Bank, marketplace, and PO reconciliation — automated, not just alerted
UTR and reference-number matching that survives partial payments, split settlements, and marketplace deductions. Confidence-scored per line, auto-clear above your threshold, human review below it.
AI automated reconciliation matches bank statements against Sales Invoices and Payment Entries, marketplace settlement reports (Shopify, Amazon, Flipkart) against ERP revenue, and Zoho Inventory Purchase Orders against Books Bills — with UTR and reference-number matching that survives partial payments and split settlements. The reconciliation runs continuously on your TallyPrime, Zoho, ERPNext, or custom-ERP data and produces a match register with confidence scores per line. Auto-clear happens only above the threshold you set; low-confidence matches queue for a human reviewer. Recovers the hours finance teams still lose to spreadsheet reconciliation and catches the settlement drift that grows into meaningful revenue leakage.
Last updated 2026-07-02 · Part of the AI Copilot for ERPs early access
What this capability does
Bank statements against Sales Invoices, marketplace settlements against sales, purchase orders against bills — reconciliation is where finance teams still lose the most hours. The mechanical work is easy to describe (match this credit to that invoice by UTR) and hard to execute at scale because real payments arrive in the messy shape of business — one UTR against three invoices, one invoice paid in two tranches, marketplace settlements netted after commissions and shipping deductions, purchase orders received as partial deliveries against variable bills. AI automated reconciliation absorbs that mess. The copilot ingests bank statements through your existing banking connector (HDFC, ICICI, or the Perimattic banking API layer), marketplace settlement reports through the Shopify, Amazon India, or Flipkart connectors, and Purchase Orders against Bills from Zoho Inventory or ERPNext. Every candidate match is scored on UTR similarity, reference number match, party match, and amount match with tolerance for split and partial cases. Above the confidence threshold you set, matches auto-clear and post the reconciliation to the ERP. Below the threshold, matches queue for a human reviewer with the top three candidate invoices ranked. Reconciliation runs continuously, so the "reconcile at month-end" cycle becomes "reconcile continuously and close the books faster" — which is the real business outcome finance leads want.
How it works
End to end, from question to grounded action.
Multi-source ingestion
Bank statements from your banking connector or uploaded MT940 / CAMT.053. Marketplace settlements from Shopify, Amazon India, Flipkart connectors. Purchase Orders and Bills from Zoho Inventory / Books or ERPNext. Custom-ERP sources come through the mapped ingestion.
Candidate match generation
For every incoming credit or debit the copilot generates candidate matches from open invoices, POs, or settlements. Matching signals include UTR, reference number, party name (normalised), amount within tolerance for split or partial payments, and date proximity within your business's observed lag distribution.
Confidence scoring
Each candidate match is scored on the combined signal strength. A match with UTR + reference number + amount within 1% + party match scores very high; a match with only amount and date proximity scores low. The threshold at which auto-clear happens is configurable per source and per party category.
Auto-clear or human review
Above the threshold, the copilot posts the reconciliation to the ERP with an audit-trail note. Below the threshold, the match queues in the ERPPlugs dashboard with the top three candidates ranked. A reviewer approves, edits, or dismisses each queued match. Every action is logged.
How this shows up on each ERP
Same capability, tuned to each ERP's data model.
Reconciliation writes back as Bank Reconciliation entries in Tally. Multi-company installations get separate reconciliation queues per company. UTR and reference-number matching against Party Ledger vouchers.
Uses Zoho's Bank Reconciliation API and Inventory PO / Books Bill matching. Marketplace settlement reconciliation is particularly strong here because Zoho Books already has native marketplace order objects to match against.
Reconciliation posts as Payment Reconciliation entries against Sales Invoice and Payment Entry DocTypes. Purchase Order to Purchase Invoice matching respects Item-level tolerance settings you configured in ERPNext.
Match register is exposed in the ERPPlugs dashboard. Auto-clear write-back to the custom ERP happens over your mapped ingestion (REST call, CSV export, or middleware push).
What this looks like in practice
Real question-and-answer patterns from pilot deployments.
“UTR 20260703HDFC00238 credited ₹8,42,000 to current account 500123456.”
