Your AI licences are bought. Nobody uses them

A while ago I noticed something curious at a café. The barista had put a tip jar with a few coins in it next to the till. Nothing unusual there. But during a shift change it was swapped for an empty jar, and the tips nearly stopped. Until someone put a few coins back in.

An empty jar asks you to be the first. Most people would rather not be.

Organisations run into the same thing with almost every new tool. The licences are bought, the training is done, there's a page on the intranet. And then nobody uses it. The usual next step is to appoint champions: one enthusiastic person in every team who will show the others the way.

It sounds sensible. One of the cleanest experiments on how behaviour spreads suggests it's the wrong shape.

In 2010, Damon Centola, then at MIT, built an online health community of 1,528 people. Everyone got six or eight anonymous health buddies, and Centola decided exactly who was connected to whom. In one version of the network, a person's buddies were also linked to each other, the way colleagues in a team are. In the other version, everyone had exactly as many buddies, but the links were spread at random across the whole network. Then one person in each network signed up for a new health forum. Every time someone signed up, their buddies received an invitation.

A popular hypothesis predicted that the random networks would win. Random links work as shortcuts. They carry news across a population quickly, the way a virus travels.

The tight-knit networks won easily. The forum reached 54 per cent of them, against 38 per cent of the random networks, and it spread more than four times as fast.

The explanation sits in individual decisions. People were much more likely to sign up after a second invitation than after the first, and a third still helped. This is what researchers call a complex contagion. A virus usually needs one contact. New behaviour often needs several, because one person doing something doesn't yet tell you whether it's worth it, or whether it's for people like you.

In the SUE | Influence Framework^© we call that an anxiety: the doubts and uncertainties about the new behaviour that keep people where they are. One invitation leaves those doubts in place. A second and a third, from people around you, start to settle them.

Centola is honest about the limits of his study. Signing up for his forum was easy. He expects real behaviour, the kind that costs time, effort or comfort, to need even more reinforcement. A new tool at work is exactly that kind of behaviour. People have to change a routine, risk looking clumsy in front of colleagues, and invest time before they see anything back.

Now look at the champion model again. One champion per team means everybody gets one signal. That's the random network. The champion is enthusiastic, nobody around them has moved, and for everyone else the doubt stays.

That doubt is where the work is. The licences are there and the training happened, so skill isn't what's missing. What's missing is the reassurance of seeing the new way several times, from people you work with every day.

So cluster your early users. Instead of one in every team, put three in the same team, doing the same kind of work, using the tool where the others can see it. When that team works this way, let the next team see them. It's slower to set up and it looks less impressive on the rollout slide. It's also how the behaviour spread in Centola's networks.

The café got its tips back the same way. Nobody put up a sign explaining why tipping matters. Someone put a few coins back in the jar.

A cold needs one contact. New behaviour needs a few.