By now, if you’ve been following this series, you’ve seen the pattern. The challenge with adding AI to an ERP environment usually isn’t ambition. It’s follow-through. Organizations get excited, they run a pilot, and somewhere between the pilot and production, momentum stalls out. We talked through why in the last post.
This post is about the other side of that pattern, what it looks like when an implementation goes well from the start.
We’ve had the benefit of watching this up close, over and over, across more than two decades of PeopleSoft process automation and more than a decade of enterprise AI work built on that foundation. Some of those lessons came from projects that ran exactly as planned. Some came from projects that hit friction and required us to reevaluate and adjust. Both kinds of experience shaped what we now think of as a playbook: not a rigid script, but a set of principles that separate implementations that hit the mark from the ones that fall short.
Define the problem before you choose the technology
It sounds obvious, but it isn’t practiced nearly as often as it should be. Most AI conversations start backward, with a tool or a demo, and only later circle around to what problem it’s meant to solve. The most effective implementations start from the opposite direction. They name the specific operational pain first, whether it’s a backlog of manual approvals, a help desk drowning in repetitive questions, or a process that only one overworked employee fully understands, and only then ask what technology fits.
Invest in your data and content before you invest in the AI
An AI agent is only as good as what it has to work with. Knowledge articles, workflow documentation, historical case data: this is the raw material, and it takes deliberate work to get it in shape. The organizations that treat this as a one-time cleanup project tend to plateau quickly. The ones that treat it as ongoing operational infrastructure, something that gets maintained the way you’d maintain any other system of record, are the ones whose AI keeps getting more useful over time instead of less.
Choose depth over breadth
It’s tempting to try to solve everything at once. Resist it. The implementations that hold up start with one well-defined use case, get it right, prove the value, and expand from there. Depth first, then breadth. Trying to solve everything in phase one is one of the most reliable ways to end up with nothing that works particularly well.
Understand the platform you’re integrating with
This is where a lot of AI initiatives run into trouble. Take something like checking seat availability in a class. On paper, it looks like a simple lookup. In practice, getting the answer right depends on how the platform structures that data. For a multi-campus system like VCCS, that means knowing which college a student belongs to, how their program and plan are recorded, and who’s authorized to see that information. None of that is visible from a generic integration point. It only surfaces once you know the platform’s data model well enough to know where to look and what the relationships mean. Miss one of those pieces, and the result isn’t just a wrong answer. It can mean exposing data that should have stayed protected.
Platform fluency isn’t optional. It’s what separates an implementation that works from one that doesn’t. A lot of our platform fluency was built over two decades of PeopleSoft process automation, long before we applied any of it to AI. But the underlying discipline isn’t PeopleSoft-specific. Knowing a platform’s data model, its business rules, and its edge cases well enough to build AI around it responsibly, that applies whether the platform is PeopleSoft, Oracle Fusion Cloud, or Workday.
Work with a partner who stays engaged
It’s not that organizations can’t do this work on their own. It’s that the pace of change in AI would strain any internal team, however capable, whose core focus isn’t AI. The organizations that succeed most consistently pair their own expertise with a partner who is genuinely invested from the first discovery conversation to the last post-launch challenge, not one who builds in isolation and hands you a finished product at the end. That kind of investment shows up in the results: every Ida implementation we’ve delivered has stayed on budget, and we’ve never been the reason a project’s timeline slipped, though clients sometimes revise their own plans along the way.
That kind of track record doesn’t happen by accident.
It comes down to discipline, not complexity.
Good AI implementation in an ERP environment isn’t a technology story. It’s a methodology story. The same instincts we’ve applied to PeopleSoft process automation for more than two decades are now pointed at a faster-moving problem. The technology matters. The discipline behind it is what makes sure it hits the mark.
This is the fourth post in our five-part series, Steady in the Storm: What We’ve Learned About AI in ERP. The next post, Where We Go From Here, publishes Wednesday, September 30.
