Meta Pixel

How to Get Your Whole Team Actually Using Claude, Not Just Paying for It

Licences are not adoption. How New Zealand teams move from a few keen users to everyone using Claude on real work: one workflow at a time, timed against their own baseline, taught in the room.

James Oldham

James Oldham

Founder, Sentry AI

22 October 2026

In short

Most teams that buy Claude end up with a few enthusiasts and a lot of unused seats, because a licence gives people a tool but not a job to do with it. Adoption comes from picking one recurring workflow per person, timing how long it takes today, building it into a Claude skill with the person who does it, and checking the time again a week later. Teaching works best in the room, on the team's own work, not in a generic course. The measure is the time each workflow takes against its own week-one baseline, not how many people logged in.

The pattern is common enough to have a shape. A business buys Claude for the team. Three people use it every day and talk about it at every meeting. Everyone else opened it once, asked it something general, got a general answer, and went back to how they did things before. Six months later someone asks whether the seats are worth renewing.

Nothing is wrong with the tool. A licence gives people a capability and no job to do with it. Adoption is the job.

Start from a workflow, not a feature

The question that moves a team is not "what can Claude do?" It is "which part of your week do you least want to do again?" Every role has one: the weekly report, the proposal first draft, the quote that needs the same five checks, the inbox triage on a Monday, the meeting notes that become actions.

Pick one recurring workflow per person. Not five. One, done properly, is what changes behaviour, because the person feels the time come back on something they do every week.

Time it before you touch it

Before anyone builds anything, time the workflow as it runs today. Ten minutes with a stopwatch and the person who does it. Write the number down with its frequency: forty minutes a week, three hours a month.

This is the step most rollouts skip, and it is the one that decides whether anyone can say, three months later, whether it worked. A baseline turns "people seem to like it" into "this took forty minutes and now takes fifteen". Our default goal for a workflow is half its own baseline, measured on the same job a week later.

Build it with the person who does the job

The fastest route from a recurring workflow to something Claude does reliably is a skill: a short written set of instructions Claude follows the same way every time, with the rules, the format and an example. The person who does the job knows the rules nobody wrote down. They have to be in the room when it is built.

A skill built this way does two things a course cannot. It makes the good result the normal one, because the rules travel with the instructions. And it gives the person something they own and can improve, which is what keeps them using it. If you want to try this on your own process, our free process-to-skill skill walks Claude through writing one with you.

Teach in the room, on their work

Generic AI training has a short half-life. People remember the demo and forget the method, because the examples were someone else's. Teaching on the team's own documents, clients and numbers sticks, because the first thing they make is something they needed anyway.

That is why we run training as a day in the client's office rather than a webinar: everyone set up, everyone with one workflow built and working before we leave. What a Claude training day covers.

Check again a week later

A week after the build, time the workflow again. Three things can happen:

  • It is faster. Write down the new number. That person is now the example for the next one.
  • It is not faster. Usually the skill is missing a rule, or the input is in a place Claude cannot reach. Fix the skill, not the person.
  • They stopped using it. Ask why. The answer is the most useful thing you will hear all month.

Then pick the next workflow. Teams that keep this rhythm, one workflow at a time, timed before and after, end up with Claude in the everyday work rather than in a tab nobody opens.

What to measure

Not logins, not prompts sent, not seats active. Those count activity. Measure the time each named workflow takes against its own week-one baseline, and the number of workflows that have moved. That is the figure a director can put in a board paper, and it is the one that answers whether the seats are worth renewing.

If you are deciding between plans first, Claude for business sets out what each plan gives a team.

Questions people ask

How long does it take a team to start using Claude properly?

The first useful workflow per person usually lands in the first day if it is built on their real work in the room. Habits take a few weeks. The sign it has stuck is that people reach for Claude on the workflow without being reminded, and the timed baseline has come down.

Should everyone get a Claude seat on day one?

Give a seat to everyone who has a named workflow to use it on. A seat with no job attached is the one that goes unused. Add people as their first workflow is ready.

Is Claude or ChatGPT better for team adoption?

Both work. The choice matters less than how it is rolled out. We compare them for business use in Claude or ChatGPT: which one.

Do we need an AI policy before we start?

You need a short one: what data may go in, which plan you are on and what it does with your data, and who to ask. One page is enough to start, and it should be written before the first training session, not after.

Written by

James Oldham

James Oldham

Founder, Sentry AI

James Oldham founded Sentry AI in Auckland and runs it: an AI engineer who designs, builds and operates the AI Operating System for businesses across New Zealand and Australia, and who does the engineering on the work described in these posts.

  • Founder and sole director of Sentry AI Solutions Limited (NZ company 9399152, incorporated 20 January 2026)
  • Built custom AI systems for enterprise teams, including the NBA, before founding Sentry AI
  • Delivered the voice, knowledge-graph and agent systems in the named case studies on this site, across recruitment, healthcare, retail and financial services
  • Taking Sentry AI through its own ISO 27001, SOC 2 and ISO/IEC 42001 programme
  • Auckland University of Technology
More about James

Ask an AI about this page

Opens the assistant with this page loaded: read it, summarise it, cite it.

More from the blog