An AI demonstration can make a complicated task look simple. Ask a question, review the answer, and move to the next decision.
Inside a business, the harder questions come first. Which records support that answer? Do they reflect how the company operates? Who will check the result, and can that person act on it?
On The ABCs of ERP & Beyond, Nirav Shah and Peter Nicholson discuss practical questions about AI, implementation, employee skills, information overload, and privacy. Those questions offer a useful starting point for evaluating AI and ERP together.
Key AI and ERP readiness takeaways
- Start with a specific business task and a useful, testable result.
- Check the records and assumptions behind the answer.
- Match the workflow to the people who will run it.
- Define access, review responsibilities, and the next action.
- Measure the whole process, including corrections and follow-up.
1. Choose a task your team can explain
Begin with work employees already need to complete. Where do they spend time gathering information? Which repetitive task delays a decision? What would a better result look like?
Consider a purchasing team preparing to review overdue orders. A possible pilot could produce a draft summary from approved records, showing the order, promised date, latest recorded update, and person responsible for follow-up. This is a hypothetical planning example, not a case study from the episode.
Before selecting a tool, agree what the summary must include and how a buyer will check it. A useful test needs recognizable success and failure, not just an impressive demonstration.
Keep the first task small. Identify who will use the output, what decision it supports, and what remains outside the pilot. That makes it easier to evaluate whether the proposed workflow actually helps.
2. Check the data behind the answer
The podcast returns to the relationship between available information and useful insight. For an ERP team, that means examining the records behind a proposed use case.
Are dates updated consistently? Do employees use the same definitions? Is relevant information recorded in ERP, or does it remain in personal notes and email?
An overdue-order summary may miss important context if buyers record revised supplier dates only in their inboxes. Decide which record represents the current commitment and how updates should reach it.
Assign owners to important fields and document known gaps before testing. Compare the output with source records a knowledgeable employee has already checked. Include incomplete records and exceptions, not only straightforward examples.
The goal is to understand what the available data can support before employees rely on the result.
3. Match the workflow to your team’s capacity
The hosts describe manufacturing requirements that sound sensible until the team considers the effort needed to carry them out: detailed bills of materials, tracking each routing step, and additional inventory movements.
Does the production floor have enough terminals? Can employees record each step without creating a queue? Are material handlers available for the extra movements?
These questions apply to AI-supported processes, too. More detail creates work for someone who must capture, maintain, review, and use it.
Walk through the proposed process with the employees responsible for it. Include busy periods, missing information, and exceptions. Ask what decision each additional step supports.
If a simpler starting workflow fits the operation better, document that choice and what would justify expanding it. A process needs to work with the staff and resources available, not only in a demonstration.
4. Set access boundaries before testing
The episode raises concerns about confidential information and differences between departmental roles. A purchasing agent, salesperson, and shipping employee do not necessarily need access to the same records.
For each proposed AI workflow, establish:
- Which information the task requires and who may access it.
- Where the information will be processed and retained.
- Who will review the configuration and investigate inappropriate outputs.
Review those boundaries with the people responsible for your systems and data before introducing company information to a tool. Use an approved test environment and an appropriately limited data set.
Give employees clear guidance on approved tools, permitted information, and where to ask questions. Access decisions should be part of the plan, rather than something to resolve after a pilot has already exposed information.
5. Train employees to review the result
Access to a tool is not the same as knowing how to use it within a business process.
Training should show employees how to describe the task, provide relevant context, compare the result with source records, and recognize when further review is necessary. Use examples from their actual responsibilities.
Include an incomplete or misleading result. Ask users what they would check, what they would correct, and when they would involve someone else. Make sure the review process is practical enough to use during a busy day.
The hosts also discuss reviewing generated product descriptions and adapting them for the business. The same principle applies more broadly: people retain responsibility for the finished work. Identify who approves an output before it reaches a customer or changes an operational record.
6. Give each finding an owner
More information creates little value if nobody can act on it. The podcast highlights the bandwidth needed to analyze findings and the risk of overwhelming employees with additional information.
Before a pilot begins, name the reviewer, review frequency, and approval authority. Define what makes an issue urgent and where the decision will be recorded.
In the purchasing example, a buyer would check the important exceptions and follow the agreed supplier follow-up process. A summary alone does not complete that work.
Measure preparation, review, correction, and follow-up time together. Record whether the output helps employees make the intended decision. Otherwise, a fast answer can conceal a slower overall process.
7. Evaluate ERP fit with a real scenario
Bring a defined business task to a discussion about Microsoft Dynamics 365 Business Central, Acumatica, or another ERP platform.
Ask the provider to demonstrate what is available for your version and configuration. Clarify required integrations, permissions, costs, ongoing responsibilities, and employee review steps. Include missing information and exceptions in the demonstration.
Use the results to judge whether the approach fits your processes, staff, and priorities. A feature list becomes useful when the team can connect each capability to a business requirement.
Build a foundation your team can use
Start with one process. Map its records, people, and decisions. Identify gaps before expanding the scope, then test whether the proposed improvement makes everyday work easier.
Not sure where AI fits into your ERP processes? Talk with AdCirrus ERP about your data, workflows and practical next steps for Business Central or Acumatica.
Talk to an ERP expert: https://adcirruserp.com/contact/
Watch the full podcast episode
AI & ERP – What does the future look like??