The 30-60-90 rollout: introducing AI to your IT team

Here's how AI adoption actually fails in IT teams, because it's rarely dramatic. Someone senior gets enthusiastic. Licences are bought. There's a kickoff, maybe a demo that lands well. Three months later, two people use the tool daily, one of them is pasting customer data into it, half the team tried it twice and drifted back to old habits, and nobody can say whether it's working because nobody defined what working meant.

No scandal, no decision to stop. Just a slow leak of intent, and a line item someone will question at renewal.

The failure has two roots, and they're both leadership work, not technology. Nobody set the rules, so cautious people (your best people, usually) stayed away. And nobody set a baseline, so there was no way to show a result, and initiatives that can't show results lose to whatever's louder that quarter. A 90-day structure fixes both.

Days 1-30: rules, baseline, and a narrow pilot

Resist the urge to roll out to everyone. The first month has three jobs.

Set the data rules before the first prompt. One page, not a policy binder: what may go into external AI tools, what never does (customer identifiable data, credentials, anything under NDA), and how to anonymise a ticket before pasting it. Publish it, walk the team through it once, and you've done more for safe adoption than most companies manage in a year. Do this first; every week of AI use before the rules exist is a week of habits you'll have to unwind. The data-safety checklist in the Opstimio library is a ready-made version of this page if you'd rather adapt than write.

Measure the before. Pick the two or three workloads the pilot will target (first responses and escalation summaries are the usual suspects) and spend a week measuring current state. Minutes per first response. Escalation bounce rate. It's unglamorous, and it's the difference between "the team feels faster" and "first-response time dropped 38 percent" when someone asks in Q4 what this initiative delivered.

Pilot with three or four people, chosen carefully. Not the four biggest enthusiasts. Two enthusiasts, one respected sceptic, one ordinary busy engineer. If it works for the sceptic and the busy one, it works. The enthusiasts alone can make anything look good for a month.

Days 31-60: turn wins into shared assets

The pilot will surface what actually helps, and it's usually less flashy than expected. Month two is about converting those wins from personal tricks into team property.

The habit that matters most: when a pilot user finds a prompt that works, it goes into a shared library, named by task, findable by someone who wasn't in the room. Personal prompt collections die with holidays and job changes; a shared library compounds. This is also when you write the first playbooks: this ticket type, this prompt, review the draft, send. Boring documentation, and it's what makes week ten look like week six instead of like week one.

Expand from four users to the wider team now, with the pilot users as the local experts. Peer help beats a training session, and the sceptic-turned-user is your most credible advocate by far. Keep measuring the same numbers you baselined; month two is where the curve should visibly move.

Days 61-90: standardise and decide

Month three is judgement month. Compare against the baseline honestly. Some workloads will show strong gains, one will be marginal, and the honest report says both; killing the marginal use case in public buys you credibility that carries the strong ones for a year.

Standardise what earned it: the working playbooks become the default way those tasks are done, new-hire onboarding includes them, and the data rules get their first scheduled review (things you allowed or banned in month one will look different with 90 days of evidence). Set the ongoing rhythm now: a monthly half hour on what's working, what drifted, and what's next keeps this from being a one-quarter story. Then take the next two workloads and run the same loop, faster this time because the rules, the library, and the measurement habit already exist.

What this looks like from above

Run it this way and the 90-day report writes itself: here's the baseline, here's the after, here's the annualised hours returned, here's what we standardised, here's what we're doing next quarter. Leadership gets numbers instead of vibes.

But the deeper change is in the team. AI stops being the champion's hobby or the mandate from above. It's just how first responses get drafted now, the way ticket templates are just how tickets get logged. The full week-by-week version of this rollout runs as a checklist in the Opstimio library, and teams tell us the checklist's real value is pacing: it stops the month-one enthusiasm from skipping the boring steps that make month three possible.

Ninety days, three phases, two numbers measured before you start. That's the difference between the initiative that fades and the one that quietly becomes the operating system of the desk.