AI won't fix your service management problems

In 2019 I wrote about organizations that replace their IT service management (ITSM) tool every four or five years, expecting the new tool to solve their problems. It rarely does. Often they could make better use of what they already have, but buying something new seems more attractive.

I'm beginning to wonder whether the same thing is happening with artificial intelligence (AI). AI can do some impressive things. It can help analysts find information, recognize patterns in incidents, and deal with routine requests. But I'm concerned about organizations that see AI as an alternative to understanding and improving their services.

If your processes are unnecessarily complicated, your knowledge is unreliable, and your teams don't work well together, adding AI won't make these problems disappear.

Are you just doing the wrong things faster?

Imagine an organization where it takes five working days to give a new employee access to the applications they need. Their manager completes a complicated form, three people approve the request, and several IT teams each do their bit. Much of the time is spent waiting between teams.

The IT department introduces an AI assistant to help managers complete forms and send requests to the correct teams. Forms are completed faster and the service desk saves time. Unfortunately, it still takes five days before the employee can do their job.

What if the organization had looked at the whole value stream first? Why are three approvals necessary for standard access? Why does the request have to pass through so many queues? Is the risk really high enough to justify all these controls? Perhaps standard access could be agreed in advance and provided automatically, with extra checks for unusual requests. That might remove days of delay without using AI at all.

One of the ITIL guiding principles is Optimize and automate. The order matters. As I've said before, you should optimize the work BEFORE you automate it, otherwise you may simply be doing the wrong things faster.

What about knowledge?

I think one of the best opportunities for AI is helping people use knowledge. But knowledge management isn't just about putting articles in a repository. People also learn from colleagues, from experience, and from understanding their customers. An answer that is technically correct may still be no use to someone in a particular situation.

Suppose your knowledge repository contains several different instructions for resolving an incident, including one that became obsolete two years ago. AI may turn those instructions into a very clear and convincing answer. That doesn't make it the right answer. In fact, its confidence may make it harder for a user or an inexperienced analyst to notice a mistake.

AI may help identify outdated articles or gaps in knowledge. But somebody still needs to decide which information can be trusted, and to understand when a standard answer isn't appropriate. A new search interface isn't a substitute for managing knowledge properly.

Where could AI make a real difference?

Now consider a service desk that struggles with incidents affecting several applications. The analysts do a reasonable job, but information is scattered across incident records, change records and monitoring tools. They sometimes miss connections between incidents because nobody has time to read everything.

The organization starts using AI to collect relevant information when a new incident is logged. It suggests similar incidents, recent changes and relevant knowledge. The analyst checks these suggestions, talks to the user and decides what to do next.

The problem management team also uses AI to look for patterns across thousands of records. This may help it identify problems earlier, but a pattern isn't necessarily a cause. Most problems have several contributing factors, and finding them still needs investigation and judgement.

Your service desk probably doesn't have perfect data or processes, and it doesn't need them. You can start with one useful application of AI, check the results, improve the information it relies on, and then decide what to try next.

I would expect this approach to be much more useful than putting an AI chatbot in front of every user simply to reduce the number of calls reaching the service desk.

How should you get started?

I'd suggest applying a simple three-question test to every proposed use of AI.

1. What are we trying to improve? Describe the outcome for your customers or users. Perhaps employees need access sooner, incidents need to be resolved faster, or analysts need help to find the information they need. Decide how you'll know whether things have improved. If you can't describe the improvement without mentioning AI, I'd question whether you've understood the requirement.
2. Why isn't it working well now? Talk to people who use the service and people who support it. Look at the end-to-end value stream, not just the part you think needs fixing. Find out what really happens, not what your process documentation says should happen. Look for unnecessary work, delays between teams, missing knowledge and badly defined responsibilities. You may find that the most effective improvement doesn't need any new technology at all.
3. Why do we think AI is the best solution? Consider simpler alternatives, including changing the process or using conventional automation. If AI offers something these alternatives can't, ask what it will cost and what might go wrong. Your risk appetite for AI recommending a knowledge article should be very different from your appetite for AI making changes to a production system.

You can't simply copy the answers from another organization. Your circumstances will be different. Use the questions to guide a discussion, not to create another approval form!

If you decide AI is worth trying, start small. Involve the people who will use it, agree what success looks like, and learn from what happens. Check for incorrect advice, leaks of confidential information and cases where a person needs to intervene.

And be careful with your measurements. A chatbot that claims to resolve 80% of enquiries may look successful, but not if users still can't get help. As I've written before, your metrics are not your goals. Find out whether the service is actually better for the people who depend on it.

Don't forget the people

I particularly dislike AI systems that make it difficult for users to reach someone who can help. Sometimes a person needs to explain an unusual situation, or a service desk analyst needs the freedom to depart from the normal procedure to meet an urgent business need. An AI agent can be useful, but it shouldn't become an impenetrable barrier between users and support staff.

The people doing the work often know where AI could help, and where it could cause problems. Involve them in designing and testing it. Help them develop their skills and use their judgement, rather than simply telling them that a tool will now do part of their job.

Summary

I'm enthusiastic about what AI could help us achieve, but I think we need to be clear about why we're using it. A tool that helps people make better decisions or removes genuinely unnecessary work may be worth investing in. One that disguises poor knowledge, bureaucratic processes or unhelpful measures probably isn't.

So before you buy an AI solution, perhaps you should take another look at the problem you're hoping it will solve.

And finally

This blog was written almost entirely by AI. I used ChatGPT, Gemini and Claude to analyse my writing style, create drafts, review each other's suggestions, and help me improve the wording. BUT I am responsible for the content. I reviewed the drafts, decided what I agreed with, rejected things that I didn't like, and made changes until I was happy with the result.

I think this is a good example of how AI can help us work more efficiently without taking away our responsibility for what we produce. The AI tools did most of the work, but I made the decisions, and I take responsibility for the result.

As always, if you try any of these ideas, please let me know how well they work for you.

 

Image credit

Photo by Kier in Sight Archives on Unsplash 

 

This work is licensed under CC BY-SA 4.0