Vendors fortify AI in ERP to assure buyers, deter frontier AI rivals

Major ERP product introductions and conferences this fall provide a snapshot of the fast-evolving role of AI in ERP. They show vendors striving to prove they have the infrastructure, development tools and user experiences that organizations need to implement AI confidently, and backing it up with real-world examples.

The SAP Connect conference in Las Vegas and Infor Velocity Week in Orlando, Fla., last week followed Oracle's unveiling in late September of Fusion Claw runtime software, which provides 25 agentic AI applications with the built-in governance and security viewed as essential to scale AI across the enterprise.

At Connect, a conference SAP inaugurated just last year to unify separate conferences for specific business applications like finance, HR and spend management, the actual news was sparse. The company introduced payment software called SAP Pay and said it would acquire TechWolf, which makes an AI model that captures an organization's work, tasks and skills.

But SAP was forthright that the main purpose of this year's Connect was to announce that the agentic AI products unveiled in May at its biggest conference, Sapphire, are available this month. It also showcased high-profile early adopters on the keynote stage: Nestle, Novartis, PwC and Morgan Stanley.

Meanwhile, Infor spotlighted C-suite leaders from Aston Martin and Team Air Distributing after unveiling its Infor Industry AI architecture. The new release includes AI agents with expertise in industry micro-verticals (Infor's longtime selling point), an architecture designed to work with third-party applications, a UX said to adapt to individual users' needs and a governance, risk and compliance layer that agents must pass through.

These events and discussions with vendors and analysts point to four trends solidifying amid the broader effort to augment ERP systems with AI.

1. ERP vendors have almost finished their AI UXs

Soon after ChatGPT burst onto the scene in November 2022, sparking today's AI craze, ERP vendors began promising that the technology would finally let them put a user-friendly front end on systems long derided as complex and hard to use.

The recent introductions suggest those promises have mostly been fulfilled. They all use generative AI (GenAI) for their UX but closely integrate it with agentic AI, so users can make requests in their native languages that the AI agents can translate and execute. In SAP's case, the UX is Joule Work, a significantly expanded version of the GenAI assistant introduced in 2023.

Joule has grown from a small, natural-language chat client to a full web, mobile and desktop UX, said Manoj Swaminathan, president and chief product officer of SAP Autonomous Suite and keynote speaker, in an interview previewing the conference.

"We are pivoting to a user experience that is no longer the static set of experiences where we expect people to be going to a particular location and completing tasks," he said. "It's going to be a purpose-driven, targeted experience for an event, a situation or a particular task that they need to complete. In that context, Joule Work becomes a single engagement layer for how the users could interact."

Joshua Greenbaum, principal at Enterprise Applications Consulting, said SAP and other ERP vendors are meeting an important market need for user-friendly front ends and adding an agentic layer beneath to translate natural language into action. But he cautioned that the new AI-driven ERP UX remains aspirational.

"It's there in terms of the design and the experience, but it's not there in being able to say in the near future that we will flip a switch and everybody will have this experience," Greenbaum said. "Those agents still have to be built, vendors have to build the infrastructure to support it, and they have to build security infrastructure to make sure that stuff doesn't get misused or misdirected."

2. AI is challenging SaaS ERP strictures against customization

The variety, specificity and personalization of the agentic AI apps that users of multitenant SaaS ERP can program themselves or with the help of vendors and system integrators are strongly reminiscent of on-premises ERP customizations. SaaS ERP was supposed to discourage customization by making it difficult to impossible.

The trend is partly driven by ERP vendors' tendency to pitch the personalization and customization features of their AI.

"We're a firm believer that there is no such thing as out-of-the-box AI," said Rick Rider, Infor's senior vice president of AI innovation, in a pre-conference interview. He cited an Infor survey showing most business decision-makers feel the same.

Rider said Infor's Velocity program is designed to give customers "unlimited" access to the entire Velocity Suite of AI technologies, so they can quickly deploy industry-specific AI apps and customize them. It can also help customers become less dependent on Infor consulting services, though they remain available.

"We want you to have the ability to have a conversation with your system, to build your extensions, change your user experience, make it more adaptive, build more agents, or build an integration with a copilot or third-party systems," Rider said.

Jon Reed, industry analyst and co-founder of Diginomica, agreed AI is enabling a new kind of ERP customization.

"If you do it in a well-governed way, it does potentially make it easier to cut through some of the trade-offs there used to be between vanilla SaaS and more extensible SaaS," Reed said.

Vendors have traditionally pushed software extensions as the way to achieve limited customization, but they require organizations to jump through hoops and sometimes find a partner to help, he said. AI agents can be a faster way to augment ERP with useful capabilities.

"It's different than just vibe coding at home because you're doing it on a platform that's giving you a head start with all the concerns you might have around privacy, data access and authorizations," Reed said.

