How To Roll Out Claude Across Your Business | Alyra

Rollout order decides whether you get one way of working or six. Start with one whole team on work that matters, then test whether it travels.

The common failure is not that it does not work. It is that it works differently on every team, and nobody notices until the output starts disagreeing with itself.

Most rollouts start the same way. A handful of enthusiasts across several teams get access, everyone reports back that it is impressive, and six months later the business has no shared practice and no evidence worth showing anyone.

The order you do this in matters more than the configuration.

Start with one team, not one volunteer per team

Scattered individuals produce scattered learning. Nothing accumulates, because there is nobody beside them hitting the same wall on the same work.

One volunteer on each of six teams gives you six pilots and no signal. One whole team gives you conventions, because they have to agree to work together.

A whole team produces conventions, because they have to agree in order to hand work to each other. Those conventions are the actual deliverable of the first phase. Whatever gets built is secondary.

Pick the work that matters

The instinct is to start somewhere harmless, on an internal document nobody depends on. That teaches you very little, because the thing you need to learn is how the output holds up when somebody is relying on it.

Choose work that already has a checker in the process: a second pair of eyes, an approval, a reconciliation. If the assistant gets something wrong there, something catches it, and you learn where your own checks have holes. That is the finding worth having in week one rather than month six.

Measure what survives a hard question

Seats issued is not a measure. Neither is messages sent, which counts activity rather than outcomes.

The numbers that hold up are how long a piece of work takes end to end, how much rework comes back after review, how often something reaches a client needing correction, and how long a new joiner takes to be useful. Take all four before you start. A baseline collected afterwards is an argument, not a measurement.

The second team is the real test

The first team's result is interesting. The second team's result tells you whether what you learned travels, or whether it depended on four unusually motivated people.

If the second team has to invent its own way of working from scratch, the first phase did not produce what you thought it did. Go back and write the conventions down properly before anyone else gets access.

Widen on evidence, stop on silence

When the numbers repeat, widen. When they do not, find out why before the third team rather than after the thirtieth. The expensive version of this is not a failed pilot. It is a successful pilot that quietly becomes six incompatible ways of working, each with people willing to defend it.

Start a conversation