Support teams usually inherit their operating routine by accident. Someone sets up a mailbox, someone else adds a chat channel, and a year later nobody can say how a request becomes a finished job. This guide gives you eight concrete steps to replace that accident with a light, written ITOps routine. You can complete the whole sequence in two or three working weeks without buying anything, and every step produces a small artefact you can show a client or an auditor.
The structure follows the scope outlined in the Ops Atlas, which sums up the discipline on its ITOps reference page as the group keeping IT services alive: requests, incidents, cause hunting, change control and an estate map. We will build each of those pieces in turn, in a sensible order.
Step 1: Create two intake lanes
Start with a board or a shared sheet that has two clearly marked lanes. The first lane is for requests, which are routine asks with a known answer. The second is for incidents, which are things that worked and stopped. Both lanes need the same minimum fields: reporter, time received, description in the reporter’s words, current owner and status.
Do not try to categorise anything beyond this yet. Over-engineering the form is the most common way support teams stall at step one. Run for a week and see what actually arrives.
Step 2: Write a priority rule you can compute in your head
Pick two scales, one for impact (how many people or how much money is affected) and one for urgency (how quickly harm grows). Combine them with a fixed formula. A sample rule quoted in the Atlas, from one major platform, takes impact plus urgency and subtracts one; on three-point scales that gives five levels. Copy the shape or invent your own, but write it where everyone can see it.
Then attach target times to each level. A reasonable first draft, based on the lab table in the Atlas, ranges from four hours around the clock at the top, through eight business hours and one day, to a lowest level with no target. Adjust to your contract, and note which hours count.
Step 3: Define your clocks in writing
For every target time, answer three questions: when does the timer begin, what stops it, and which days does it skip? Waiting for a customer’s reply is a typical pause. Weekends and holidays are another. The reason to settle this now is shown by an Atlas example in which a ticket stayed open 92.5 hours on the wall clock but consumed just 4.5 hours of its allowance.
If you promise a response time to a customer, break it down. The Atlas illustrates an eight-hour promise as thirty minutes of triage, two hours of diagnosis, four hours from a carrier and ninety minutes of buffer. Your own breakdown might have only two or three pieces. Add them up and make sure the total is at or under the promise.
Step 4: Split changes into three routes
Draw three boxes. In the first, list routine changes that you will pre-approve, such as creating a user in the standard way or restarting a known service. In the second, put everything that needs assessment. In the third, define an emergency route that lets you act first and document afterwards, with a mandatory review within a few days.
Give each pre-approved recipe a failure limit. The Atlas example pauses a recipe once it has failed more than once in its last twenty uses. This tripwire stops a shortcut from quietly becoming a habit. It costs one extra column in your list and prevents long arguments later.
Step 5: Add a three-factor risk score
For the second box, a short score decides who looks at a change. Choose three factors, such as the tier of the affected system, whether data structures are altered and whether the work falls outside the agreed maintenance window. Give each a number and set the threshold for a second review. In the Atlas worked example, the factors scored 3, 2 and 1, and the total of 6 triggered board review.
Add one shared calendar of planned work. Look for overlaps every time you add an entry, and block known freeze periods, such as a payroll cut-off. The Atlas describes a lab where a freeze from 26 to 31 October 2026 flags any clashing item.
Step 6: Handle automated changes carefully
If you use pipelines or scripts that raise changes, a script may open the change record, yet only a person or a policy rule may sign it off. The Atlas is firm on that split, which protects independence. If your platform uses roles to guard its change interface, grant the narrowest one that works. For one popular platform, the Atlas notes that certain API operations accept three different roles and advises using only the middle-tier `itil` role rather than higher ones.
For practical context on how automation fits into delivery habits, this overview of principles behind effective software delivery pairs well with the record-keeping rule above.
Step 7: Build a small estate register and keep it honest
Create a list with six columns: item, owner, where it is hosted, what depends on it, renewal or expiry date, and last confirmed. Start with the ten things whose failure would hurt most. Each month, confirm a few rows and remove anything nobody can vouch for.
The Atlas supplies a rot gauge, applied to ten sample records, that tests completeness, staleness beyond sixty days, orphans and repeated serial numbers. You can run those four checks on your own list in twenty minutes. Resist the urge to expand the register until the first ten rows are reliable. A register should contain what decisions require, not everything that exists.
Step 8: Wire up alerts, then test them
Monitoring is the last piece because it depends on all the others. Decide how raw alerts become incidents and who receives them. An Atlas lab shows why window settings matter: thirteen alert rows folded into two incidents at 8:20, while three latecomers after the 8:30 cutoff opened more, which left five incidents and one that resolved itself.
Then treat the alert path as production. Send a harmless test alert each week and confirm a human received it. The Atlas calls out a silent failure of the event layer as a classic cause of hidden outages. If your paging tool is scheduled for retirement, as one vendor’s product is in April 2027 according to the Atlas, plan the switch early.
Review rhythm: what to do each week and month
- Weekly: sort last week’s tickets, check clocks on any that look late, and review every failed or emergency change.
- Monthly: confirm a handful of register rows, check the freeze calendar, and look for incidents that share a cause.
- Quarterly: re-read the priority rule, the change recipes and the failure limits, and remove anything nobody uses.
Problem work belongs in the monthly slot. Whenever two incidents look alike, open a short investigation with one owner, and write the root cause and the fix in the ticket. That is how a support team turns repeat firefighting into fewer fires.
What not to do
Three warnings from the Atlas apply to small teams. Do not install heavyweight service management on one team running one service. Do not add new rules in response to every failure. And do not run side spreadsheets that duplicate the official lists. If you want a free or open-source tool to hold all this, the Atlas names GLPI and iTop among the options, but a shared sheet is enough for your first month.
When your routine is running, look at how the common role titles in this field map to responsibilities to decide who should own each step as the team grows.
That is the whole sequence: two lanes, a priority rule, written clocks, three change routes, a score, safe automation, a small register and tested alerts. Do the steps in order, keep each artefact to one page, and your support team will have a routine that a newcomer can learn over one coffee.