← Blog Automating a process
Unidentified bank payments: how to match them without applying them wrong
Unidentified bank payments need a method, not random searches in the bank statement. The idea is to match bank data, ERP records, invoices, and receipts, then apply only when there’s enough evidence.
Key takeaways
- Don’t apply by similarity. If the amount matches but several customers are possible, the payment should stay pending with evidence and human review.
- Ask for different data by payment rail. In PSE, the CUS helps. In SPEI, ask for the tracking code or CEP. In local transfers, ask for the full bank receipt.
- Automate repeatable rules first. Exact reference, exact invoice, tax ID, and expected amount usually work better than asking AI to decide on its own.
- Measure 4 things. Identified percentage, average time, false positives, and aging of open items show whether the process is improving or just moving the problem.
How do you identify unidentified bank payments?
The goal isn’t to find a fast match. The goal is to leave a defensible accounting application, with a trail of the data used.
- Separate received cash from the accounting application. When a payment arrives without a clear reference, record the cash in a suspense account or unclassified item. This keeps you from applying it to the wrong customer while you investigate. In Colombia, the 2016 Resolution 193 of the General Accounting Office requires periodic reconciliations. It also requires analysis against sources like banks and accounts receivable, plus working-paper evidence to support adjustments.
- Normalize the bank statement. Turn each bank movement into comparable fields. Start with date, amount, origin account when available, and bank. Then keep description, reference, and channel. Add currency and time when available. If the bank sends files in different formats, an integration avoids copy and paste. In many cases, this is solved with integrations and APIs, not AI.
- Match against invoices and open receivables. Look for exact matches first: invoice number, collection reference, customer document, and amount. Then look for close candidates: partial payments, bundled payments, discounts, interest, or overdue invoices with the same amount.
- Ask for the right receipt. Don’t ask for “payment support” in a generic way. Ask for the data that traces the rail used: CUS in PSE, CEP or tracking code in SPEI, the full receipt for a bank transfer, and a debit screenshot only as supporting evidence.
- Apply with evidence and save the decision. The application should save which fields matched, who approved it, when it was applied, and which invoice or obligation received the payment. If the system proposed candidates, it should also save the score or rule used.
What data should you ask for by payment type?
Asking for the right data reduces back-and-forth with customers, support, and banks. It also lowers the risk of applying a collection to the wrong customer.
| Payment rail | Data to ask for | How it’s used |
|---|---|---|
| PSE Colombia | CUS and confirmation email | Used to validate status and match the transaction to the invoice. |
| SPEI Mexico | Tracking code or CEP | Used to retrieve the receipt and validate the payment. |
| Bank transfer | Full receipt | Helps match date, amount, bank, and origin account. |
| Branch or correspondent payment | Printed reference or receipt | Helps rebuild the collection when the description is poor. |
| Corporate bulk payment | Detail file | Used to split invoices inside one deposit. |
In PSE, users can pay by debiting a savings account, checking account, or electronic deposit. If a transaction was approved but the merchant didn’t apply it, PSE tells users to validate it with the merchant. They should use the CUS, or “Código Único de seguimiento”, sent by email. The business module also lets companies search by CUS/T.C., customer number, invoice, and status. It also shows cycle, amount, and extra transaction data. PSE explains this in its business help center. For Mexico, Banxico explains that SPEI uses “Concepto del Pago” and a reference number. To obtain the CEP, it asks for data like date, reference, or tracking code. It also asks for sender bank, receiver bank, beneficiary account, and amount. This appears in its SPEI and CEP FAQ.
What does an unidentified collections case look like in operations?
This is an illustrative example, not a customer. It helps ground the flow for a company that receives many bank deposits.
A B2B company receives a deposit for the same amount as an overdue invoice. The statement says “vendor transfer” and has no tax ID, invoice, or clear name. In receivables, three customers have invoices for the same amount.
