Originally published here at CIO

The test automation startup’s ERP Lifecycle Optimization Platform aims to revolutionize how enterprises configure, test, and support ERP operations via AI agents.

Opkey, a startup with roots in ERP test automation, today unveiled its agentic AI-powered ERP Lifecycle Optimization Platform, saying it will simplify ERP management, reduce costs by up to 50%, and reduce testing time by as much as 85%.

“If you look at a lot of ERP project failures, the problem is not the ERP itself,” says Pankaj Goel, CEO and co-founder of Opkey. “The problem is how you are implementing it, how you are testing it, how you are supporting it. That is what we’re attempting to solve with this agentic platform.”

In 2019, McKinsey reported that 75% of ERP transformation projects fail, and nearly two-thirds risk budget overruns of an average 45%. Goel says the enterprise approach to ERP implementations and operations have largely remained static for 20 years. He believes enterprises need to reimagine these transformations by leveraging AI agents for deploying, operating, and optimizing enterprise applications.

Platform powered by specialized ERP SLM

Starting with its expertise in test automation, Opkey built a small language model (SLM) called Argus AI specialized for ERP. Argus AI orchestrates the ERP Lifecycle Optimization Platform’s virtual agents in the execution of complex, multi-step workflows across the lifecycle of applications that support HR, manufacturing, supply chain, and finance processes.

Opkey has created four AI agents to assist customers across their ERP lifecycles:

• Configuration agent: This agent is responsible for configuration mapping, migrating planning, and execution to accelerate complex deployment workstreams.
• Testing agent: The testing agent performs test automation and patch updates leveraging AI-powered, self-configuring test scripts and pre-built test scripts.
• Training agent: This agent is for change management and end-user enablement. It uses test case data and AI-generated user guides to assist users via job aids, training, and personalized in-app prompts.
• Support agent: The support agent assists end users at scale by analyzing user journeys, identifying root causes, and prompting the user to resolution.

“You can call them digital workers,” Goel says. “Each of these have their own independent thinking.”

Goel notes that, for now, humans remain in the loop. An AI agent, for example, could monitor a system and if a business user encounters an error, it could provide help in-browser. If that doesn’t resolve the issue, an agent could then help the business user log a ticket with a single click. Another agent at the back end could then view the data gathered and provide a recommendation. But, at least for the next several years, a human will still look at the data, look at the recommendations, and then make a call.

“Any successful agent AI, particularly for our domain of ERP implementation, it will be a long time before it starts working autonomously,” Goel says. “Human-in-the-loop is required for the right governance, to make sure that all the right decisions are being made by the AI, and then gradually and slowly improve it going ahead.”

But even with humans overseeing the work of AI agents, Goel says agents are poised to dramatically reduce the time required to implement and operate ERP systems.

“Why do you require a year to configure a system? Today, you require a human to understand what the customer wants and put it in a configuration worksheet, but then it should be done overnight, running on 100 machines,” he says.

As of today, the platform and agents have been trained to support more than 15 Oracle ERP applications. Goel says support for Workday integration is in the works and will launch soon. Other implementations will follow.

“Training the small language model for specific ERPs is a non-trivial problem, so we are doing it in phases,” Goel says, though he notes the testing agent is already available for many existing ERP systems.

He notes that training the lifecycle optimization platform for Oracle applications took about a year but training it for applications from other vendors should be much faster because the core nerve center has already been built.

Opkey works directly with customers but also partners with large systems integrators such as KPMG, Huron, and PwC.