Rogers, hybrid corn, and why your AI pilot will not spread by itself

In the 1940s, Iowa farmers facing the same weather, the same markets, and broadly the same evidence did not adopt hybrid corn at the same pace. Some planted early. Neighbours waited. A few never switched. Bryce Ryan and Neal Gross documented that unevenness. Everett Rogers, who grew up in that rural world and later worked as a rural sociologist at Iowa State, spent a career showing it was not really a corn problem.

It was a diffusion problem. Health systems still have it. Your AI pilot sitting in one clinic while the neighbouring unit ignores it is not a mystery unique to machine learning. It is the same S-curve wearing a different badge.

What Rogers actually contributed

Ryan and Gross's hybrid corn study found adoption did not rise in a straight line. A few early experimenters, then a steep climb as the practice caught on, then a plateau as remaining holdouts converted or dug in. Rogers combed through hundreds of studies across agriculture, education, and community development and found the same shape again and again. In 1962 he published Diffusion of Innovations, which gave the field a vocabulary that is still in use, for better and for worse.

Adopter categories, with the classic percentages: innovators (about 2.5 percent), early adopters (about 13.5 percent), early majority and late majority (about 34 percent each), laggards (about 16 percent). Treat those numbers as heuristic, not physics. The useful idea is that people do not all decide at once, and that the people you need for scale are not usually the first people who love the demo.

Rogers also named attributes that predict whether something spreads: relative advantage, compatibility with existing values and workflows, complexity, trialability, and observability of benefit. If a tool is hard to try, hard to see working, and incompatible with how a ward already runs, technical accuracy will not save it.

His deeper move was reframing diffusion as social, not merely technical. Relationships, norms, and how an innovation is communicated matter as much as whether the underlying idea "works" in a paper.

Where digital health keeps relearning this the hard way

Walk into most health systems and you can map Rogers without needing a seminar. One service line has an AI triage tool in production. Another has the same vendor deck and a polite "not now." A third ran a pilot, declared success on accuracy metrics, and never crossed into routine use.

Technical performance matters. So do trust, transparency, and whether early adopters can show sceptical peers a benefit that is visible in their own workflow. Programmes that treat adoption as a training deficit ("if we explain the model harder, they will use it") keep stumbling. Programmes that treat adoption as coalition work, local adaptation, and honest handling of risk have a better chance of getting past the pilot slide.

Compatibility is the attribute students underestimate. A model that requires six extra clicks, or that surfaces recommendations at the wrong moment in the chart, fails Rogers' test even when the AUROC looks excellent. Relative advantage has to be advantage for the person expected to change behaviour, not advantage for the analytics team that built the dashboard.

Trialability and observability matter for the same reason. If clinicians cannot try a tool safely, or cannot see what improved when they did, the S-curve stalls in the early adopter pocket and never reaches the early majority. That stall gets misread as "doctors are resistant to innovation." Often it is "the innovation never became legible inside the workday."

The questions worth stealing from Rogers

Rogers is most useful when you stop using him as decoration in a literature review and start using him as a checklist in an implementation meeting.

1. Who is this for, specifically, and who decides whether it spreads beyond the pilot site?
2. What is the relative advantage in their terms: time saved, fewer callbacks, safer handoffs, less documentation, clearer decisions?
3. Where does it fight existing values or workflows, and are you naming that conflict or papering over it with change-management language?
4. Can someone try it without betting their shift or their licence?
5. What would a sceptical peer need to see to believe it worked, and is that evidence being collected?
6. Who benefits first, who bears the failure modes, and are those the same people?

In global health and in under-resourced clinics, these are not academic questions. An innovation designed without its end users, or rolled out without attention to capacity and context, will diffuse unevenly no matter how good the underlying technology is. Equity failures are often diffusion failures with a body count attached.

A note on "laggards"

Rogers' language gets abused here. Calling a clinician a laggard is a great way to end a useful conversation. Late adoption is sometimes wisdom: waiting for evidence, waiting for workflow fit, waiting until the liability story is clear. In health, that caution can be the feature, not the bug.

Use the categories to describe timing and social position, not moral worth. The person who declines your pilot may be protecting patients from an unvalidated tool. Your job is to find out which Rogers attribute failed for them, not to win the argument by labelling them behind.

What to do with this on Monday

If you are a student writing about "barriers to AI adoption," do not stop at "lack of training" and "resistance to change." Those phrases hide the mechanism. Name the Rogers attribute that is actually broken. Is the tool incompatible with the medication-administration workflow? Is the benefit invisible to the nurse who has to click it? Are the early adopters people with no opinion leadership on the unit?

If you are sitting in a go-live meeting, ask which part of the S-curve you are actually in. A room full of innovators celebrating a pilot is not evidence that the early majority will follow. Ask what would make the next third of users try it once, see a result, and tell a colleague.

Rogers watched hybrid corn move unevenly across farms that looked identical from the road. Hospitals look identical from the org chart and still adopt at different speeds for the same reasons: social proof, local fit, visible benefit, and trust. Innovations do not spread because they are good on a slide. They spread because enough people, in enough relationships, decide the change is worth the cost of changing.