How to use AI to capture your maintenance team’s knowledge and turn it into action

Your maintenance team is only as effective as its most knowledgeable technician. But what happens when that technician leaves? 

When today’s maintenance managers look to the future, many of them see obstacles to keeping that knowledge in-house. Maintenance continues to face a looming talent crisis, and the numbers are worrying:

• 30% of maintenance leaders cite a skilled labor shortage as a top challenge.
• As much as 40% of the maintenance workforce could retire over the next five years.
• 1.9 million manufacturing jobs could remain unfilled by 2033 due to the skills gap (Deloitte).

While maintenance leaders know this is an important and pressing issue, it’s also one that introduces logistical challenges. Moving all of this knowledge out of work orders and the head of a technician takes time and effort. 

The good news: AI can help. We can use it to capture that valuable knowledge so that it doesn’t walk out the door when your best technicians retire.

In this article, we’ll show you how to properly document, organize, and build AI prompts based on that knowledge to help your team make informed, data-backed decisions to make sure everyone on your team can act like a veteran tech, now and in the future. 

For a complete playbook on using AI in maintenance, Download the AI Quick Start Guide 

Key Takeaways

• Capturing tribal knowledge from veteran technicians starts with collecting data. Organized, consistent, and clean data sets you up to produce useful outputs.
• Constructing an AI prompt takes some skill. Including details like data sources and ranges, timelines, audience, and desired output formats will ensure your results are insightful and actionable.
• Tracking metrics such as knowledge capture % and MTTR can help you measure the success of the changes you put in place using AI insights. 

How to prepare your data

Knowledge is data. And as we’ve said many times before, data is the fuel that makes AI run well (or makes it go up in flames—bad data infrastructure is to blame for 20% of AI project failures). 

The specific data you need to collect depends on your operation and the type of knowledge you’re trying to capture, but typically includes things like:

How to clean and organize your data

Unless your systems are already streamlined and automated, simply collecting data as-is isn’t enough. You need to clean and organize data properly in order to produce helpful results. It’s worth it to take each of the data types listed above through a checklist to ensure it’s in its most useful form.

Data organization checklist

• Standardize fields to ensure consistency across all files. Does each asset have its own asset ID? Are failure codes organized according to a consistent and legible code? It takes time, but ensuring there’s one unified language across files will increase your AI mileage.

• Use pre-set options wherever possible. In other words, don’t fix what isn’t broken. This applies to values like failure codes, equipment name, spare parts IDs, and technician names, which can be configured on the backend to appear as a dropdown menu.

• Split complex fields into subsections. This applies to data like work order notes. Entering a wall of text is far less helpful than following a template with mandatory prompts/fields, such as “Issue/Category/Description.”

• Fix typos, shorthand, or inconsistent language. Your technicians might know what an obscure acronym means, but an AI assistant won’t. Making sure your data is legible to a person outside of the organization will also ensure an AI system can correctly contextualize it. 

How to construct prompts

Once you’ve captured, cleaned, and properly organized your data, you can get to the fun part: constructing AI prompts. 

It takes a certain level of knowhow to construct an AI prompt that produces useful results. Here are some key elements you should include when you’re building one:

• Define the data source and range. Instead of entering something like, “Why is downtime increasing?” be specific. Example: Analyzing work orders from the past 30 days, do you see any trends that are impacting downtime?

• Ask for a specific output and format. Think about how your most actionable data and insights are normally served to you, then ask for that. Example: Summarize roadblocks and recommend fixes in a chart with impact, effort, and feasibility levels for each.

• Identify who the output is for. C-suite leaders and line technicians need different data points to do their jobs effectively. If you have an intended audience in mind, say so. Example: Use technician notes to create an onboarding guide for junior techs.

• Specify the context or patterns to look for. Are you curious about a specific trend? Want to dive into a particular failure code? Spell it out. Example: Flag comments about safety concerns.

Sample prompts

Once you’ve mastered the art of building prompts, you’ll be able to ask your AI assistant specific, context-rich questions—just like you would with your most seasoned technician. 

Here are some examples of prompts you can feed into an AI system that, armed with good data, will produce actionable insights.

1. “Compare official PM procedures for [ASSET] with work order and meeting notes from the last three months. Identify steps that aren’t in the procedure, but are frequently completed by technicians. Recommend updates to procedures that better align with real-world conditions.”

2. “Review work order notes, meeting notes, and RCA logs from the last six months. Extract patterns of technician workarounds, observations, or undocumented fixes for standard procedures. Summarize what was done, the associated asset, and the trigger.”

3. “Review work order notes, RCA reports, and meeting notes from the last three months across all sites. Identify major obstacles and frustrations for technicians. Recommend solutions and prioritize them with implementation timelines for each.”

4. “Analyze technician notes and team meeting transcripts related to [ASSET TYPE] over the past 90 days. Extract recurring language, terminology, and troubleshooting patterns. Use this to create a draft training module for junior technicians that reflects proven in-field strategies.”

5. “Compare the average time to complete [TASK] across similar assets. Identify outliers and investigate root causes using work order notes and technician feedback. Recommend adjustments to the procedure or training to reduce variability and improve efficiency.”

AI in action: From prompt to proposal

Now that we know the ingredients an AI assistant needs to produce meaningful outputs, let’s take a look at an in-depth example of AI in action. 

In this example, we’ll show how an AI assistant takes a month’s worth of notes from maintenance team meetings and, with a good prompt, uses them to spot trends and identify where maintenance leadership can remove obstacles to their team. 

The prompt

The results

‍

Using these outputs, a maintenance manager could:

• Create a meeting agenda based on recent work and roadblocks.

• Build a knowledge base for onboarding and training new technicians.

• Summarize work orders to create shift changeover notes.

• Update procedures to reflect real-world actions of technicians and equipment performance.

• Find and address common roadblocks for technicians.

In this example, the outputs are useful because the inputs are organized, thoughtful, and based on existing knowledge. This is not a “garbage in, garbage out” scenario—it’s a case of giving an AI assistant exactly what it needs to support the expertise and instincts your seasoned technicians have carefully built over the years.

How to measure the impact of AI insights

Once you take AI outputs out of the system and onto the shop floor, how do you measure success? The same way you’d measure the success of any new maintenance initiative. 

If you’re testing out new processes recommended by AI, you could track and measure results such as:

• Knowledge capture (% of work orders initiated with notes, % of procedures updated)

• Time to onboard new technicians

• Mean Time to Repair (MTTR)

• First-time repair rates

• Inventory usage/costs

Any or all of these metrics can show you whether your new processes are making an impact on maintenance operations over time. The more you measure the impact of your outputs, the more easily you’ll recognize when your AI assistant has made a great recommendation. 

A final note on AI and worker knowledge

Useful AI systems will never replace your best employees. You’ll always need skilled, informed workers to ensure the inputs and outputs of AI systems will benefit the organization as a whole. But when you can train your AI systems to complement and support that knowledge, you’ll be in a much better position as you navigate the labor challenges that face the maintenance industry.

Want to learn more about how AI can transform your maintenance operations? Download the AI Quick Start Guide here.