AI adoption in pharma: a context problem, not a tooling problem

Most pharma AI programmes do not fail because the technology is wrong. They fail because the leaders running them misread the people.

Ganes Kesari's piece in MIT Sloan Management Review this week is not aimed at pharma specifically, but every line of it lands in life sciences. He names three reasons adoption lags: AI feels inaccessible and scary, it looks like a lot of avoidable work, and the benefits are communicated in the wrong currency.

Kesari's sharpest point is that what looks like resistance to AI is usually exhaustion from earlier rollouts that demanded too much for too little. That rings true in pharma. The objection is rarely "I do not believe AI works." It is "I do not have the bandwidth to make it work right now."

His prescription has three pillars. Demystify AI through everyday analogies, not technical lectures. Embed AI into systems people already trust — renovate, do not relocate. Quantify impact using metrics people already get judged on: MLR cycle time, sample compliance, signal detection latency, call plan adherence, cost per audit finding.

The pilot works. The change does not. The reason is almost never the model. It is the failure to build adoption around the human and operational context the model has to live inside. AI in pharma is not a tooling problem. It is a context problem. Compliance, trust, change fatigue, and operating metrics, in that order, are what determine whether a model actually changes behaviour at the front line.

This is also the precise gap that the next wave of pharma-native AI integrators is built to close. Big consulting firms have the slideware. Hyperscalers have the platforms. What is missing is the operator-grade middle layer — partners who carry both regulatory scar tissue and implementation craft, and who understand that adoption in life sciences is won at the workflow level, not the platform level.