Ask three people at your organization what “AI” means for your ERP and you’ll probably get three different answers. They’re likely talking about three different things.
In my last post, I talked about a pattern we’ve seen before. When the market feels uncertain, organizations rarely sit on their hands entirely. They stay busy, evaluating, piloting, comparing options. All of that activity is meaningful, but it doesn’t always produce a clear answer, and for some organizations, that search for clarity can stretch into months or even years. With AI, that pattern is everywhere right now. Almost every organization we talk to is doing something. Far fewer have a clear strategy behind it, one that holds together across the whole organization rather than living in scattered pockets.
That gap matters because the word “AI” gets stretched to cover a handful of genuinely different categories of solution. Thinking about AI in these three categories can help you set appropriate expectations for what success looks like.
Category 1: What you already have — the AI built into your ERP platform
The most common starting point is AI that lives inside the systems you’re already running, whether it showed up as a built-in feature or you specifically paid to add it. What that looks like depends on which platform you’re on, though less than you’d think. Oracle and Workday have both shipped AI directly into their cloud applications. Oracle alone has rolled out hundreds of AI agents across Oracle Cloud’s finance, HR, supply chain, and customer service modules over the past year, included at no extra cost. Workday has done something similar. If you’re running PeopleSoft, the experience looks different. Oracle’s strategy there is to provide some of the building blocks that let AI reach your data, rather than hand PeopleSoft organizations a finished AI experience the way it has with Oracle Cloud. So there are really two situations here: platforms where the company delivers the AI itself, and PeopleSoft, where Oracle hands you the building blocks and leaves the rest to you.
Whatever AI a company builds for you, or hands you the building blocks to build yourself, is contained within that company’s own world. It can answer questions using data that already lives there. Reaching beyond that, into your other systems, takes deliberate work. Sometimes that means extending what Oracle provides for PeopleSoft. Sometimes it means layering an orchestration tool on top of Workday or Oracle Cloud to reach further than either was built to go on its own. Either way, nothing connects automatically just because a feature got turned on.
There’s also something easy to miss in the moment you turn one of these features on. You don’t always know what’s doing the work behind it, or where your data goes to generate an answer. That’s not a reason to avoid these features. It’s a reason to know what you’re agreeing to before you use them.
None of this is a knock on the platforms. Building AI into a major ERP, whether as a finished feature or as a building block for others to build on, is a significant engineering investment. Oracle has taken that approach across its platforms, shipping finished features directly into Oracle Cloud and providing the building blocks that PeopleSoft organizations and their partners can build on. Workday has made a similar investment on its own platform.
Back when a different technology had people predicting that a vendor-delivered tool would make partners obsolete, our CEO Paul Taylor said something that still resonates. “People are saying it’s all going to be free now. It’s not. And if Oracle delivered through PeopleSoft a tool that made our product offerings obsolete, we would use it.” That’s still how we think about it, on any platform, including with our AI offerings. If a platform ships AI that genuinely solves the whole problem for your organization, that’s a win, and we’d be the first to say so. Most of the time, though, it doesn’t, because the problem that matters most usually lives across more than one system, or can’t be addressed with an out-of-the-box process or agent. That’s the part we know how to fix.
Category 2: A great tool for one job
The second category is the single-purpose tool, AI built into one specific application to do one specific thing well. Unlike the AI built into your core ERP platform, these tools are narrow by design and aren’t built to orchestrate or expand across your systems. Think of Co-Pilot helping you search your email, or transcription and summary features in your meeting platform. For the job they were built for, it’s hard to beat what a purpose-built tool delivers.
The picture gets more complex when requirements reach farther across the enterprise than the tool was built to go. A student advising platform with a built-in assistant shows how quickly that complexity surfaces. On paper it checks every box. It’s already part of the advising workflow, the institution already owns it, and it has “AI” right there in the feature list.
We worked with the Virginia Community College System (VCCS), which ran into exactly this. They wanted their advising AI to answer student questions, such as: can I transfer these credits, what classes should I take next based on my degree, are there open seats in the sections I need. These were reasonable questions for a student advising platform to field, and reasonable to assume the tool already in place could handle them. The problem was where the answers lived. Transfer equivalency data sat at the state level, outside the advising platform entirely, and the course catalog synced over from PeopleSoft in nightly batches rather than real time. Seat availability wasn’t something the tool was ever built to check at all.
