How to Use AI in Lean Operations Without Automating Waste

John R. Olson, Ph.D.

2 hours ago

5 min read

AI in Lean operations means using artificial intelligence to help people understand and improve how work flows. It can flag patterns in production or service data, summarize recurring problems and help teams prioritize where to investigate. Lean methods provide the discipline to verify what is happening, remove waste and test a change before automating it.

The starting point is not a new tool. It is a business problem your team can define, observe and measure. For a production or service company, that might be late orders, inconsistent handoffs, excessive work in process or recurring rework.

Why combine AI and Lean?

Lean makes the process visible and focuses improvement on value to the customer. AI can help a team review information from that process, such as order records, downtime notes or customer inquiries, and point to patterns worth checking. The output is a lead for investigation, not proof of a root cause. AI is most useful in the Measure and Analyze phases of a Lean Six Sigma (DMAIC) project, where people can use it to examine process data and investigate patterns.

Without a clear process and reliable information, a tool can make a poor process faster or make a weak recommendation sound convincing. A team still needs to observe the work, ask why a problem occurs and verify the evidence with the people who do the work. When AI is used as a decision making tool and not the solution it empowers people to move forward!

Start with one bottleneck or recurring delay

Choose a process where the problem has an owner and the team can compare performance before and after a change. Map the main steps, handoffs and queues. Ask operators, supervisors or service staff where work stops, where it returns for correction and what they must work around.

Next, define a measure that matches the problem. For a late order problem, use on-time delivery and order lead time. For rework, track the number of items returned for correction and the reason. Record the period, definitions and data source so the team compares like with like.

Use AI to find questions, then verify them

A practical first use is to group existing delay or downtime notes by theme. An AI tool may identify repeated descriptions that point the team toward a shift, handoff or category of exception. Before acting, check the original records for missing entries, inconsistent labels and errors. Observe the process and ask the people involved whether the pattern fits what they see.

For example, consider a hypothetical manufacturer whose order records show repeated late releases from one production step. An AI tool groups the notes and flags missing material as a common theme. The team checks the underlying records and walks the process. It may find that material availability is the issue, or discover a different cause such as incomplete job information. The example illustrates a method, not a reported client result.

Other useful applications include summarizing recurring service requests, highlighting orders that need human review or helping prepare a list of questions for a Kaizen event. Each output should make the next investigation clearer, not replace it.

Run a small test before scaling

Define the decision. Specify what the team needs to decide and who owns that decision. For example, a planner reviews a flagged order rather than allowing the tool to change the schedule.
Check the inputs. Identify where the data comes from, who can access it and whether it is complete enough for the question. Remove or protect confidential information before using an external tool.
Set a baseline. Record the current measure and the period it covers. Note unusual conditions that could affect the comparison.
Test in a limited area. Use a selected product family, order type or service queue. Have a qualified employee review each recommendation before action.
Review the results. Compare the same measure over a comparable period. Check for unintended effects, such as moving a queue to another department or increasing rework.
Standardize or adjust. If the evidence supports the change, document the new process, train the people involved and continue to monitor it. If not, revise the approach or stop the test.

Where human judgment stays essential

AI output can be incomplete or wrong, even when it sounds confident. Keep people accountable for safety, quality, customer commitments and decisions that affect employees. Set clear limits on what a tool may access and what it may recommend. Do not give an AI tool authority to change a production schedule, supplier order or customer commitment without an appropriate review. A successful test should fit the way work is performed, not add another layer of hidden work.

What should you measure?

Choose measures that reflect the problem rather than the tool's activity. Depending on the process, that could include lead time, on-time delivery, first-pass yield, rework, work in process or time spent resolving exceptions. What is most important is that you are consistent in how you measure the problem. If possible use repeatable and well documented measurements. A sneaky way to use AI is to have it create data collection worksheets and journals.

Where simple AI agents fit

Simple AI agents can help prepare routine work without taking the decision away from the people responsible for it. An agent could draft a reorder request when an inventory alert appears, summarize daily shift or quality reports for a supervisor, or prepare a follow-up when a supplier confirmation is late. A person should review and approve each action before anything is sent or changed. Start with a clearly defined trigger, a reliable source of information and a named reviewer. Check the underlying records and set limits on access, especially when information is confidential. If the draft is incomplete or the alert is misleading, correct the process and its data before giving the agent more work. The goal is a useful handoff, not automatic decisions that hide errors.

Good first AI uses vs. uses to wait on

Good first AI uses	Uses to wait on
Summarize shift or quality notes for review	Make safety or quality decisions without qualified review
Flag inventory exceptions and draft reorder requests	Place supplier orders automatically without approval
Group delay notes to guide investigation	Change production schedules based only on an AI recommendation
Draft supplier follow-ups for a person to check	Make or change customer commitments without review

From our work: Total Process Consulting helps production and service companies improve operations through consulting, training and education.

Frequently asked questions

How can I use AI in a Lean improvement project?

Start with a defined process problem and use AI to surface patterns for your team to investigate. Check the original data and observe the work before deciding what to change.

Can an AI agent place supplier orders on its own?

A safer starting point is to have the agent draft a request for a person to review and approve. The reviewer checks the inventory information and confirms the order before anything is sent.

How do I know whether an AI pilot improved the process?

Compare a relevant process measure with a baseline under comparable conditions. Check for effects on safety, quality and downstream work, and remember that a change in the measure alone does not prove AI caused it.