AI based problem-solving

A playbook for AI superpower in your structured problem-solving

“Hey, there is a problem, let’s fix this fast and keep on working for our common goal.” I’m pretty sure you have heard that sentence at least once.

Did you also recognize: „Hey, the problem from 2, 3, 4, x months ago is there again, I thought we fixed it? Why can’t we get this under control?“

AI generated, Solutions Workshop

---

Fixing vs. understanding problems

There is an interesting debate in the industry going on, if we shall move fast and release often to reduce the number of issues in our solutions or if we shall work with a quality focus to avoid known problems and minimise the effect of reoccurring issues.

There are two different work styles and workflows fighting for the priority of your management or facilitation. You also might know this by terms like “technical debt” or “lessons learned” processes. All of them root back to the structured problem-solving (SPS) approach that tries to give guidance on how problems can be analyzed in a way, that we as humans understand them nicely. That an organization can learn about it and may adapt their guidelines, processes and rulebooks to avoid these problems happening again.

Structured problem-solving as a method to avoid reoccurring issues

The fundamental idea is something that is nicely summarized in a so called A3 canvas. It is a typical Lean Management tool that helps to structure your thought process towards the relevant focus. So if someone asks you: “What is the problem?” you can answer with the A3 in mind to give a structured answer:

1. You clarify the problem with explaining what it is and what not.

2. In your breakdown of the problem you explain the related processes, workflows and other topics that might have an impact on the problems cause.

3. Defining target KPIs and their expected values, e.g., throughput or failure rates, you create a measure to validate your changes and observe the long-term stability of your system

4. With using techniques such as 5 Why Method you analyze the problem and try to identify the root cause of the problem, not just the obvious one however the more hidden causes.

5. Creating countermeasures is your design and validate cycle to get to a more optimal solution. The countermeasures might not work, they are your action list in the beginning.

6. You implement your countermeasures and check against the defined target KPIs. Typically, the list of countermeasures is prioritized and as soon as your targets are achieved, you can stop implementing further measures.

7. The constant monitoring of the target KPIs help you to ensure the problem is really solved

8. Last but not least, you are expected to change the underlying documentation, processes, methods, tools, whatsoever to ensure that the learning of the A3 is fully integrated, and the problem will not reoccur in the future.

Sounds easy, right?

In praxis, you need to be organized, not skilled

In fact, it isn‘t that easy as it looks, this is why you typically have a kind of “neutral” coach or leader is involved. You will hear the leader often hear asking: “What are your countermeasures and their status?” While the coach is checking for things like “Are you sure this is why it happened?” Or “Is this a real indicator for that type of outcome, or is it just measuring output?”.

Now there is a strong debate going on if a manager, coach or any other expert is needed to run such a process. The more fundamental topic in my honest opinion is the question: What are you doing with the learning/result?

Measure the outcome, not the output

The learning of such an A3 or any type of inspect and adapt, lessons learned, retrospective or validation flow is the outcome for your organization. It is not the type, format or method that counts, neither the number of times you did this process. In the end, what counts is the improvement in the financial numbers.

Within the industrialization of products, this is typically easy, as we have a long history of data and learning already. The coach, leader or expert here is just the superpower for the team, to find relevant countermeasures faster. Or better KPIs that have more relevance to the underlying problem. This way you can more easily decide if the number of bugs, defects or the throughput of the line is the relevant measure to prove your solution works.

In engineering, especially in software-based products, the identifiers are totally different. The duration and impact of a changing environment, e.g., a reconfiguration of the production line, is much smaller. And the user feedback is much faster and important. Therefore, other measures, e.g., the time to market might be more relevant. And this also explains the different perspectives on our initial discussion. Almost fixing, or in deep understanding, is a question of the consequence on the outcome.

In my practical experience, the most difficult thing is to share the knowledge and learning. It is really hard to communicate the concepts in a way that other people are willing to learn from them. There are many cultural aspects that prevent either sharing the knowledge or the effort for sharing is just too much in the day-to-day work. Focusing on the outcome of learning and make this as easy as possible for all of your team is crucial for me. For me, it is actually irrelevant how the new learning is achieved, as soon as it is proven to be right, validated and measured: you should make it as easy to share and access, with as little bias’s and a maximum of transparency and accessibility.

So what are the skills needed?

Summarizing all of this, you might think: “Ok, I can do this myself with a little help of my AI friend.”

And I would agree to it. Especially if you can use an AI co-pilot that has access to your knowledge base. The neutral perspective of the coach or manager is easily achievable through a suitable prompt.

The skills of the process are very well documented and known to the AI models. Therefore, it is typically not a big deal if you are not that familiar with their concepts. The methods such as A3 thinking or 5 Whys are logical and simple to follow.

