Genpact makes Onyx available for governed enterprise AI agents
Genpact projects up to 80% touchless invoice processing through AP Suite while Onyx grounds enterprise agents in process context, earned autonomy and audited controls.

Why it matters
- Genpact announced the availability of Onyx on November 16, 2026.
- Genpact unveiled Onyx at its third annual Agentic Operations Day in New York on November 12, 2026.
- Genpact projects up to 80% touchless processing for AP Suite invoice workflows.
- Genpact projects more than 95% duplicate prevention for AP Suite.
- Genpact projects automatic resolution of up to 90% of supplier queries.
By Rebecca Stone
Genpact projects up to 80% touchless processing for invoice workflows through its Genpact Accounts Payable Suite, alongside more than 95% duplicate prevention and up to 90% automatic resolution of supplier queries. Genpact announced the availability of Genpact Onyx on November 16, 2026, a platform for building, running and governing AI agents in enterprise operations.
The disclosure follows Genpact’s unveiling of Onyx at its third annual Agentic Operations Day in New York on November 12, 2026. For enterprise buyers, Genpact frames the stakes as trusted execution rather than simply adding more agents. That places auditability, exception handling and expert supervision at the center of the product claim.
What changes with Onyx’s availability?
Genpact says its Process Intelligence Core gives agents operational context drawn from workflows, policies, exceptions and human decisions. The company says this lets an agent begin with knowledge of how a business process is actually handled, including cases that static process maps or prompt libraries often omit.
Genpact is using Onyx across its own agentic products, custom client systems and partner-delivered solutions. Its proprietary product line starts with the Genpact Accounts Payable Suite. That gives Onyx a concrete commercial test: whether its operating model can reduce routine accounts-payable work without removing exception paths or accountability.
How does Genpact define earned autonomy?\n
Priya Vijayarajendran, Genpact’s chief product and platform officer, said: “Enterprises get value from AI when they turn operating knowledge into trusted execution, not simply by adding more agents.” That statement ties Onyx to a specific enterprise problem: converting internal process knowledge into supervised execution. The release treats governance and operating context as core platform capabilities, rather than an add-on after agents reach production.
Under Onyx’s operating model, agents execute configured activities, people supervise outcomes and manage exceptions, and orchestration software coordinates work across systems, agents and people. Human experts retain accountability for results. Genpact’s sequence makes supervision an operating condition rather than an optional handoff.
Genpact calls the authority model “earned autonomy.” Agents begin by proposing actions for human approval. They can receive greater decision responsibility after they demonstrate performance under human supervision. Authority therefore expands from reviewed results rather than from a fixed role assigned at launch.
Reviewed exceptions and expert decisions flow back into the Process Intelligence Core. Genpact describes this as a learning loop intended to make later cases faster and more autonomous. Its product documentation ties any improvement to configuration, governance, human oversight and approved operational data. The documentation therefore treats autonomy as a configured and governed outcome.
What governs an agent’s actions?\n
Genpact says every Onyx agent action is controlled, auditable and traceable to the reason for the decision. Supported model calls and tool actions pass through governance controls that, depending on configuration, can mask personally identifiable information, enforce policy and record audit logs. Genpact says its responsible-AI controls also address bias mitigation and transparency.
These controls place the audit trail inside the execution path rather than beside it. That distinction supports Genpact’s claim that an agent can gain responsibility without surrendering institutional oversight.
Onyx’s knowledge layer spans finance, risk, supply chain, banking, insurance and procurement, with prebuilt workflows and agents. Genpact says it also records exceptions, edge cases and human decisions that rarely appear in prompt libraries or static process maps. Those records give deployment its starting context while orchestration software coordinates work among systems, agents and people.
How does Onyx fit into existing systems?\n
Onyx supports compatible models, frameworks and cloud infrastructure, subject to technical requirements. Genpact identifies AWS, ServiceNow, Microsoft and Databricks among the data and AI technologies it lists for potential integration. The company says the platform is designed to fit a client’s existing technology environment and draw on its data and AI foundations. Compatible models can also come through Genpact’s partner ecosystem.
Routing-layer configuration can let an organization change certain components without rebuilding the full solution, depending on architecture and integration requirements. That approach gives Genpact a path into installed enterprise systems without requiring every model, framework or cloud service to use the same configuration.
Through a context interface, agents can be made accessible in supported third-party applications. Genpact says the interface tailors information to configured data, permissions and workflows for users such as architects, delivery managers and developers.
The company says metrics aligned with established key performance indicators can evaluate agent performance and inform outcome-based commercial models. That feature links technical operation to contracting by giving buyers measured performance data against which to assess commercial terms.
How does AP Suite apply the model?
Genpact’s AP Suite divides accounts-payable work into four modules:
- AP Capture handles invoice intake, classification and enrichment.
- AP Advance manages three-way matching, exceptions and non-purchase-order invoices.
- AP Trace looks for duplicates and unusual patterns.
- AP Assist identifies intent and automates responses to supplier queries in multiple languages.
Across the workflow, agents help validate and prioritize invoices, manage auto-posting and route cases to people when needed. Genpact presents the suite’s main outcome figures as projected estimates: up to 80% touchless processing, more than 95% duplicate prevention and up to 90% of supplier queries resolved automatically.
Genpact labels the numbers as projections, positioning them as targets for implementation rather than established performance averages. The AP Suite therefore serves as the first named product through which buyers can examine the operating model, controls and escalation paths behind those targets.
The suite is available as a software-as-a-service offering. Genpact says clients can license and deploy it without Genpact managed services, while implementation, advisory and support remain separate services. Audit trails, exception escalation and role-based access are built into the product.
Genpact says client data stays within the environment set by the agreed deployment model and is not used to train models for other clients. That commitment makes the selected deployment model part of the suite’s data-governance proposition, alongside its built-in controls and escalation mechanisms.
What will Onyx have to prove?
Genpact says Onyx connects enterprise AI models to day-to-day execution through process knowledge, governance and expert supervision. AP Suite puts that approach into accounts payable, where routine volume meets defined exception paths and accountable review.
The forward test is whether the projected touchless-processing, duplicate-prevention and query-resolution figures hold in deployments while earned autonomy preserves clear escalation, audit trails and human accountability.
Source: unite.ai
More from Rebecca Stone
Show full bio
Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
728 articles
Related articles
- Onyx and Glean integrate controls for enterprise AI agents
- OpenAI launches ChatGPT agent, folding Operator into ChatGPT
- Genesys Sells AI Agents Without the Cloud Migration, Priced by Effort
- Over 40% of Agentic AI Projects Face Cancellation by 2027
- OpenAI's GPT-6 agents now do medical paperwork for 8.5 cents a task