What does a payments back office actually do?
- Merchant onboarding, underwriting intake and agreements.
- Ledgers for every capture, refund, fee and chargeback.
- Fees, reserves and partner or ISO commissions.
- Disputes and chargeback handling.
- Payouts to merchants, partners and vendors.
- Support, risk monitoring and management reporting.
Where does AI help, and where shouldn't it?
AI is strong at the work around the money: reading onboarding documents, classifying and routing support tickets, spotting anomalies, explaining statements and drafting responses. It is the wrong tool for the money math itself. Ledger postings, fees and commissions should be deterministic, versioned code with an audit trail, so the same inputs always produce the same result.
The best division of labor is simple: code calculates, AI proposes and explains, and people approve anything that moves funds.
What does the architecture look like?
- 01
A double-entry ledger first
Post every capture, refund, fee and chargeback as balanced entries so the books reconcile to the cent. Everything else reads from the ledger.
- 02
Onboarding with e-signature
Collect applications, documents and signed agreements in one guided flow, with activation gated on what underwriting requires.
- 03
A commissions engine
Support the models partners actually use, such as buy-rate splits, revenue share and flat rates, computed automatically and paid on schedule.
- 04
Risk at the moment of payment
Enforce card fingerprinting, attempt limits and block lists when the payment happens, not in a report the next morning.
- 05
Payouts with guardrails
Batch payouts across ACH rails with balance checks before every send, bank holiday calendars and settlement tracking.
- 06
Agents with human approval
Let agents triage tickets and exceptions and propose payout batches, and require a person to approve before money moves.
- 07
Partner APIs
Document APIs and embeddable checkout so partners and point-of-sale systems can integrate without custom work for each one.
Should you replace a legacy back office?
If the back office runs on a system controlled by a third party, or on spreadsheets for commissions and payouts, every month of growth adds risk and manual work. Owning the platform removes the bottleneck. A practical migration path is to onboard new merchants onto the new platform going forward and let the legacy book run off, which avoids a risky one-time data migration.
How do you keep it compliant?
Keep raw card numbers out of your systems by using tokens from the gateway's vault, which shrinks compliance scope. Separate roles so the person who proposes a payout is not the only one who can approve it. Log every action, and build bank holiday and cutoff rules into the payout schedule rather than relying on someone to remember them.
Key takeaways
- Build on a double-entry ledger; everything else depends on it.
- Keep money math deterministic and auditable. Use AI to triage, explain and propose.
- Require human approval before funds move.
- Migrate by onboarding new merchants to the new platform rather than importing the old book.