Anyone who consults on IT operations eventually collects stories. A client who proudly shows a spotless ticket dashboard while the same server falls over every month. A manager who asks for “more process” the day after an outage. An engineer who quietly keeps the real inventory in a personal spreadsheet because nobody trusts the official one. Those stories repeat so reliably that they start to look like laws. Here are eight of them, pulled together around ITOps, the day-to-day function that keeps services running, with notes on what to do when you meet each one.
1. The dashboard is not the discipline
When a team says it “does ITSM”, ask what it records and what it decides. A tool can hold tickets, approvals and clocks beautifully while the habits behind them stay chaotic. The ITOps entry in the Ops Atlas makes the point in a sentence: the platform records the practice, it does not substitute for it. In consulting terms, never let a software demo be the first meeting. Start with the people who take the calls.
2. Requests, incidents and problems are three different jobs
Picture a client with a single queue for everything, whose busiest person spends each day firefighting the same four faults. Splitting the work into requests (routine asks), incidents (restore the service) and problems (find the cause) gives her a reason to protect time for the third kind. The shift is not technical. It is a decision, reachable in a one-hour workshop, to treat repeat failures as a separate stream with its own owner.
3. Not every change deserves a committee
One of the most useful sentences to hand a nervous client is that ITIL does not require a board for every change. Low-risk, rehearsed changes can be pre-authorised. Others are assessed. A few are rushed through, with review afterwards. Research on software delivery that the Atlas cites found no evidence that outside approval cuts failure rates and leans toward peer review. A consultant’s job is to help draw the lines between lanes, not to defend the meeting.
4. Priority needs arithmetic, not volume
A good priority rule is dull. The Atlas mentions a sample in a popular service platform where priority is impact plus urgency minus one. What matters is that two people applying it get the same answer. Pair it with clocks that reflect reality: the Atlas lab sets its P1 at four hours on a 24×7 clock, and its P5 carries no target whatsoever. Teaching clients that “no target” is a legitimate, honest answer is half the battle.
5. Service promises are chains, not numbers
A customer promise of eight hours does not belong to one team. The worked example on the Atlas page builds it from desk triage at thirty minutes, a two-hour network diagnosis step, a four-hour carrier commitment and a ninety-minute cushion at the end. Each segment has an owner, and each owner can see where their part goes wrong. When a client’s targets keep slipping, draw the chain on a whiteboard before touching any tool.
6. A prediction model still needs a base rate
Machine-learning risk scoring for changes is tempting, and some research is promising. A preprint from a bank’s research group, posted in August 2026 and not yet peer reviewed, reported that its model flagged 63 percent of the changes that later caused major incidents, compared with roughly 20 percent for the bank’s questionnaire. Precision was 0.55 on 203 changes, 35 of which caused major incidents. The Atlas adds the sobering base-rate caveat: apply a similar model to 10,000 changes with a one percent failure rate and only around six percent of the flags would be genuine.
Nothing there argues for abandoning such research. It does argue for treating any scorer as a prompt for a human second look, and to ask vendors for precision, not only recall.
7. Silent monitors are the costliest
The event layer, meaning whatever turns raw alerts into incidents, is easy to demote to a side project. The Atlas lists that demotion as an anti-pattern: when the alert path fails quietly, outages sit behind a calm screen. A consultant’s audit should include one blunt test, which is to trigger a harmless alert and watch who gets it. Also check the roadmap of every alerting product in use. One widely used paging product from Atlassian is scheduled to shut down on 5 April 2027, so any client relying on it has a migration to plan.
8. The inventory that nobody trusts
Configuration records fail in small, human ways. In the Atlas health check, a ten-item sample had nine complete records, two stale entries older than sixty days, one orphan and one duplicated serial number. Do not respond with a campaign to capture everything. Ask which decisions the record supports, such as impact checks before a change, and keep only the fields those decisions use.
Pushback you will hear, and replies that work
Every one of these lessons meets resistance, and rehearsing the replies saves energy.
- “We are too small for this.” Often true for the heavy version. The Atlas itself lists heavyweight service management imposed on a team running one service as a mistake. Offer the light version: a shared queue, a priority rule, a change calendar.
- “Our auditors want a committee.” Possibly, and in some regulated settings the law asks for independent approval and after-the-fact sign-off on emergency changes. Check the actual requirement instead of the folklore, and ask a qualified adviser where the stakes are high.
- “We tried this once and it died.” Usually because process was added after every failure until nobody could follow it. Start with one page and prune quarterly.
- “The vendor says the tool does it all.” Ask which decisions the tool makes and which people still make. Vendors themselves disagree on how the terms nest, so insist on definitions in writing.
A story about a spreadsheet
One more picture, an invented composite that tends to land well. Imagine a mid-sized firm whose official inventory lists forty servers while one engineer keeps a private sheet of forty-six, annotated with which ones really matter. When that engineer is away during a hardware fault, the team loses half a day working out what the failed box does. Nobody was negligent. The official tool simply lacked the columns the engineer needed, so the sheet grew beside it.
The remedy is not a ban. Adopt the sheet as the starting record, add owner and dependency columns, and reconcile it monthly with the official tool until the two match. Shadow lists are an honest signal that the formal system is not fit for purpose.
A bonus lesson: know your neighbours
Finally, a note on scope. ITOps stops where live major-incident command, service design, delivery pipelines and endpoint patching begin. Those areas have their own disciplines, and a good consultant knows when to hand over rather than stretch a single framework across everything.
Turning the list into a first workshop
If you have an hour with a client, you can run all eight in a single session:
- Ask them to describe the last outage and sort the work into request, incident and problem.
- List the last ten changes and mark which lane each would use.
- Write the priority rule on a flip chart.
- Draw one customer promise as a chain.
- Trigger an alert test the same week.
Our own notes on how smart automation improves internal workflows show what to automate once the lanes exist, and a companion article on business portals and the communication they improve suggests where the request intake could live.
The biggest lesson of all is that the dull habits win. Teams rarely fail for want of a clever platform; they fail because nobody wrote down who owns the repeat problem. Start there and the rest of ITOps has somewhere to stand.