The system doesn’t apply the payment. It creates an item to classify and proposes three candidates. For each candidate, it shows evidence. That includes open invoice, amount, due date, and payment history. It also shows recent customer email and receipt matches if present.
Support asks the payer for the full receipt. The customer replies with a transfer that shows origin bank, date, amount, and company name. The system updates the case, rejects two candidates, and leaves one with enough evidence.
The receivables person approves the application. The ERP records the collection against the correct invoice, closes the suspense item, and saves the decision trail. If the same customer pays again with a similar description, the rule stays available. It still asks for review when the same amount matches several customers.
This kind of process connects to automatic bank reconciliation. The difference is that the focus here is on ambiguous items, not movements that already match exactly. If the problem is concentrated in PSE, it’s worth reviewing the specific flow for PSE payment reconciliation.
When should you use rules, automation, or AI?
Unidentified payments don’t all need the same treatment. The decision depends on data quality, volume, and error risk.
| Option | When it fits | Limit |
|---|---|---|
| Bank or ERP tool | Few cases and well-filled references | Struggles with poor descriptions. |
| Reconciliation rules | Consistent references, amounts, and invoices | Fails with partial or ambiguous payments. |
| Workflow automation | Several areas are involved and there’s rework | Needs orderly data and clear owners. |
| AI for extraction | Receipts arrive as PDFs, images, or emails | Should propose, not decide on its own. |
| Custom development | Many banks, ERPs, and business rules | Requires testing and measurement before running. |
AI helps extract receipt data, normalize text, detect candidates, and summarize evidence. For ambiguous payments, it shouldn’t apply accounting entries on its own. At Quarl, we usually design these flows as AI automation: rules when they’re enough, AI when it adds value, and human review when there’s risk.
What metrics show whether the process is improving?
An unidentified payments dashboard should measure quality, not only processed volume. If the team identifies payments faster but makes more mistakes, the process got worse.
% identifiedPercentage of collections classified against a customer, invoice, or obligation.
Average timeTime between bank deposit and application or documented rejection.
False positivesPayments applied to the wrong customer or invoice.
AgingValue and number of unidentified items by day range.
What we do
Frequently asked questions
How do you identify a payment with no bank reference?
Start by separating the cash into an item to classify. Then match date, amount, origin account, description, reference, and channel against open invoices, receivables, and customer receipts. If several candidates are possible, don’t apply by similarity. Ask for the right receipt and leave evidence for the decision.
What data should I ask the customer for to trace a transfer?
Ask for date, amount, origin bank, destination bank, beneficiary account, payer name, full receipt, and reference used. If it was PSE in Colombia, ask for the CUS. If it was SPEI in Mexico, ask for the tracking code or CEP. Avoid loose screenshots if they don’t show verifiable data.
How do I apply a collection if I don’t know which customer paid?
Don’t apply it right away. Record it as a collection to classify and look for candidates in the ERP. If you find one exact and unique match, apply it with evidence. If there’s ambiguity, escalate to receivables or support to request the receipt. The rule is simple: without enough evidence, the payment shouldn’t close an invoice.
What is the PSE CUS and what is it used for?
The CUS is the Código Único de Seguimiento for a PSE transaction. It is used to trace payment status and help the merchant validate an approved transaction that still hasn’t been applied. In companies, it helps match the payment with customer number, invoice, amount, and status.
How do you avoid false positives when assigning payments to invoices?
Use exact matches before approximate matches, require more than one signal when the same amount repeats, separate partial payments from full payments, and send ambiguous cases to human review. It also helps to measure false positives and correct rules. Automation that applies fast but gets it wrong creates more accounting work.
Written by
Quarl engineering team
We are an AI engineering team in Medellín. We build and run systems that reach production — automations, assistants and agents — and for three years we ran an assistant for two million users across ten countries.
See how we work →Want to apply this in your company?
In fifteen minutes we tell you whether it is worth it, free and with no pitch. The proposal arrives within 48 hours.