Solving it meant reaching into the systems that held those answers: the state’s transfer database, PeopleSoft’s live course data, real seat counts as they changed through the day. That’s not a fix you make to a single-purpose tool. It’s a different category of AI.
That advising assistant wasn’t broken. It was just answering from inside one room, while the answers needed were scattered across several others.
Category 3: The AI Orchestrator
What that community college needed wasn’t a better advising tool. It needed something that could reach across all those rooms at once. That’s where AI orchestration comes in, and that’s what we deployed for that client, using our Ida AI platform to pull answers from the systems that held them. Orchestration needs like that come up consistently at enterprise scale.
Andrew Bediz, who leads our AI and UX practice, likes to put it this way. “Imagine you had one question about your organization, and the only way to get it answered was to schedule a separate meeting with each department head, one at a time. HR this week, Finance next week, IT the week after. By the time you’d gathered everyone’s piece, you’d have spent more time reconciling answers and connecting dots than solving the problem.”
That’s effectively what built-in platform AI and single-purpose tools give you, each working alone. Each one gives you a confident, well-informed answer about its own corner of the organization. A whole category of orchestration tools now exists specifically to sit on top of your various systems and bring those answers together, and that’s a meaningful step forward. As Andrew puts it, “The most valuable AI use cases are connecting the different enterprise dots. AI inside one system that only understands what’s going on in that system is nice. It’s not a game changer. AI that can talk to your CRM and your financial system and your HCM system and your document management system? That’s game changing.”
Going back to the meeting analogy, an AI orchestration tool theoretically gets everyone in the meeting at the same time. But getting people in a room doesn’t automatically make a meeting productive, and having an orchestration platform doesn’t get the work done on its own. Much of the market is still selling orchestration as if it does.
The question worth asking is whether the tool can also fill the gaps where delivered AI doesn’t exist. Can you deploy custom functionality where you need it, not just connect what’s already there? For organizations with significant custom requirements, or platforms where vendor-delivered AI is limited, that distinction matters a great deal.
Getting your platforms to work together, especially when one of those systems is something like PeopleSoft, with its own non-standard logins, integration points, and years of organization-specific configuration, requires a flexible tool. But it also takes someone who has done it before. A general-purpose orchestrator doesn’t arrive already knowing your platform’s particular quirks. When building in an environment like PeopleSoft, the team matters as much as the tool. It takes people who know both well enough to know where the landmines are.
Which category of AI tool are you actually getting?
Before adopting any AI tool, thinking through which of these three categories it falls into will help you set the right expectations going in. Your built-in ERP features won’t cover every point solution need, and they won’t serve as your orchestrator. Some point solutions are more extendable than others, but the question worth asking is whether they understand your specific systems well enough to work across them safely, securely, and effectively. For an orchestration platform, ask one more question: how well does it handle environments where AI isn’t already built in, PeopleSoft being one of the more common examples, and how much effort does bridging that gap take?
A lot of the disappointment in this space isn’t really about AI underperforming. It’s about expecting a tool built for one job to handle a problem that calls for a different kind of tool.
Telling the difference takes more than reading a feature list. Andrew has spent his entire career in both worlds, AI and PeopleSoft together, going back to before either one was considered exciting. “I worked at PeopleSoft myself, so we understand the platform as well as we understand AI. We understand both, and hardly anybody else does.” That’s PeopleSoft. The same kind of gap shows up on other platforms too, though PeopleSoft’s age and lack of standard integration points make it a particularly wide one to cross. Closing it, wherever you’re starting from, is the actual work. We’ve spent the last decade in exactly that space, understanding both sides well enough to know where AI genuinely helps and where it doesn’t.
We’ll get into what it costs an organization to get this wrong in a later post. For now, if you’re trying to sort out which category of AI you’ve been sold, or which one your problem calls for, that’s the kind of conversation we like to have, wherever you are in your AI journey.
This is the second post in our five-part series, Steady in the Storm: What We’ve Learned About AI, ERP, and Getting It Right. The next post, Why Most AI Initiatives Stall, publishes Wednesday, September 2.
