Saturday, November 14, 2026

Why Every Industry Needs Its Own AI Adoption Plan

Making AI Work in the Enterprise  |  03 of 07

The same technology can support very different work, so leaders need to judge adoption in its operating context.

By Rinoo Rajesh

Five architectural settings for banking, life sciences, media, property and software surround one central technological core.

Editorial illustration of a shared technology serving five different industry settings.

A leadership team returns from an AI conference with a compelling success story. Another company has accelerated a process, reduced effort and impressed its customers. The obvious next step seems to be replication. Yet the proposed application touches different information, different customers and different consequences when something goes wrong. Before commissioning a similar tool, the team needs to establish which parts of the example are transferable and which depend on conditions its own organisation does not share.

McKinsey's collection on AI in five industries provides a useful starting point. It covers life sciences, banking, film and television, real estate, and software development, emphasising differences in workflows and adoption barriers.[1] Reading across the collection suggests a practical discipline for project leaders: describe the specific decision or service being changed before discussing the technology that might support it.

The life sciences explainer distinguishes scientific applications from operational ones, including work involving substantial documentation.[2] My recommendation is to make the evidence requirement explicit for each proposed application. Helping somebody locate an approved document and producing a scientific claim are different assignments. A project brief should identify the intended user, the status of the source material and the person qualified to accept the result. That precision makes a smaller initial scope easier to justify.

Consider a hypothetical document-support pilot in a pharmaceutical company. The team could begin with a controlled set of approved procedures and ask users to assess whether answers point to the correct passages. Missing evidence would become a recorded failure requiring attention. The acceptance record could identify the procedure version and the reviewer who checked the answer. That information would help the team investigate disagreements and decide whether the application was ready for a broader trial.

The banking explainer discusses systems progressing from assistance toward actions within operational processes.[3] For a project team, I would translate that observation into a map of permitted actions. Retrieving a customer's case history, proposing a response and changing an account record should be treated as separate capabilities. The project can then decide which capability to test first and what evidence would justify extending access. A broad label such as intelligent service would conceal those important distinctions.

In film and television, the linked discussion considers production work beyond generating finished video, including planning and postproduction activities.[4] A producer evaluating a hypothetical tool could ask whether it helps prepare a more accurate day's shooting plan. The useful measure might be fewer avoidable changes once the crew arrives. Generating many attractive planning options would be less persuasive if staff still had to reconstruct the practical schedule manually before anybody could use it.

Real estate offers another context. The explainer identifies leasing, property operations and administrative work as areas for AI applications.[5] A property manager choosing an initial project might favour a service with clear ownership and an observable completion point. For example, routing routine enquiries could be easier to assess than a broad promise to improve tenant relationships. The latter remains important, but it needs more careful interpretation than a simple count of messages answered.

The software discussion highlights the wider development process as a source of constraints when coding accelerates.[6] A software leader could therefore examine the time between an approved customer need and a usable release. That boundary includes decisions made outside engineering. Applying the same analysis to a laboratory or a building would require different acceptance evidence, which illustrates why a common technology platform cannot supply a complete common operating plan.

A shared project template can still help. I would require each proposal to name the person affected, the action being assisted, the evidence needed for acceptance and the route for recovering failed work. Add a short account of the current process, including its known weaknesses. These fields let a portfolio team compare the clarity of proposals without pretending that every department has the same needs or should pursue the same application.

The approval conversation should then examine readiness. Is there enough reliable information to test the proposal? Can the responsible team absorb the change? Will somebody be available to review results? A seemingly modest application with clear answers may deserve priority over an ambitious demonstration with unresolved dependencies. Recording the reason for that choice also helps teams understand what they must improve before their own proposal can move forward.

For organisations serving several industries, this calls for deliberate local involvement. Keep reusable infrastructure where it helps, but ask people who understand the customer's work to approve the service assumptions. Their contribution should influence scope and acceptance, rather than arrive as a final endorsement. The next time a success story inspires an AI proposal, use it to begin an informed comparison. A well-designed adoption plan should explain why this application makes sense here.

Sources

  1. McKinsey & Company. AI in five industries: What's now, and what's next. 1 July 2026.
  2. McKinsey & Company. AI in life sciences explained: The technology that could reinvent medicine. 9 June 2026.
  3. McKinsey & Company. Banking and AI: When the tech starts doing the work, not just assisting it. 8 June 2026.
  4. McKinsey & Company. Lights, camera, algorithm: How AI is rewriting the rules of film and TV. 1 July 2026.
  5. McKinsey & Company. Where AI is creating real value in real estate. 15 May 2026.
  6. McKinsey & Company. AI-powered software development: How technology is rewriting the rules. 24 June 2026.

No comments: