AI has arrived in your ERP. Are your people ready?

Written by Lyndsey Martin

A change approach designed for one country rarely survives contact with a second one. The system configuration doesn’t change. The go-live date doesn’t move. Yet the same rollout plan can produce strong adoption in one region and quiet non-compliance in another, and the gap almost never traces back to the technology.

It traces back to assumptions built into the rollout that nobody examined, because they didn’t look like assumptions. How formal communication should feel. How much context people need before they’re willing to act. Whether group training builds confidence or shuts people down. How directly a problem gets raised, and to whom. These aren’t preferences at the edge of the project. They sit underneath every training session, every change message, every go-live briefing, and a rollout plan built around one cultural default will consistently underperform wherever that default doesn’t hold.

Most ERP vendors are slowly adding AI to the systems people use every day. Sometimes this comes with a big software release, but other times it appears after an update as a new suggestion, a flagged exception, a forecast, or a new field.

A feature may be ready to use, but that does not always mean people are prepared for the changes it brings.

Release notes can explain what a feature does, but they do not always tell employees how much to trust it, when to question it, or what to do if a recommendation does not make sense in their daily work. This is often where real adoption challenges start.

What AI in ERP changes for employees

For most employees, AI in ERP probably will not appear as a completely new system. Instead, it will likely show up within processes they already use.

An accounts payable clerk might see a suggested GL code next to an invoice line, while a stock controller could notice an anomaly highlighted during a count. The system is no longer waiting for someone to make every decision from the beginning. Now, it recommends, flags, and sometimes even automates parts of the process.

That sounds efficient, and it can be. But it also subtly changes the employee’s role. The question is no longer only, ‘What do I need to do?’ It becomes, ‘Is what the system is suggesting actually right here?’

That is a different kind of responsibility, and it needs a different kind of preparation.

The problem is not only whether people trust AI

A lot of AI adoption work focuses on getting employees comfortable enough to use the technology. That matters, but another risk deserves just as much attention: people trusting the system too quickly.

If employees do not trust an AI feature, they may ignore it, work around it, or keep doing the task manually, which means the investment sits underused.

If they trust it too much, the opposite problem appears. A recommendation may be accepted simply because the system produced it, even when something about the transaction, customer, supplier or stock position should have prompted a closer look.

Neither outcome is helpful. In one case, people don’t use the feature; in the other, they use it without enough judgement.

The best approach isn’t just to tell employees the AI is reliable. Instead, help them understand what the AI bases its recommendations on, its limits, and when human review still matters.

The task often changes before the job does

Many people worry that AI will replace jobs. In ERP settings, though, the change is usually less immediate. The role remains, but parts of the job begin to change.

Manual entry, matching, and routine checks may take less time. Instead, employees may spend more time reviewing exceptions, questioning recommendations, and deciding what needs attention.

Consider an accounts payable clerk who used to enter every invoice line by hand. If the system can now code most lines accurately, the employee will spend less time on data entry and more time reviewing the invoices the system could not process well. The job still exists, but the judgement needed for the role has changed.

The same goes for a planner who used to create forecasts from scratch. If the system now makes the first forecast, the planner’s value is in knowing if the assumptions are reasonable, what the model might have missed, and when to question the numbers.

In many cases, that actually raises the bar. Reviewing a recommendation properly requires a strong understanding of the underlying process, because the employee now needs to spot a mistake they didn’t make themselves.

Training has to go beyond showing people where to click

Traditional ERP training usually covers process and navigation. Which field do you fill in? Which button comes next? What is the approval sequence? This approach works when the system mainly asks users to follow a set transaction path.

With AI involved, employees may need to interpret a confidence score, notice when an output is technically correct but contextually wrong, or know when to double-check a recommendation before accepting it.

This is not just about learning the interface, but rather about building judgement.

The most helpful training will focus on the decisions each role now faces. People need to know what the feature does, when they can trust it, when to question it, and where to go if something seems wrong.

Adding a short AI module to an ERP course might check a training box, but it will not always prepare someone for these real-world decisions.

What to check before switching an AI feature on

Before an AI feature goes live, it is worth asking a few practical questions:

  • Which specific roles will interact with the feature?
  • Can someone explain, in plain language, what the recommendation is based on?
  • Where does human sign-off remain mandatory?
  • What should happen when the system gets it wrong?
  • Who is responsible for spotting and correcting the problem?
  • Have employees been trained on when to question an output, not only how to accept one?
  • If something looks off, do people know exactly who to ask?

A feature might pass every technical test but still not be ready for daily use if those questions are not clearly answered.

Adoption does not arrive with the software update

The technical rollout and the human rollout are not the same thing. One can be completed while the other is still very much in progress.

That is how organisations end up with AI functionality that is live but barely used, or with employees accepting recommendations without the level of scrutiny the process still requires.

The human side needs the same careful approach as any other ERP change. Explain why the feature is being added, train people for the new decisions they will make, make escalation routes clear, and check after go-live whether the feature is being used as planned.

Turning the feature on is usually the easy part. The real challenge is making sure people know how to use it once it is available.

If you found this article helpful, you might also like to read Your Dashboard Is Talking. Is Anyone Listening? and How to Build a Quick Reference Guide Your Team Will Actually Use.