As of
Should every service request be a ticket?
Every service request needs a record. Only the requests a person has to work need a ticket, and an AI agent's record can be a better audit trail than the ticket it replaces.
For as long as most service teams can remember, the rule has been short: if it isn't in a ticket, it didn't happen. The ticket was the unit of work, the thing the dashboard counted and the thing the auditor sampled. It started in IT, and HR, Facilities, Finance and Legal adopted it as their requests moved onto the same kind of desk.
That rule made sense when people did every step. It gets strange when an AI agent does the step. A request comes in through Slack, the agent checks the policy, grants the access and replies, all in under a minute. Teams that keep the old rule open a ticket for it anyway, assign it to the agent, and close it a few seconds later. The queue fills with tickets no person ever touched.
The question isn't whether to keep a record. You need one for every request, whichever team it goes to. The question is whether a ticket is the right shape for a record of work nobody had to pick up.
Why tickets became the default
Tickets solved a real problem, and they still solve it well. Service work passes between people: the person asking, the analyst or coordinator, the approver, the system owner. A ticket gives that work a place to wait, an owner while it waits, and a history once it's done.
That's three useful things. The queue shows what's outstanding. Ownership means somebody is accountable when a request stalls. The history lets the next person see what was tried, and lets an auditor see that a change was asked for and approved.
So the ticket also became the audit trail, almost by accident. It was the one object that existed for every piece of work, and it had a requester, timestamps and a comment thread. When an auditor asked who approved a change, the answer was in a ticket.
Should every service request be a ticket?
No: every service request should have a record, and only requests that need a person to pick them up should be tickets.
The two jobs got bundled because, for people, they were the same object. A request that an AI agent resolves within written policy doesn't need a queue, an owner or a hand-off, whether it's an app license, a letter from HR or a desk booking. It still needs to be accountable: someone has to be able to see later what happened and why it was allowed.
That second job is where a ticket is weakest. A ticket is a form a person fills in, so it holds what someone chose to write. An agent's record can hold what the agent actually used.
What an AI agent's record has to hold
For a request the agent resolves on its own, a complete record answers the questions an auditor asks, without anyone reconstructing them later.
- Who asked, when, and through which channel.
- Which policy applied, and whether it required an approval and who gave it.
- What changed, in which system, and when.
- The facts the decision depended on, where each came from and when it was observed.
The first three are what a good ticket holds on a good day. The last one is what tickets almost never hold, and it's the one auditors end up asking about.
A sales operations manager asks in Slack for a Salesforce Sales Cloud license in March, and the agent grants it under the team's policy. By November she works in finance, and the SOC 2 auditor samples her grant. A ticket that says "granted per policy" sends someone to today's HR system, which shows a finance employee holding a sales license, and the afternoon goes on rebuilding March from exports and old threads.
"She was in sales operations" is a fact with a source (Workday) and a time (March, before the move). A record that keeps that fact as it was observed answers the auditor's question in one look. A record that points at the current HR system answers a different question: what is true today. Nothing went wrong in March; the grant was correct. The ticket just recorded what was typed into it, and not what was true when the agent acted.
What still deserves a ticket
Tickets are for people's work, and plenty of work still needs people. Incidents need someone to own them until service is back. Exceptions need judgment: a request the policy doesn't cover, an approver who says no, a contractor who needs production access over a weekend. Anything that waits on a person, or passes between people, belongs in a queue with an owner.
The same line runs through Legal. A mutual NDA on your standard template, which Legal's policy says needs no review, gets a record: who asked, which template, which rule allowed it. A counterparty's redline to the liability clause gets a ticket, because a lawyer has to read it.
So the useful line is simple. If the agent can resolve it within policy, it gets a record. If a person has to pick it up, it gets a ticket, and the ticket should arrive with the context already attached so nobody starts by asking the requester who they are.
This also keeps the queue honest. When every resolved request becomes a ticket, the queue mixes work done with work waiting, and the real backlog is harder to see. Your team should be able to open the queue and find only things that need them.
Measuring workload without counting tickets
Ticket counts were always a proxy for work. They stop being a good one when a large part of the volume is closed by an agent in seconds.
Report the two kinds separately. Count what the agent resolved, by team and request type, and what it routed to a person. Then look at each team's queue the way you always have: time to first response, time to resolution, what's aging. A rising count of agent-resolved requests with a flat person queue means the agent is taking on routine work. A person queue full of one request type is a policy someone should write.
None of this means employees lose track of their requests. They still need one place to see what they asked for and where it stands, whether it was a ticket or not.
How STLabs records what an AI agent did
STLabs keeps one searchable log of every AI decision, workflow run and configuration change. A request the agent resolves is in that log with who asked, the policy that applied and what changed. When a request needs a person, it becomes a ticket in the service desk, arriving with the requester's identity, device, access and recent changes attached, and every resolution shows the reasoning behind it. Teams start in IT and can extend to HR, Facilities, Finance and Legal without adding another tool, so a Legal approval gets the same audit trail as an access grant.
Three parts do the work. Policies & Approvals is where your team writes what's allowed: a policy can allow a request, block it or require approval, and the workflow waits for the decision before it acts. Workflows carry out the change across your connected tools, with a record of every run. And Axiom™, the live model of your organization built from HR, identity, device and app systems, supplies the facts.
That last part is what makes the record more than a ticket with better formatting. Every fact in Axiom™ records which system it came from, when it was observed and how confident Axiom™ is in it, so any answer can be traced back to its source. This is the AI-native part: the facts and the action live in the same place, so the record of one includes the other.
When an auditor samples a grant from months ago, the record shows the request, the policy that applied, the change it made, and the fact the policy turned on, with the system it came from and when it was observed. Nobody reconstructs anything. For access in particular, STLabs also produces audit-ready access review reports for SOC 2 and other audits.
How to test this on your own systems
Pick a request one of your teams resolves the same way every time, such as a standard license. Find one granted six months ago, to someone who has since changed teams.
Now answer the auditor's three questions from the record alone, without opening the HR system or asking anyone. Why did this person qualify? Which rule allowed it? What did they belong to on the day it was granted?
If the answer means reading today's org chart and guessing backward, your record holds the request but not the reason. Then try the same with a request an agent handled, if you have one, and see whether its record does better than the ticket did.
Record everything, ticket the work
None of this means turning off the service desk you already run. STLabs works alongside it with two-way sync, so tickets that need people stay where your team works them, and teams go from integration to results in five days. The plan is in Moving to STLabs from your current service desk, and ITIL without the paperwork covers how this fits the practices your team already follows.
Keep the ticket for the work that waits on a person. For everything else, keep a record that remembers what was true. A ticket tells you someone asked; a record tells you why the answer was yes.
Frequently asked questions
Should every ITSM service request create a ticket?
No. Every service request needs a record, but only work a person has to pick up needs a ticket. Requests an AI agent resolves within written policy can be logged with who asked, which policy applied and what changed.
Will auditors accept a record instead of a ticket?
Auditors want evidence of who asked, who approved, what changed and when. A searchable record that holds those facts meets the same need. STLabs logs every AI decision, workflow run and configuration change for this.
How do you measure service workload if not every request is a ticket?
Report agent-resolved requests and person-handled tickets separately. Track resolved volume by team and request type, and keep response and resolution times for the queues people work. A recurring request type there suggests a policy to write.
Which service requests still need a ticket in ITSM?
Incidents, exceptions the policy doesn't cover, denied or unusual approvals, and anything that waits on or passes between people, in IT, HR, Facilities, Finance or Legal. Those need a queue, an owner and a history.