Payment operations could answer repeated merchant and scheme questions from controlled knowledge before invoking a model. This thesis models lower inference spend without treating generative AI as a payment authorisation engine.
Industry thesis — modelled, not deployed. Figures are potential savings.
$2.4tnGlobal payments revenue in 2023McKinsey, 2024
236.6bnUS noncash payments in 2024Federal Reserve, July 2026
~75%Share of noncash value on ACH, 2024Federal Reserve, July 2026
52%Euro-area cash share by transaction countEuropean Central Bank, 2024 (latest SPACE wave)
Payments — the cheapest trustworthy source answers first; frontier reasoning is paid for only when it earns its price.
Where the money goes today
Payments: the cost of asking
McKinsey's 2024 Global Payments Report put global payments revenue at $2.4 trillion in 2023. Revenue scale makes operating efficiency relevant, but it does not establish the size of any processor's AI budget.
The Federal Reserve's 2025 Triennial Payments Study, released in July 2026, recorded 236.6 billion noncash payments in the United States during 2024, with ACH accounting for nearly three-quarters of noncash value for the first time. Those figures cover cards, ACH and cheques, illustrating the breadth of procedures that operations teams must navigate.
The European Central Bank's 2024 SPACE study found that cash accounted for 52% of euro-area point-of-sale transactions by number, down from 59% in 2022. Processors still face a mixed payment environment rather than one uniform digital workflow.
Three pressures
What makes this industry different
Scheme rules keep changing
Operational answers depend on effective dates, regions and payment schemes. A fluent answer based on an old rule can create avoidable rework.
Sensitive fields travel easily
Support tickets and dispute files can contain cardholder data. Redaction must happen before provider selection, not after a prompt has left the environment.
Exceptions consume expert attention
Most questions are repetitive, while a small minority require cross-system investigation. A single premium route would price both categories alike.
Five applications
How Qua would run inside payments
Each application is a real workflow, mapped to the tier that could answer it. The waterfall tries the cheapest trustworthy source first and only pays a frontier model when the expected value clears the gate.
Merchant settlement timing answers
Resolves at Personal Knowledge~60,000 questions/month
Support teams could reuse approved explanations for standard settlement schedules. Merchant-specific exceptions would leave the cache route rather than receive a generic answer.
Potential saving — 90–100% of query spend.
Apply merchant entitlements and prevent account-specific data exposure.
Match the question to a verified settlement-calendar or cut-off answer.
Return the current cached answer with a receipt; route exceptions onwards.
Chargeback evidence requirement lookup
Resolves at Enterprise Search~60,000 questions/month
Dispute teams could identify the required evidence without regenerating scheme guidance. Staff would verify deadlines and approve any submission.
Potential saving — 80–100% of query spend.
Mask cardholder fields and enforce scheme-document access rights.
Check verified answers, then search rules by scheme, region and effective date.
Return cited evidence requirements and a receipt inside the perimeter.
Reconciliation exception runbook retrieval
Resolves at Enterprise Search~30,000 questions/month
Analysts could locate the correct procedure for known mismatch patterns. Qua would not move funds or modify ledger entries.
Potential saving — 80–100% of query spend.
Apply processor and merchant access boundaries before retrieval.
Check verified fixes, then search approved reconciliation runbooks.
Return the relevant procedure and source references with a receipt.
Merchant incident update drafting
Resolves at Fast Models~40,000 questions/month
Operations could turn approved facts into consistent merchant updates. An incident owner would approve messages before distribution.
Potential saving — 80–95% of query spend.
Remove cardholder data and apply communication-provider restrictions.
Retrieve approved incident facts and check reusable update templates.
Use a permitted Fast Model to draft the update and record its receipt.
Complex settlement break investigation
Resolves at Pro Models~10,000 questions/month
An allowed Pro Model could compare competing explanations for a complex break and attach a cost-and-privacy receipt. Analysts would validate the diagnosis before any financial correction.
Potential saving — 0–20% of query spend.
Mask transaction identifiers and apply maximum-tier and provider policies.
Check known resolutions and retrieve permitted evidence across affected systems.
Escalate only if lower tiers are insufficient and expected value clears the premium gate.
The modelled ledger
What the same year of questions could cost
A thesis, not a case history. The assumptions are stated so you can replace them with your own numbers — which is exactly what a pilot does in week one.