Companies must nevertheless put governance and other controls around employee AI development to avoid duplication and ensure the apps are usable across the enterprise.

"There's still a role for IT if you're going to propagate the app and make it available to others," he said.

Greenbaum sees the customization trend as more of a risk than an opportunity because it could encourage a reversion to bad habits that result in over-customized, hard-to-maintain ERP systems and negate some advantages of multitenant SaaS ERP, with its generic industry best practices.

"Now, customers can build functionality with AI that they could otherwise get in an enterprise software package," he said, citing the example of a Danish company that used Joule to build an app that duplicates a standard feature of SAP Ariba procurement software. "Now that customer basically has a customized environment to do something that rightfully is already a best practice in an existing piece of SaaS software, which is really where the customer should be."

Furthermore, using AI this way for short-term gain could undermine the foundations needed for governed, accurate AI in the long run, Greenbaum said.

What SAP is more concerned about is becoming an API call for someone else's AI.

That risk was confirmed for him when every customer on an SAP Connect panel said SAP's clean core approach was vital to their AI deployment plans. Clean core calls for leaving the ERP itself largely untouched, with differentiating features programmed as extensions in the vendor's platform-as-a-service development environment, Business Technology Platform.

"That was a really strong statement, unprompted by SAP, that said we as practitioners in the vanguard know that you're not going to fake it customizing agents," he said. "We're dancing in a dangerous sort of direction with this unlimited agentic tool approach because you still need that discipline to keep this stuff from being needlessly and excessively customized."

Greenbaum said organizations will still view customization as a competitive advantage, which is fine if they respect its limitations and stay disciplined.

3. ERP vendors compete more with hyperscalers and frontier model makers than each other

A common theme the ERP executives and analysts expressed is that competition for AI is less between ERP vendors and more with the companies claiming to offer an alternative AI UX for ERP. They said hyperscalers like AWS and Google Cloud and frontier-model vendors, such as Palantir and Anthropic, can't deliver the required accuracy and governance because they lack full access to the data and business logic in ERP systems.

Reed said the issue was so much on the mind of SAP CEO Christian Klein that frontier models were the first thing he mentioned in his keynote.

"They're trying to show that SAP has unique abilities to do this kind of stuff with business data," Reed said. "What SAP is more concerned about is becoming an API call for someone else's AI. They don't want to be perceived as a system of record."

Reed said frontier model vendors play fast and loose by telling customers they can avoid the heavy lifting on data and get going right away using their AI to interact with ERP systems.

"I've seen some customer situations where it hasn't worked very well," he said.

Greenbaum calls this the Palantir palliative or fantasy. "Palantir gets into a space where they think, 'we can screen scrape our way into providing you next-generation agentic AI.' But because it doesn't have the ability to find the really deep nuances in the metadata underneath the enterprise, it's not going to do a good job."

Greenbaum said a welcome distinction is emerging between the enterprise capabilities of general-purpose large language model vendors and those of enterprise software companies.

"If I had to bet money on which kind of company would do a better job really meeting the needs of the enterprise and driving very deep into enterprise requirements, functionality and experiences, I would trust any of those vendors over a Palantir and Anthropic or anybody else who isn't distinctly steeped in the historical understanding of what makes enterprises tick."

4. Vendors are anxious to prove agentic AI is safe for enterprise use

Oracle emphasized the governance and security controls in Fusion Claw, and competitors are doing the same with their agentic AI rollouts.

For example, Swaminathan said SAP's Autonomous Enterprise products come standard with the authorizations and authentication established for human users in the ERP system and can be tailored by each customer.

"The [agentic] runtime is natively part of our platform and will be an SAP-managed solution," he said. "That means we provision, enable, activate and manage the end-to-end lifecycle of the platform and anything and everything that runs on top of it. SAP-provided agents, the ones that are extended, [and] custom agents developed by partners and customers on top of this platform, will have the same governance promise."

The analysts noted how the security message was reinforced by the appearance at the end of the Connect keynote of Marielle Ehrmann, SAP's chief security officer.

"She did a really good job of talking about how it's not just bad actors. It's that agents without proper governance can also do a lot of harm without anyone intending them to," Reed said.

There's reason to believe organizations are getting the message.

Greenbaum said most customers he talks to are moving carefully to cover the basics of AI deployment, which include getting their data and process houses in order and setting up change management ahead of time, so AI deployments and ERP upgrades go smoothly.

Despite pressure from boards amid an incessant drumbeat for AI in the market, they're putting in the hard work of laying a strong foundation, he said.

"Whether that foundation allows them to jump into something really interesting with agentic tomorrow or in two years or five years, success on any timeline will depend on getting the foundation right."

David Essex is a senior editor who creates in-depth content on enterprise applications, emerging technology and market trends for TechTarget's CIO group.