What the science of how behaviour spreads says about how many champions a team really needs, and why your best-connected people may be your weakest bridges.

 

When building a champions network, the instinct is to focus on coverage. Every team should have someone. Every business unit should have a name on the list. We measure success by how many people in the organisation had a champion within reach, and by that measure, we did well.

I believe that measure is answering the wrong question. Coverage tells you how far information can travel. It tells you almost nothing about whether behaviour will change. Those are different processes; they move through networks in different ways, and the difference explains why so many champion programmes report high engagement and flat outcomes.

Information travels light. Behaviour does not.

Mark Granovetter's classic finding is that information spreads fastest through weak ties and long bridges (Granovetter, 1973). If you want a message to reach the whole organisation, you want a few well-connected people with contacts in every corner. One exposure is enough. Someone hears it, and now they know.

Damon Centola and Michael Macy argued that this logic breaks down in the case of behaviour (Centola & Macy, 2007). When adopting something is costly, socially risky, or of uncertain legitimacy, people do not act on a single signal. They wait until they have seen it from several independent sources. They call this 𝐜𝐨𝐦𝐩𝐥𝐞𝐱 𝐜𝐨𝐧𝐭𝐚𝐠𝐢𝐨𝐧, and it changes what a network needs to look like. A single contact can tell you something. It takes several contacts, close together, to convince you that doing it is normal and safe.

Centola then tested this directly (Centola, 2010). He built online health communities of identical size, where every member had the same number of contacts, and varied only the network's shape. In one version, contacts were clustered, so people shared friends. In the other, ties were spread randomly across the whole community. The behaviour spread further and faster in the clustered network. The reason was redundancy. In a cluster, people saw the same behaviour from more than one person, and that repetition was what tipped them into adopting it. The long ties that make information move quickly actually slowed behaviour down, because they carried one signal into a place where nobody else was reinforcing it.

Most security behaviour is complex contagion

Think about what we actually ask a champion to spread. Challenging a senior colleague's unusual request. Slowing down a delivery to check a change. Reporting your own mistake and trusting that the response will be fair. None of these is an information problem. People already know what they are supposed to do. What they are waiting for is proof that other people like them do it and nothing bad happens.

That is the textbook condition for complex contagion. And it means that one champion, however committed, is often below the threshold that would move anyone. They are a single signal. Their colleagues hear the message and file it under things the security person says.

There is a useful exception. Small habitual behaviours with almost no cost, such as locking a screen, may spread as simple contagion, in which a single prompt is enough. Knowing which of your target behaviours fall into which category is worth the effort, because it tells you which ones need a network and which just need a reminder.

Two champions in one team beats one champion in each of two

This is the uncomfortable part. The standard playbook, one champion per team, spreads single signals thinly across the organisation and leaves almost everyone short of the threshold. Two champions in a team of twelve will change that team. One champion each in two teams of twelve may change neither.

Density within a team matters more than company-wide headcount. Centola's later work on tipping points found that a committed minority of roughly a quarter of a group was sufficient to flip an established convention, and below that threshold the minority had almost no effect (Centola et al., 2018). Whether the same threshold applies to security norms is an open question, but the shape of the finding is what matters. There is a floor, and a champion programme that never reaches it in any single team is spending its effort in the wrong places.

The practical move is to stop optimising for reach and start optimising for clusters. Pick the teams that matter most, put two or three champions in each of them, and accept that some teams will have none for now. A network that has genuinely changed behaviour in twenty teams is a better foundation than one that has a name in every team and has changed nothing.

Your bridges need to be wide

The second implication concerns how behaviour crosses from one part of the organisation to another. The usual approach is to find the well-connected person, the one who sits on every cross-functional call, and make them a champion. They are a bridge, and bridges are good.

For information, yes. For behaviour, a single well-connected person is what Centola calls a narrow bridge. They carry the message into the neighbouring department, but once it arrives, it is one signal again, and it dies. Behaviour crosses between groups when the bridge is wide, meaning several overlapping relationships connect the two sides, so that someone in the receiving department hears it from more than one place.

This has a sharp consequence for how we choose champions. Guilbeault and Centola showed that the standard ways of picking influential people, by how many contacts they have or how often they sit on the shortest path between others, select the wrong seeds when the goal is spreading behaviour rather than news (Guilbeault & Centola, 2021). The people who appear most central on a network chart are often those whose ties are spread widest and thinnest. What you want instead are people whose relationships overlap with those of other champions, so the signals reinforce one another.

What this does to measurement

If your champions programme reports coverage, you are measuring readiness for information to spread. If you want to know whether it is set up to change behaviour, ask different questions. What proportion of staff have two or more champions among the people they actually work with? How many pairs of departments are joined by more than one champion relationship? Where would the loss of a single champion drop a whole team below the threshold?

That last question matters for both resilience and design. In a complex contagion, a single departure does not remove a single node. It can take a cluster from above the line to below it, and the behaviour that cluster had adopted can quietly unwind. Anyone who has watched a strong team drift back to old habits after their champion moved on has seen this happen without having a name for it.

What we are testing

I want to be honest about the limits. Centola's experiments were in online communities, not in a company trying to get people to report phishing emails. Whether the threshold shape shows up in security behaviour is something we have to test, not assume, and observational data alone cannot settle it, because people with similar habits tend to cluster together anyway (Shalizi & Thomas, 2011). The clean test is to hold the number of champions constant, randomise whether they are clustered or dispersed across comparable business units, and observe behaviour over the following year.

That is the study we are building into Heroes. In the meantime, if you run a champions network and have one person per team, consider expanding by adding a second or third person in critical areas first. This will also help build your network's resilience.

 

References

Centola, D. (2010). The spread of behavior in an online social network experiment. Science, 329(5996), 1194–1197. https://doi.org/10.1126/science.1185231

Centola, D. (2018). How behavior spreads: The science of complex contagions. Princeton University Press. https://doi.org/10.23943/9781400890095

Centola, D., Becker, J., Brackbill, D., & Baronchelli, A. (2018). Experimental evidence for tipping points in social convention. Science, 360(6393), 1116–1119. https://doi.org/10.1126/science.aas8827

Centola, D., & Macy, M. (2007). Complex contagions and the weakness of long ties. American Journal of Sociology, 113(3), 702–734. https://doi.org/10.1086/521848

Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78(6), 1360–1380. https://doi.org/10.1086/225469

Guilbeault, D., & Centola, D. (2021). Topological measures for identifying and predicting the spread of complex contagions. Nature Communications, 12, Article 4430. https://doi.org/10.1038/s41467-021-24704-6

Shalizi, C. R., & Thomas, A. C. (2011). Homophily and contagion are generically confounded in observational social network studies. Sociological Methods & Research, 40(2), 211–239. https://doi.org/10.1177/0049124111404820