprise tier | platform policy |
| Average month-end close time | 2–3 days | categorization step | ~ 5 days | rev-rec step | ~ 7–10 days |
FAQ
If you have a sixth, we'd rather answer it on a call than write it here.
Accuracy improves with operator feedback. Out of the box, Accounting AI uses the operational graph — payment, invoice, order, vendor — to propose categorisations and matches. The operator confirms or corrects, and the model learns the chart of accounts, the vendors, and the channel patterns. Per-pattern auto-rate is shown alongside each match, not hidden. The unusual gets routed to operator review with a suggested match and a confidence score — nothing is auto-posted without your rules.
Book a 30-minute demo. Bring last month's bank statement, a Stripe payout, a marketplace settlement — whatever your finance team is currently typing memos on. We'll connect it to your operational graph during the call and walk through the categorization, the reconciliation, the anomalies, and the NL queries.