Smart workflows in TOPdesk: what automation solves

A great deal is promised about AI in service management and rather little demonstrated. That is a shame, because there is real ground to gain in a TOPdesk environment today — it just rarely sits where the brochure looks for it. The largest time saving does not come from AI, but from ordinary automation that was never set up.

This article covers both: what rules solve, where AI genuinely adds something, and in which order to tackle it.

First the dull part: rule-based automation

With Events & Actions, TOPdesk has had a capable automation mechanism for years. It works on rules: if this happens, do that. No model, no prediction, fully traceable. In the environments we take over, this is almost always where the most unused potential sits.

What it automates away is exactly the work nobody benefits from:

• Tickets routed straight to the right team by category, instead of the service desk acting as a relay.
• An acknowledgement stating the expected resolution time, so the caller does not ring after two days to ask whether it arrived.
• A warning well before an SLA is about to be breached, rather than at the moment itself.
• Onboarding requests that automatically create a set of tasks: account, hardware, access, workspace.
• Standard changes that move themselves to the next stage once approved.

None of this is exciting, but it is where the hours are. An organisation that has this in order has time left for the work that does need attention.

Start with what happens often, not with what is complicated

The temptation is to start with the most complex process, because that is where the frustration is. Do not. Look in your own figures for the action that occurs most often and requires the least thought. Passing a ticket to the same team ten times a day is a better first candidate than an annual migration.

Where AI does add something

AI is good at tasks for which no watertight rule can be written, because the input is free text. Within service management that comes down to a handful of applications.

Classifying and routing

A caller who picks their own category often picks the wrong one — not out of unwillingness, but because the category tree was built for administrators. A model that predicts from the text of the ticket where it belongs tends to beat the dropdown in practice. Importantly: let it suggest, not silently decide, and track how often the agent corrects the suggestion. If that percentage does not fall, it is not working.

Summarising

A ticket handed over four times and carrying twenty notes costs the fifth agent ten minutes of reading. A summary at the top saves those ten minutes. This is the lowest-risk application: the original stays put, and nobody decides anything on the summary alone.

Suggesting and writing knowledge

Two sides of the same coin. While handling a ticket, a model can surface existing knowledge items that resemble it. And when closing a recurring ticket, it can draft a knowledge article the agent only has to check. That last one removes the biggest barrier to knowledge management: the writing itself.

Draft replies to the caller

A draft the agent adjusts is faster than an empty field. Do watch the tone: a reply phrased too smoothly reads as a form letter and does more harm than a short sentence of your own.

What it does not solve

AI does not make a bad process good. If tickets land with the wrong team because the shape of the organisation no longer matches the shape of TOPdesk, smarter routing does not fix that — it hides it. The same goes for a knowledge base nobody maintains: a model surfacing outdated articles does more damage than no suggestion at all.

There is a second limit, and it is a governance one. As soon as a model decides how something is handled, you have to be able to explain why it went the way it did. With a rule that is trivial; with a model it is not. In healthcare and government that is not a theoretical objection but a question you get at the first review.

A workable order

1. Measure first. Which actions occur most often? Without that figure you automate on gut feeling, and gut feeling usually points the wrong way.
2. Clean up the categories. Automation driven by category is exactly as good as the structure underneath it.
3. Build the rules. Assigning, acknowledging, escalating, creating tasks. This is the bulk of the gain.
4. Apply AI where text is the problem. Category suggestions, summaries, drafts for the knowledge base.
5. Measure again. Not whether it works, but whether it saves anything: resolution time, number of handovers, how often a suggestion is overruled.

In closing

The organisations that get the most out of automation are rarely the ones with the newest features. They are the ones that know what happens in their own environment and tackle one thing at a time. AI changes little about that; it only shifts what is possible at step four.