Matched to Sales Invoices TSA/2026/1108 (₹5,20,000) + TSA/2026/1112 (₹3,22,000) from buyer "Rajarshi Industries Ltd". Confidence: 0.94 (UTR + party + total amount + reference-number match on the payment advice). Auto-cleared; posted as Bank Reconciliation in Tally.
“Shopify settlement report for 1–7 July: gross ₹12,84,500 net ₹11,92,340 after fees.”
Matched to 148 Shopify orders (147 in Zoho Books, 1 missing). Reconciliation posted with Shopify fees ₹92,160 booked to "Payment Gateway Fees" ledger. Missing order #SPY-2026-4471 flagged for review — order exists on Shopify but was not synced to Zoho Books; suggested fix ready to queue.
“Purchase Invoice PI-2026-0294 received for ₹4,80,000 from Vendor "Kirti Metals".”
Matched to Purchase Order PO-2026-0180 (originally ₹5,20,000). Difference ₹40,000 due to short-supply on 2 line items — matched to the corresponding Delivery Note DN-2026-0221 which shows the short delivery. Reconciliation queued for your approval since the price variance is above your default tolerance.
Why this matters
Continuous, not month-end
Reconciliation runs as soon as data lands.
Handles the messy cases
Split payments, partial deliveries, netted settlements.
Confidence-scored auto-clear
Auto-clear above your threshold; human review below.
Every reconciliation audited
Every match, reason code, and reviewer action logged.
Frequently asked questions
How does the copilot handle one UTR against three invoices?+
The candidate generator considers combinations of open invoices whose sum is within tolerance of the incoming credit, filtered by party match. A UTR of ₹8,42,000 credited from Buyer X, matched to two open invoices of ₹5,20,000 and ₹3,22,000 for Buyer X, scores very high on the combined signal. This is the split-payment case that spreadsheet reconciliation handles badly and where most manual errors happen.
What about partial payments?+
Partial payments are matched against a single open invoice with the outstanding balance updated. If the buyer's historical pattern shows they routinely pay in 60/40 tranches, the copilot recognises the split without waiting for a second credit. On the second credit, the copilot matches the balance against the same invoice and closes it. Partial-payment history is one of the signals in the debtor-cadence baseline the cash-flow forecast uses.
How is marketplace settlement reconciliation different from bank reconciliation?+
Marketplace settlements arrive netted after commissions, shipping, refunds, and returns. The copilot ingests the settlement report line-item detail (Shopify, Amazon India, Flipkart all expose this) and matches each order to your ERP's sales entries, then posts the deductions to the appropriate expense ledgers automatically. Without this, the "settlement reconciliation" is a spreadsheet exercise that takes days per week for a mid-sized D2C brand.
What does the auto-clear threshold control?+
It controls the minimum confidence score at which the copilot auto-posts a reconciliation to the ERP without human review. Default is 0.90 for bank reconciliation and 0.85 for marketplace. You can tighten or loosen per source and per party category — e.g. auto-clear enterprise-buyer credits above 0.85, keep SMB credits at 0.90, and hold all cash credits for review regardless of confidence. Every threshold change is versioned in the audit log.
How does this compare to Zoho or Tally's native bank reconciliation?+
Native features do rule-based matching on exact UTR or reference number. The AI Copilot layer adds fuzzy UTR matching, split/partial payment handling, party-name normalisation, and confidence scoring — the cases where the native matcher gives up and passes the row to a human. The two layers coexist: native runs first for the clean cases, the copilot picks up what native could not match.
Is the reconciliation audit trail acceptable to a statutory auditor?+
Yes — that was a design constraint. Every candidate match, its confidence score, the auto-clear vs manual-review path, the reviewer identity for manual decisions, and the reason code for each reconciliation is logged with timestamps. During an audit the trail can be exported to CSV or read directly from your audit storage (S3 or on-prem bucket). Pilot customers have used this in India Rev-Sec audits and CFO board reviews.
Related AI Copilot capabilities
Every capability composes with the others.
Available today on these ERPs
Each edition uses the same connectivity ERPPlugs already ships.
Want this capability running on your ERP?
Join the AI Copilot early-access pilot. We calibrate against your own data during the first 2–4 weeks and lock in early-adopter terms before general availability.
Join Early Access