How to Construct AI Prompts for Maintenance Teams

Introducing any new system to a facility (and the team within that facility) always comes with a learning curve. 

Adopting an AI system is no different. It’s not something you can implement today and magically reap rewards from tomorrow. To get real results, you first need to make sure your data is in order. Then, you need to learn how to interact with AI tools to get the best possible results.

In this article, we’ll show you how to construct AI prompts that generate useful, reliable answers that your team can use in their day-to-day work. 

Key takeaways

• Prompts can make the difference between results that are “just okay” and results that truly improve the company’s bottom line.
• Weak AI prompts are vague and lack context. Strong AI prompts reference your exact equipment, problems, desired outcomes, and intended audience to produce the best possible answers.
• Depending on what you’re using your AI system for, you can follow a prompt template to produce reliable results every time.

Why prompting matters

While AI has made vast improvements over the years, it’s still no crystal ball. Just like a technician reading through work instructions, an AI system needs structured, well-thought-out inputs to be able to do its job well. Otherwise, it’s just guessing.

The difference between AI outputs that are passable and ones that actually improve your operations and impact your bottom line often comes down to how well the AI assistant was prompted. So let’s look at how to write a good prompt.

Not all prompts are created equal

To compare weak vs strong prompts, let’s humanize the experience. Imagine a person came up to you and asked, “How many pizzas should I order?” There’s good news here: it sounds like you’re going to get some free pizza. But there’s also a lot of missing information. 

For example: What’s the occasion? Is this person ordering pizza for a team lunch, or just for the two of you? Is it for an afternoon snack, or a full meal? How big are the pizzas in question? Is it part of a larger food order? (Might be nice to throw some wings in there too, no?) We could keep going here, but the point is, the initial question is vague and lacks the necessary context to be able to give a useful answer.

Weak AI prompts suffer from similar problems. Often, a prompt is weak because it’s:

• Vague: If a prompt doesn’t specify which assets, timeframes, failure codes, or circumstances need to be analyzed, it’s tough to get a useful answer in return.
• Starved for context: Just like in the pizza example, if you don’t know the details surrounding the question of how much pizza to order, you could come up with any number of seemingly plausible answers. Without context, your AI system lacks a clear starting point.
• Written without a desired outcome and audience in mind: The only way to get the outcome you want is to ask for it. If you don’t specify what you want—and who will be using it—the results won’t be as useful as they could be for you and your team.

The main ingredients of a strong AI prompt

A strong prompt, on the other hand, sets up your AI system with exactly the right ingredients to give you the output you need. 

Using the example above, it’s like someone coming up to you and saying, “I want to reward our team of seven with a pizza lunch to celebrate a great month. I’m adding three orders of garlic knots, too. Each pizza is 14” in diameter and has six slices. How many pizzas do you think we need?”

Much like this example, a strong AI prompt asks for exactly what it needs, giving the right level of detail, context, and desired result. In other words, a strong AI prompt is:

• Specific: Whenever possible, clearly name the asset, part, issue, metric, and/or task you need information on. 
• Context-rich: Including context such as failure history, a time frame, or observations can help narrow down a problem or question.
• Clear about desired outputs: Stating the format you want for your output or answer is important. If you know you’d like your answer written as a step-by step procedure, a table, or a side-by-side comparison, include that in the prompt. 
• Tailored to its audience: Different audiences need different information in different amounts of detail. Specify who this information is for, and you won’t have to go back and try the prompt a different way.

Repeatable frameworks to get the results you want

Of course, you didn’t come to this article to get advice on how many pizzas to order. So let’s look at some real-life examples of how to construct AI prompts to get the best, most useful answers. 

Example 1: How to prompt AI for repair assistance and troubleshooting

To get help with a repair or to troubleshoot a problem, you’ll need to clearly state which asset is having trouble and what you’re observing, as well as any other useful information. 

In this example, let’s say the fan on your FANUC robot needs to be cleaned, but you’re afraid that doing so will damage the wiring inside. Here’s how you would prompt your AI assistant:

The fan on our FANUC Delta Robot needs to be cleaned. How can I do this without damaging the internal electronics?

More broadly, this type of prompt should normally be structured like so:

The [SUB ASSET] on [ASSET] is [OBSERVED ISSUE]. What steps should I take to inspect and confirm the source of the issue?

Why this prompt works:

• It’s very clear about what needs troubleshooting and why.
• It asks for a clear outcome, in the form of inspection steps to confirm the source of the issue.

Example 2: How to prompt AI to analyze maintenance metrics

To get an AI assistant’s help with analyzing maintenance metrics, you’ll need to be clear about who these metrics are for, what needs to be analyzed, and how you want the analysis to be structured. 

For example, let’s say you have a big meeting with the exec team coming up, and you want to show how your team has reduced downtime month over month. This AI prompt might look like:

I’m presenting to the exec team tomorrow, and I need to summarize maintenance performance using downtime and MTBF. Please take these metrics from the last 12 months and translate them into a concise narrative that ties maintenance performance to overall cost savings for the company.

In a general sense, this type of prompt should be structured like this: 

I’m presenting to [INTENDED AUDIENCE]. Summarize maintenance performance using the following metrics: [METRICS]. Present this analysis in [DESIRED FORMAT] and tie the results to [COMPANY TARGET].

Why this prompt works:

• It states an intended audience right off the bat. 
• It asks for the right measures. Providing the metrics that should be analyzed is key. 
• It asks for a clear output, in the form of a concise narrative that ties maintenance performance to a specific target.

Example 3: How to prompt AI to generate procedures

Sometimes you need to generate a procedure, but you don’t have time to write it out yourself, step-by-step. In this case, you can provide your AI assistant with manuals and past PMs to create procedures for you. 

In this example, let’s say you need to generate a number of inspections for your CNC machine to make sure it’s running smoothly. Here’s what that prompt would look like:

Create a PM program for our CNC machine with daily visual inspections, weekly functional checks, and monthly deep inspections. Pull technical steps from attached manuals and PMs.

The generic version of this prompt would be something like:

Create a PM program for [ASSET] with [DESIRED PROCEDURES]. Pull technical steps from [ATTACHED SOURCE MATERIAL].

Why this prompt works:

• It states what it wants the AI assistant to use as a source. Attaching manuals and PMs is a great way to ensure that the generated procedure is useful for the facility—and asset—in question. 
• It suggests a procedure framework. Asking for specific procedures on a specific timeline ensures that the procedures will fit into existing team operations. 

AI systems can do a lot, but you need to approach them with the right information and intended outcome. Once you understand how AI prompts should be structured, you’ll be able to find new ways to interact with your system and produce interesting results.  

Want to learn more about how to use AI to improve maintenance operations? Download the AI Quick Start Guide.