The crucial knowledge is your domain knowledge about existing problem, solution fields, relevant metrics for measuring, implementation techniques. Even here the AI solutions might have hints and advice, together with your skills that should suit you to help.

As said in the end, the crucial outcome is to bring back the learning into the knowledge base. And here again the AI can help nicely. Either by transcribing your A3 canvas into a nicely formatted Markdown document to be stored or supporting you even further. So let’s have a try.

AI superpowers as SPS coach

There are three elements that are relevant for you to become your SPS coach using the AI superpowers.

1. Setting and understanding the AI context

2. Having the AI to help you with the analysis

3. Using the AI to be creative in finding potential solutions

Setting and understanding the AI context

All AI-based chats are based on specific models and the context you use for your interactions with the AI. The model does not need to be super-intelligent or specific. Likely a fast model will serve your needs. It is also not needed to have a specific reasoning model, as your system as the primary reasoning is going to happen with the checks in your own ecosystem.

Please respond as expert and coach, be precise in your answers and if you do not know the answer, do not pretend it instead say it clearly and ask for further guidance.

We are going to run an structured problem solving session with you being the expert and coach guiding me through the process. 

First we will talk about the context of our environment, organization and domain in which the problem occurred. This is to give you more context about the problem that we have and allow you to ask context or domain specific questions.

After the general context is understood and agreed, you will guide me through the following 7 categories of the A3 problem solving canvas.
1. Problem clarification
2. Breakdown of the problem using processes
3. Definition of target KPIs
4. Use 5 Why method to find the root cause
5. Creative process to find counter measures
6. Suggest and validate activities and priorities to implement counter measures

For each category you may:
1. Collect my input and ask further clarification questions
2. Ask for relevant further documentation, descriptions that might be needed
3. Suggest improvements to my input based on your knowledge and further documentation you obtained
4. As soon as I'm confirming that the category is complete, we move to the next one

All my answers and the documentation of the process is going to be documented in a separate document, that is in markdown format and shall be following the guideline of the typical lean approach.


As soon as all 7 categories are completed the structured problem solving A3 canvas is complete. Then you run the result through your knowledge and try to identify further improvement potentials. You create a management summary out of the result and a status summary of the agreed actions and current KPIs comparing them with the target KPIs. All of this is added in as introduction to the documentation.

With this system prompt or context, we give the AI GPT a very specific setup. It is very similar to what your expert or coach already knows. Asking an internal coach would always ensure that the person knows about the general context and the expertise of the coach is about the method and flow of questions.

It is important that you push the AI GPT to research the knowledge graphs it has. At this point, it is best if the AI GPT would be connected to a dedicated, knowledge base of your organization. This way, it would be able to link you better with already existing solutions.

Let’s start, defining the general context

Depending on the system you are using, it might be needed to run the system prompt as a first chat request to your chat. After that, if I try to check with the chat if all is understood using a question like:

Are there any further questions to the structured problem solving method we are going to use or the process flow?
If not let's start to clarify the general context.

Within the general context, describe your environment as good as you can. E.g., it makes a difference if you are within software engineering for a pure software-based business or if your software is going to run on industrialized hardware products.

Be aware that depending on what chat solution you are using, all the information you provide can be used to train the models. You may not disclose any sensible, confidential or similar information according to the regulations and rules of your company. In case you do not know, please check with your data governance regulations.

Running through the categories

As soon as you are done with the definition of the general context, tell the chat so that it can guide you through the results. Again, it will depend a bit on the service you use, if it can summarize the result automatically in a separate document. Most professional solutions can do so already.

Ok we are done with the general context, let's move on to the problem that we like to analyze.

Describing the categories

Running through the various categories is typically not so complicated. Try to be as specific as possible as it will help you to document and use the result more easy. In case you have to avoid internal figures or confidential data, use placeholders for the same datapoints. This way you can replace the information in your final documentation before publishing internally.

Be explicit when you are done with the definition of one category so that the chat can guide you to the next one.

This category is complete, let's move on to the next.

Or

We are done, please summarize the result as defined.

Bringing all together

By using these prompts, you are going through the actual flow of a classic structured problem-solving. The expert and coach is typically your starlings partner to ask the challenging questions. It might be that the AI chat isn‘t that challenging for you, as it has maybe not all the relevant information. However, it can for sure guide you through the thinking process. Best case you can run this session, with multiple of your affected people or a separate person and use the chat only as your guide through the journey.

At the end we recommend that you bring everything together in one document and share it with others in your company. Use the action plan and management summary to discuss with your stakeholders and may be other internal experts.

Enjoy and I would be happy if you can share your experience with this TEP article. What was working good for you, what did you to be improved?