Ledger rates are the vendors’ own published list prices: $0.0150 per premium question and $0.00065 per fast question at 1,500 input / 700 output tokens. Payment economics use Stripe's published 2.9% + $0.30 card fee and $0.07 Radar screening fee.
Modelled annual comparison
Workload: 200,000 questions/month for 12 months, at 1,500 input and 700 output tokens per question.
Published list prices, not estimates — premium baseline xAI Grok 4 at $3.00/$15.00 per 1M tokens = $0.0150/question.
Cheap metered tier xAI Grok 4 Fast at $0.20/$0.50 per 1M tokens = $0.00065/question.
Terminal-tier mix: 75% verified or in-perimeter, 20% Fast Models, 5% Pro Models. Rows exclude Qua fees, retrieval infrastructure, integration and human review.
Payment economics anchor: Stripe's published 2.9% + $0.30 per successful card charge and Radar for Fraud Teams at $0.07 per screened transaction.
WorkloadFrontier-onlyWith Qua
Verified answers and in-perimeter search1,800,000 annual questions resolved with no external model call.$27,000$0
Fast Models (Grok 4 Fast)480,000 annual questions at xAI's published Grok 4 Fast rate ($0.00065/question).$7,200$312
EV-gated Pro Models (Grok 4)120,000 annual questions keep frontier reasoning at the published Grok 4 rate.$1,800$1,800
At published vendor prices this thesis models $33,888 of avoided annual inference spend — $36,000 down to $2,112, a 94.1% reduction — before the excluded costs above.
Controls that matter here
Policy runs before routing, not after
Constrain cardholder data exposure
To support PCI DSS v4.0.1 requirements, a pilot would exclude sensitive authentication data from prompts and mask unnecessary cardholder fields before routing. Provider blocks and maximum-tier clamps would prevent prohibited external inference; Qua is not presented as PCI DSS certified.
Keep retrieval within scope
Under GDPR data minimisation and security obligations, Enterprise Search would remain inside the defined perimeter and respect merchant-level entitlements. Deployment would be assessed across Qua Cloud, customer VPC and air-gapped options rather than assumed to remove compliance scope.
Evidence operational AI decisions
For payment entities in scope of DORA, receipts would record exact inference cost, configured premium baseline cost, savings and what stayed private. Pro Models would require an expected-value gate after policy approval, and Open Models would be user-picked only.
Week one would measure settlement and dispute questions resolved locally, segmented by scheme and rule effective date.
Week one would compare inference cost and reviewer-rated accuracy with a matched premium-only baseline.
Week one would use synthetic cardholder records to test masking, merchant isolation and prohibited-provider blocking.
Sources
Every figure on this page, traceable
Market figures come from the publishers below. Qua savings are modelled from the vendors’ published list prices — they are not customer results.
[1]McKinsey & Company, Global payments in 2024: Simpler interfaces, complex reality (2024). Source Global payments revenue reached $2.4 trillion in 2023; this measures industry revenue, not processors' AI expenditure.
[2]Board of Governors of the Federal Reserve System, 2025 Triennial Payments Study — initial findings (2026). Source The 2025 Federal Reserve Triennial Payments Study recorded 236.6 billion US noncash payments in 2024, with ACH reaching nearly three-quarters of noncash value for the first time.
[3]European Central Bank, Study on the payment attitudes of consumers in the euro area (SPACE) — 2024 (2024). Source Cash represented 52% of euro-area point-of-sale transactions by number in 2024, down from 59% in 2022; this is not its share of all payments.
[4]PCI Security Standards Council, Payment Card Industry Data Security Standard: Requirements and Testing Procedures, Version 4.0.1 (2024). Source PCI DSS establishes requirements for protecting account data, restricting access by business need and logging access within the cardholder-data environment; routing data to another service can affect compliance scope.
[5]European Union, Regulation (EU) 2016/679 — General Data Protection Regulation, Articles 5, 25, 28 and 32 (2016). Source GDPR supports purpose-limited retrieval, data minimization, processor safeguards and appropriate security when payment workflows involve covered personal data.
[6]European Union, Regulation (EU) 2024/1689 — Artificial Intelligence Act (2024). Source The AI Act requires logging and oversight for applicable high-risk systems, subject to phased application; payment-processing or operational routing alone does not establish high-risk classification.
