“We should be using AI.”
Maybe it came up in a meeting. Maybe you’ve seen another company using it and wondered whether you’re falling behind. Or perhaps your ERP vendor has introduced AI features, and you’re trying to figure out what they could do for your team.
It’s easy to find tools to try. Knowing where to start takes a little more thought.
In a recent Platform Talks conversation from The ERP Update, Eric Kimberling, founder and CEO of Third Stage Consulting Group, joined Amiee Keenan to explore what businesses should ask before adding AI.
The first question was simple: Why?
Think about the work that keeps coming back to your team’s to-do list. Is someone spending hours gathering data for a report? Entering the same details in several places? Working through a reconciliation by hand?
Pick something specific. “We want to spend less time preparing this report” gives you a place to start.
Then look at why it takes so long. If invoices are waiting because nobody knows who should approve them, the team needs to sort that out first. Adding AI won’t decide who should be responsible.
You don’t need every detail figured out yet. But you should be able to explain what you want to make easier and who it would help.
Start with one task that takes more time or manual work than it should. Keep the test small enough that someone can review the results and catch mistakes before they affect the rest of the business.
For a finance team, that might mean trying AI on one reconciliation or one part of month-end close. You can see how it handles the work before considering anything bigger.
Be careful about what depends on the result. If people will use those numbers to make business decisions, someone who understands them needs to check them.
Before you begin, ask:
There’s no need to involve the whole department in your first attempt. Give a few people time to try it, ask questions, and work through the problems.
Before asking AI to analyze your data, make sure you can trust what you’re giving it. If sales numbers are wrong or inventory records are out of date, AI could give you a convincing answer based on bad data.
Go through it with someone who uses it regularly. Is that spreadsheet still current? Is anything missing? If the numbers don’t match another report, can they explain why?
The people who enter and update the data can also help you understand how those problems started. Maybe an update gets missed, or people enter the same details differently. Fixing that now can save you from cleaning up the same mistakes again next month.
AI may help with some of the cleanup, but someone who knows the data still needs to check its work.
Think about handing a task to a new employee. You’d explain what you need, show them where to find the information, and check how they’re getting on.
AI needs clear instructions, too. Tell it what you want it to do, give it the details it needs, and explain any limits. A vague request leaves more room for assumptions you didn’t intend.
Someone also needs to check its work. Decide who that will be and what they should do when they find a mistake. A detailed request can help, but it doesn’t guarantee a correct answer.
As for the project itself, put someone in charge of keeping the pilot moving and helping when things don’t go as planned. The people testing it need time, too. If everyone is already stretched to thin, then start smaller or wait until there’s enough time for everyone to participate.
The software price is only part of what you’ll spend. Your team also needs time to set it up, learn how to use it, and check its work.
Compare how long the task took before with how long it takes using AI, including any corrections. If AI creates a report in seconds but someone spends an hour fixing it, that hour counts.
Ask the people testing it:
Listen to what they tell you. They may find that AI helps with one part of the task but struggles with another. That gives you something to work on.
If they’re spending more time fixing mistakes than they’re saving, you’ll want to sort that out before more people start using it. You may need to change the approach or decide this particular idea isn’t worth continuing.
Your ERP may already have AI features that could help with the task you’ve chosen. Find out what’s available, then compare it with tools from other providers.
Keeping everything in one system may make some things easier. But ask what would happen if you changed ERP systems later. Could you keep using the AI tools, or would you have to start over?
If your current ERP meets your business needs, you may be able to add AI without replacing it. Ask whether another tool can connect to it, who would help if something goes wrong, and what you’d pay to keep it running.
Have those conversations before committing to either option. You’ll want to understand what you’re taking on beyond the features shown in a demonstration.
Before changing a process, ask the person who does the work to walk you through it. Have them show you the spreadsheets, extra checks, and steps they’ve added along the way.
Ask why they do each one. That spreadsheet might track something missing from a report. An extra check might catch a mistake that has caused problems before.
Once you understand those steps, you can decide what's easier and what still needs to happen.
The people doing the work often know where things get stuck and what would help. Include them early so their experience helps shape the changes.
If your team doesn’t have time to test AI and check its work, it may be better to wait. Use that time to clean up your data and clarify where you need help.
Start with one task you’d like to make easier. Talk to the person who does it, find out what slows them down, and write down what they’d like to change.
Then, at your next software demonstration, ask to see how the tool would handle that task. You’ll have something specific to compare with how the work gets done today.
Watch the replay of the October Platform Talks.