Instant payments leave banks less time to detect, investigate and correct payment problems. Corporate clients expect rapid answers on missing payments, returns and recalls. Retail customers expect the same when they report fraud, challenge a card transaction or question an account-to-account payment. Auditors and regulators expect evidence that each decision followed policy and that higher-risk actions received the right oversight.

For executive teams, the central question has shifted from whether to automate to how to scale autonomy without introducing new operational, conduct or compliance risks.

HCLTech’s Future of Payments research captures that tension. While 99% of surveyed organizations use AI in payment operations, 91% of executives remain concerned about its risks. Meanwhile, 52% of organizations expect to become autonomous within 18–24 months, yet only 17% are fully operating in that mode.

The solution lies in trust by design, which requires clear policies, controlled decision rights, human review where judgment matters and a case record showing what happened, why and who approved it.

One payments vision, two operating realities

Corporate payment exceptions and retail disputes belong within the same strategic narrative, but they are not the same journey.

Corporate teams investigate missing or delayed payments, incomplete remittance information, status requests, amendments, returns and recalls. Their work often spans payment messages, correspondent institutions and client-service teams. Retail operations handle fraud claims and non-fraud disputes across cards, account-to-account transfers, real-time payments and digital channels, with network rules and consumer-protection obligations shaping the process.

Pega addresses these environments through related capabilities. Smart Investigate Agentic Automation supports commercial and cross-border investigations, while Smart Dispute Agentic Automation supports consumer disputes and fraud claims. Both provide case and workflow capabilities through which rules, decisioning, AI and human review can be applied to each journey.

Better automation begins with better intake

A case cannot run straight through when it starts with incomplete or ambiguous information. Poor intake pushes work downstream through callbacks, manual research, incorrect routing, missed service levels and weak audit evidence.

In retail disputes, putting a digital front end on a static form does not solve the problem. Intake should recognize the payment type and dispute reason, retrieve information already held by the bank, ask relevant questions and collect the evidence required by the workflow. Pega’s analysis of dispute intake makes the point directly: clean, adaptive intake is the foundation for higher straight-through processing.

Consider a retail customer who disputes a card transaction because the purchase was settled through another method. Guided intake can identify the transaction, capture supporting evidence, check for refunds or duplicate activity and route the claim through the applicable card-network workflow. Manual review is reserved for conflicting facts, missing evidence or cases that cross a defined risk threshold. Pega Smart Dispute supports payment-network rules, regulatory requirements and guided capture of essential information at the start of a claim.

The corporate equivalent may begin when a client reports a missing cross-border payment with partial remittance details. The platform can retrieve transaction and message data, request what is missing, apply relevant policy checks and determine whether to initiate a status inquiry, amendment, cancellation or recall. It can then coordinate the next action with a correspondent institution or counterparty.

Corporate journeys involve clients, payment messages and counterparties. Retail journeys involve customers, claims, merchants and payment networks. The orchestration model can be shared, but the language, controls and external interactions must fit the journey.

Straight-through processing before generative AI

The first operational goal should be higher straight-through processing (STP). Routine cases should progress through deterministic rules, policy logic and prescribed workflow wherever the required action is clear.

Generative AI should not be the starting point. Placing a large language model inside a fragmented process can make the outcome harder to predict and explain. Fix intake and orchestration first, then add intelligence where it removes measurable effort or improves a bounded decision.

Rules and workflow manage eligibility, deadlines, approvals, accounting actions and payment-rail requirements. Predictive AI can prioritize cases and estimate likely outcomes. Generative AI can interpret correspondence, summarize case histories, retrieve guidance and draft communications. Agents can execute approved playbooks such as requesting information, updating systems or preparing a network submission.

In every case, workflow should define what automation may do, which evidence it must retain and when a person must intervene.

Trust by design is an operating discipline

The hardest test of autonomy comes after the decision. Can the bank explain it to a customer, client, auditor or regulator without reconstructing the case from emails and system logs?

That challenge extends beyond payments. HCLTech’s recent Blueprint for AI Leadership found that only 17% of organizations show high confidence in agent-initiated actions. In payments, where an autonomous action may affect funds, customer outcomes or regulatory obligations, confidence depends on knowing what an agent can do, under which conditions and with what oversight.

A governed model should make the path visible: the data available, the rule or model applied, the action taken, the threshold that triggered review and any human approval or override. A single case record gives operations, risk, compliance and audit teams a common source of evidence.

This is where leadership and practitioner responsibilities meet. Leaders set risk appetite, decision rights, service expectations and escalation thresholds. Practitioners test whether those controls work under real conditions. Their feedback exposes missing data, excessive handoffs and thresholds that create unnecessary manual work.

As Steve Morgan, Global Banking Industry Lead at Pegasystems, puts it: “The key is balancing this autonomy with the transparency and control that builds lasting customer trust.”

Trust is not a communications layer added after deployment. It is built into intake, workflow design, access controls, decision logic, human-review points and auditability.

Move to production in bounded steps

Banks should start with one high-volume journey, one payment rail and one channel. A corporate starting point might be a missing-remittance investigation or recall while a retail starting point might be a defined non-fraud dispute or low-value fraud claim.

The first phase should be a bounded production release, not a disposable pilot. Scope should be narrow enough to reach production quickly, while allowing sufficient time to validate data, integrations, policy controls and human-review points.

Before launch, the team should agree on required data, STP eligibility, policy thresholds, human-review points, audit evidence and ownership of the outcome. Performance should be measured through intake completeness, cycle time, STP, rework, handoffs, communication quality, deadline adherence and audit-trail completeness.

Expansion can then proceed across channels, payment rails, dispute or exception types, risk tiers and geographies. The scope should grow only when operational results and control performance remain consistently strong.

The operating model shift

Autonomous payments are not simply a technology upgrade. They change how responsibility is distributed across people, policies, systems and AI.

As autonomy increases, payment operations teams can spend less time progressing routine cases and more time handling exceptions, reviewing higher-risk decisions and improving the rules under which automated work takes place. Practitioners increasingly become supervisors and designers of how work gets done, while retaining accountability where judgment and material risk demand human intervention.

For HCLTech and Pega, the opportunity is to connect platform capability with this operating model: complete intake, higher STP, selective AI, governed execution and evidence that stands up to scrutiny.

Speed is now expected. Trust determines whether that speed can scale. In payments, acting faster creates value only when the institution can prove why it acted.