A manufacturer planning an ERP replacement faces a sequencing decision: which AI work depends on that replacement, and which questions can be answered before it?
Start with how you want the company to operate as AI becomes part of the business. Then choose an operating decision that supports that direction and identify the records, people and controls it needs. A general complaint that “our data is not ready” is too broad to tell the company what to do next.
Name the decision before the dependency
Choose one decision with an accountable operating owner. What must that person decide? Which employees contribute judgment? Which records support the decision today? What happens when the records disagree?
Suppose a team is considering help with supplier-expediting decisions. This is an illustrative planning example, not a customer result. Purchasing may need an approved order, a current supplier commitment and the production consequence of a delay. The assessment should establish where those facts come from and who can resolve an exception. It should not assume a new ERP will answer those questions by itself.
Separate learning from production
Ask the team to make two lists.
For a bounded evaluation: which approved examples, employee participation, environment and review process are required to test the idea? Have the responsible technical and information owners determine what may be used. A small evaluation still needs explicit boundaries.
For operational use: which current connections, permissions, update rules, exception handling and support arrangements would be required? Who would authorize any production change, and who would remain accountable for the decision?
These lists may expose different dependencies. An evaluation that teaches the company something useful does not establish that a capability is ready for live operations. Equally, an eventual production dependency should be named specifically before it becomes a reason to postpone every learning activity.
For example, a review of approved historical cases might test how consistently the team identifies expediting exceptions. Using a capability in daily operations might require current supplier commitments and a maintained mapping between item records. Have the team establish those dependencies before choosing the sequence.
Put each ERP dependency on one page
For each proposed capability, record:
| Question | Decision to document |
|---|---|
| What would improve? | One operating result and its current baseline. |
| What must we learn first? | A bounded question the evaluation can answer. |
| Which records are needed? | Approved sources, responsible people and known gaps. |
| What truly depends on the ERP change? | The specific missing capability and the stage at which it becomes necessary. |
| What would justify continuing? | Evidence, effort, exceptions and an accountable decision owner. |
A dependency such as “we need an agreed item identifier before joining these records” is actionable. “We need better data” still needs investigation. Neither statement, on its own, proves that the company must replace an entire system before forming its AI plan.
Choose the sequence with your people
The answer may be to fix a record problem first, evaluate a narrow capability before the ERP transition, or wait for a particular system dependency. The purpose of the assessment is to choose that sequence from evidence.
Include the employees who operate the process, the ERP owner and the leaders responsible for the business result. Account for their time, the cost of maintaining temporary work, and the risk of building something the replacement system will make unnecessary.
NIST’s voluntary AI Risk Management Framework also asks organizations to define intended use, business context and system requirements. That supports starting with the decision and its requirements; it does not determine a company’s ERP architecture or certify an AI implementation. NIST AI RMF MAP guidance.
A company-level AI plan should make the relationship between business priorities and system dependencies visible. If your ERP program and AI ideas are competing for attention, bring one concrete decision to a free 20-minute introduction with Dan. We can explore whether a company-level assessment would be useful.
For the broader decision brief, read AI strategy for defense manufacturers. Explore how employees and specialists work together in the Innovation Center.
For the introduction, call Dan at 760.803.7619 at your booked time.