Before we recommend AI tooling to a single client, we're running it across our own service desk for 30 days. Here's how we've scoped it, and what we're measuring.
There is no shortage of people willing to sell you an AI strategy at the moment. There is rather less appetite for going first — for putting the tooling into your own business, on your own systems, with your own people, and then being honest about what happened.
So that's what we're doing. Over the next 30 days, iTops is running a structured internal rollout of Claude across the business, before we take a single recommendation to a client. Here's how it's scoped and what we're paying attention to.
Why internal first
We've spent twenty years telling clients not to put anything into production that hasn't been tested somewhere it can safely fail. It would be a poor look to abandon that principle the moment something new and interesting turned up.
There's a more practical reason too. The interesting questions about AI in a managed services business aren't really about the model. They're about process: who is allowed to use it, for what, with which data, and how you know whether it actually helped. You can't answer any of that from a demo. You can only answer it by living with the thing for a few weeks.
How we've scoped it
The rollout is a single ticket with a single owner, and the scope is deliberately narrow:
- Internal enablement only for the first 30 days — no client-facing work rides on this ticket
- Organisational settings reviewed up front, including what the tool retains and what it doesn't
- Domain verified properly so identity and access sit where the rest of our tooling sits
- One senior engineer and our technical lead driving team onboarding, so it isn't everyone's side project
- Notes taken throughout, because the notes are the actual deliverable
That last point matters more than it sounds. The output of this exercise isn't “the team now uses AI”. It's a documented process for how a New Zealand business of our size introduces this class of tool without creating a security or compliance problem for itself. That's the thing our clients will actually need.
What we're measuring
Vendor case studies love a productivity percentage. We're more interested in questions we can answer honestly:
- Where did it genuinely reduce time to resolution, and where did it just move the work around?
- What did people stop using after week two, and why?
- Which tasks turned out to need more review than they saved?
- What did we have to change about our documentation for any of it to work?
The organisations that get value out of this won't be the ones that adopted earliest. They'll be the ones who understood their own processes well enough to know where it fits.
The bit nobody enjoys talking about
Governance. Before a single person on our team typed anything into a chat window, we went through organisational settings, data retention, and what happens to information that goes in. If you are in a regulated sector, or you hold client data under contract, that review isn't optional and it isn't a five-minute job.
It is also entirely doable. The controls exist. Most organisations simply haven't been shown where they are, or have adopted the tooling department by department without anyone taking a whole-of-business view. That's usually where the risk creeps in — not in the technology itself.
What happens next
At the end of the 30 days we'll write up what worked, what didn't, and what we'd do differently. Then we'll start taking it to clients as separate, properly scoped pieces of work — not as a bolt-on to a support contract.
If you're weighing this up for your own organisation and you'd rather talk to someone who has actually done it than someone selling a licence, get in touch. We're happy to share the notes.
