Is GenAI a better process improvement coach than a human?

Some people have hobbies on the weekend. I think about how GenAI performs compared to a human. In this case, I wondered whether GenAI could function as a useful coach for mentoring lean process improvement learners. As a quick reminder, "lean" refers to principles that are used to improve the quality and efficiency of processes, so that products and services can get to the end users faster and without errors. An example would be figuring out how to reduce the wait time at a passport office so travelers can get their documents faster. Lean process improvement methods are particularly useful when it comes to thinking about how to improve government services without adding extra cost or resources. When someone is learning how to improve processes having access to coaching from an experienced lean practitioner is incredibly helpful, but everyone doesn’t have a coach on speed dial. Could GenAI tools like Copilot or ChatGPT be used to create easier and more equitable access to lean coaching?

Using seven recent questions from lean learners, I compared my (human lean coach) responses with two GenAI tools - MS Copilot and a custom, user-generated version of ChatGPT called "Master Black Belt Lean Advisor." For context, a Master Black Belt is the highest level of lean certification. I engineered the prompt for Copilot so that it was also asked to respond from the perspective of a Master Black Belt.

Here is an example of one of the questions and the responses:

Prompt: From the perspective of a Master Black Belt Lean Advisor what's the best way to measure time in a value stream map? Minutes or hours?

Human response: I do hours, minutes, or seconds, depending on what makes sense to the process being analyzed. It depends on how long the end-to-end process is and how long each process step is. I never do "parts of hours" because it's not meaningful to people in the same way. For example, saying .75 of an hour doesn't make sense to people the same way 45 mins does. Similarly, 10,000 minutes doesn't make sense to people in the same way that 6 days, 21 hours, 35 mins does. Our brains have a hard time understanding what hundreds of minutes mean and think that minutes mean "not a lot of time."

Copilot response: Great question! The best way to measure time in a value stream map (VSM) depends on the level of detail you need and the nature of the process you’re analyzing.

Minutes: Use minutes when you need a high level of detail, especially for processes with shorter cycle times. This granularity helps identify small inefficiencies and areas for improvement.

Hours: Use hours for processes with longer cycle times or when a broader overview is sufficient. This can simplify the map and make it easier to communicate with stakeholders.

Ultimately, the choice depends on the specific context of your value stream and the goals of your analysis. Do you have a particular process in mind that you’re mapping?

Master Black Belt Lean Advisor response: In a Value Stream Map (VSM), measuring time can vary based on the level of detail required, the complexity of the process, and the specific goals of the analysis. Generally, the measurement unit should align with the typical cycle time of each process step and the takt time of production:

Minutes: Using minutes is common when processes have shorter cycle times, as this provides a granular view of time spent on each task, waiting periods, and lead times. Minute-level data helps reveal small yet impactful inefficiencies, which can be crucial in Lean environments focused on minimizing every second of waste. This approach is particularly effective for VSMs in manufacturing with high-frequency processes, where detailed adjustments can yield significant cumulative benefits.

Hours: Measuring in hours is more suitable for processes with longer cycle times, such as batch processes, complex assembly, or service industries where tasks may span several hours. In such cases, using hours provides a clear overview without overwhelming the map with overly detailed timing, which may not add significant value to high-level decisions or improvement plans. Best Practice: When deciding, consider the total lead time you aim to reduce and the level of precision needed. For production systems where takt time or cycle time is short (under an hour), minutes will typically give the best actionable insights. However, in systems with longer cycles, hours will suffice and make the VSM more readable and effective for decision-making.

Adjustments based on findings in either minutes or hours can always be converted to finer or coarser units as needed, depending on the improvement focus or analysis depth for each specific Value Stream Map.

The above examples are typical of what I saw in the responses. The human responses were more conversational and practical, with anecdotes from lived experience as a lean practitioner. They were also more succinct, with workplace specific context added. The Copilot responses were more detailed and included questions to prompt further discussion, making it seem like a more interactive chat partner. The Master Black Belt Lean Advisor GPT responses were even more detailed, had technical depth, and included more lean tools and terminology. It was also the most comprehensive, with connections made to other lean concepts and practices beyond what was included in the prompt.

Overall I was surprised at the quality and accuracy of the GenAI responses - with one exception. Both Copilot and the MBB Lean Advisor GPT made one error (aka a "hallucination" in GenAI speak). Both tools suggested that SIPOC diagrams should be used after creating a value stream map. But in practice, lean practitioners use SIPOC as a visual tool at the beginning of projects to understand, (as the acronym suggests) the suppliers, inputs, process, outputs and customers at each step of delivering a product or service. I think of SIPOC as a high level process map. You can turn it vertically and the process column becomes the vertical for your value stream map. I confirmed my human knowledge using the ASQ Certified Six Sigma Black Belt Handbook (2017, p. 146) which states that SIPOC "can be used as a basis for constructing detailed process maps and value stream maps at future dates." I suspect that the GenAI error in response to this question was a result of how the prompt was structured. I copy-pasted all prompts verbatim from lean leaners and in this case the prompt was: For Value Stream Mapping, is that something we do after we get an idea of every stakeholder and their responsibilities? When do we do VSM? Before SIPOC? Because I indicated that VSM is done before SIPOC, I think I led the GenAI to think that was the order I was looking for. In a follow-up question I changed my prompt and asked in what order those two tools are used and this time both provided the correct answer (SIPOC first).

The above examples illustrate two important concepts when working with GenAI - prompt engineering and evaluation. To get the best content from GenAI, it's important to use good prompts. And to understand whether the tools are producing high-quality content, you need to be able to evaluate the output. This is a fair warning for learners who may not be able identify errors when evaluating output, especially because GenAI responses are crafted in a definitive tone, even if the information is inaccurate. Learners can develop skills in structuring prompts to help mitigate against hallucinations, and should consult other resources to verify accuracy. With further development, these tools may function more strongly as lean coaches in the future. I encourage you to test out these and other uses of GenAI on your lean journey, particularly to discover whether you can reduce your processing time on any onerous tasks.

And in case you might be wondering, this article was not written with GenAI!