ITSM automation: what to automate first, and what the wrong tier costs

The tool is rarely what stalls the programme

Service desk programmes stall on the records the workflow reads, not on the workflow engine. The catalogue lists services nobody requests, the asset register holds rows nobody trusts, and onboarding still starts with an email to the desk. A workflow executes bad data at machine speed, and the queue grows faster than the business case did.

ITIL 4 treats service request management as a practice sitting on top of a service catalogue, and the tools model requests the same way: each catalogue entry becomes a form with its own fields and approvals. A vague catalogue entry produces a vague automated form, and a person still reads every ticket. The first deliverable is therefore not a workflow. It is a short list of what people ask for, in their words, sorted by volume.

Three records have to be right first

The catalogue needs to reflect real demand rather than the last audit. Pull three to six months of tickets, map them to catalogue entries, and retire the entries nobody raises. Each surviving entry needs a description an employee would recognise, required fields, and a named owner. That document becomes the data model for every workflow that follows.

The asset and licence list needs to be accurate enough to act on, which is narrower than a complete configuration management database. It needs the subset of hardware, software and subscriptions the automation reads or writes, each with an owner and an authoritative source. If it cannot say whether a laptop is assigned or a licence is free, the request goes back to a person.

The identity provider needs to own joiners, movers and leavers, with HR as the system of record. Entra ID Governance and comparable products express entitlements as access packages, assignments and approval policies, and grant or revoke them on a lifecycle schedule. The automation can then ask a directory that already knows who has left, instead of a spreadsheet that nobody has updated since the last leaver.

The order that survives the queue

Sequence by dependency, not by pain. Classification and routing come first because request type, category and priority are the inputs to everything downstream, and better routing shortens handling time even when nothing else is automated. Fulfilment of the ten most requested catalogue items comes next: the volume is there, the workflow is short, and saved minutes multiply by a large number.

What the licence actually adds up to

Atlassian publishes the numbers that decide the edition. Jira Service Management is free for three agents, Standard costs $21.50 per agent a month on monthly billing, and Premium costs $55.28; Atlassian says annual billing saves up to 17 percent. A 20-agent desk therefore pays $430 a month on Standard or $1,105.60 on Premium, a difference of $675.60 a month, or $8,107.20 a year.

Usage charges sit on top and move with the estate rather than the headcount. Premium and Enterprise include 50,000 and 500,000 Assets objects respectively, and objects above the allowance cost $0.02 each per month. A 60,000-object estate on Premium pays for 10,000 objects, $200 a month, which is the kind of line that appears only after the tier is signed. Virtual service agent conversations include 1,000 a month, then cost $0.30 each; 5,000 conversations a month adds $1,200.

The practical point is that the edition decision is a data decision. Buy Premium for the asset model only if the estate actually exceeds the Standard allowance and the extra asset management changes how the desk works. Buying it for a workflow engine that the catalogue cannot feed adds cost without removing a ticket.

The metrics that move, and the ones that lie

Deflection is the most abused number in this market. It counts contacts that did not reach a person, not whether the problem went away, and it rises whenever the self-service path is hard to escape. Reopened tickets, repeat contacts and the same user asking twice in a day sit underneath the figure and rarely appear beside it.

The measures that hold up are first-contact resolution against ticket volume, mean time to resolve split by priority rather than blended, the reopen rate, the repeat-contact rate and backlog ageing. A blended mean is dominated by cheap tickets, so a real improvement in P1 handling disappears inside a flat average. Satisfaction surveys mislead in a specific direction too: they are answered disproportionately by people whose ticket went well and quickly, so automating easy tickets lifts the score while the hard queue worsens.

The decision rule before the first workflow

Choose classification and routing first if the category tree is dirty and the desk argues about queues. It is cheap, it produces evidence in weeks, and it makes the next workflow easier to measure. Choose catalogue fulfilment first if the ten most common requests are already clean and high volume; the saved minutes are larger and the workflow is still short.

Do not start with change management, however much it hurts. It is low volume, its inputs arrive from other teams, and its exceptions are the part you cannot automate. A workflow that fails on its third run teaches the organisation that automation does not work, and every later business case pays for that lesson.

Choose Standard if the desk needs forms, approvals and a catalogue and the asset estate fits the included allowance. Choose Premium only when the asset model or the advanced reporting is doing work that the service desk would otherwise do by hand. In both cases, keep approval with a named person: anything that grants access, changes a permission or closes an incident without a human decision puts the tool in the approval path, and the record should show a name rather than a queue.