The AI Adoption Curve

Picture a page from a quarterly board pack. The company is invented and so are the numbers, but most executives have seen a page like it. Four thousand AI licenses issued, up from fifteen hundred. Twenty-seven pilots running in nine functions. Sixty-one percent of people with access using the tool every week. Above the tiles, a headline: AI transformation on track.

Read the page again and ask a simple question: which of these numbers shows that any piece of work is now done differently? None of them does. Each tile measures how widely AI has spread through the organization. Not one measures how deeply it has gone into the work. The headline is a claim about depth, supported only by evidence of breadth.

That gap is what this chapter is about. It offers a way to describe where an organization really stands with AI, why the most common position is wide but shallow, and what moves an organization from one position to the next.

Two curves, not one

The phrase adoption curve comes from the sociologist Everett Rogers, whose Diffusion of Innovations, first published in 1962, is still the standard work on how new ideas spread. Rogers showed that adoption across a population tends to follow an S-shaped curve: slow at first, then rapid, then leveling off. He sorted adopters into five groups by when they take up an innovation¹.

How fast generative AI moves along this curve, and why it often spreads from employees upward, is the subject of The Speed of AI Adoption, later in this module. The point here is different. Rogers’s curve counts people. An executive also needs a second curve, one that tracks depth.

Rogers saw this himself. When he studied innovation inside organizations rather than among individuals, he found that the decision to adopt is only the midpoint. After it comes implementation: the innovation is reshaped to fit the organization, people learn what it means for their work, and finally it is routinized, absorbed into regular operations until it no longer feels like a separate initiative¹. A license is an adoption decision. Routine is the outcome.

This is the idea that runs through the rest of the chapter. AI adoption is a curve of depth, not only of spread. It rises from tools in hand to work redesigned, and your position on it is set by where the AI lives, not by how many people have access to it.

Wide but shallow

The most careful recent picture of this comes from the US Census Bureau. In 2026 it published results from a nationally representative survey that asked firms not only whether they use AI, but in which business functions and for which tasks².

Three findings matter for leaders. First, use is narrow even where it exists. Among firms that use AI, 57 percent use it in three or fewer business functions, and where workers use it in their tasks, 65 percent of firms limit that use to three or fewer tasks. Second, use is mostly additive: two-thirds of users rely on AI only to assist with tasks people already do. Third, the two kinds of adoption do not line up. Workers sometimes use AI with no formal adoption by the firm, and firms sometimes adopt AI that workers do not use in their tasks. The authors also find that firms using AI across more functions and tasks tend to perform better commercially, a correlation rather than proof of cause².

This is the board page in national statistics. Spread is real, and it is growing. Depth is thin. As AI Is Changing Everything showed, using AI and getting value from it are far apart. The adoption curve explains where that distance sits.

Where the AI lives

To read your position on the curve, ask one question about any piece of work: where does the AI live? There are four answers, and they rise from tools toward redesigned work.

In a license. The organization has bought access: logins, pilots and a good deal of curiosity. Every organization starts here, and this is where it learns what the technology can do. The risk is not experimenting; it is experimenting without a boundary, so that sensitive data drifts into unapproved tools and five departments solve the same problem separately.

In a person. Individuals now work faster. A buyer summarizes a long supplier contract in minutes; an analyst drafts a first report in an afternoon. These gains are real and well documented, and the next section returns to them.

In the process. The workflow itself depends on the AI. Nobody has to remember to open a tool, because the work arrives already prepared. People check it, handle the exceptions and own the decisions. This is Rogers’s routinization, and it is where repeatable capability begins.

In the business. AI shapes what the organization sells and how it competes: new products, different pricing, a customer journey or an operating model that would not exist without it. Marco Iansiti and Karim Lakhani describe firms whose core operations run on data and algorithms, which lets them grow without the limits that bind conventional organizations³. This position is rare and uneven, and it is not “AI everywhere”. If someone claims it, ask what changed in the customer offer.

These four places are a description, not a score. They are not stages to certify or a badge to publish. A formal assessment scale belongs to The AI Maturity Model, in Module 9. For now, the shape of the curve is enough.

The hard bend: from person to process

The most important move on the curve is from person to process, and it is the easiest to miss, because both sides look like progress.

The gains at the person level are not in doubt. In a controlled experiment with 453 college-educated professionals, Shakked Noy and Whitney Zhang found that access to a generative AI assistant cut the time taken on writing tasks by 40 percent and raised the quality of the output by 18 percent⁴. Studies like this explain why personal use spreads so quickly.

Yet a faster person is not yet a different business. A large Danish study, which linked about 25,000 workers to official records, found no measurable effect on earnings or recorded hours two years after ChatGPT’s launch; what moved was the structure of work, as employers reorganized tasks⁵. Why task-level gains so rarely reach the economic records is the subject of AI and Workforce Productivity in Module 3. The question for this chapter is where those gains live.

The difference is structural. When the AI lives in a person, the gain depends on that person remembering to use it, and it leaves when they leave. The work still moves through the same queue, the same handoffs and the same approvals, so the organization has a faster step, not a new process. When the AI lives in the process, the work arrives prepared whatever the habits of the person on shift. A new joiner gets the same result in the first week.

Survey evidence points the same way. In McKinsey’s March 2025 global survey, the redesign of workflows had the biggest effect on whether organizations reported earnings impact from generative AI, out of 25 attributes tested. Yet only 21 percent of respondents whose organizations used generative AI said they had fundamentally redesigned even some workflows⁶. McKinsey’s 2026 survey tells a similar story: outside the small group of high performers, only about a quarter of respondents reported fundamentally redesigning workflows because of AI⁷. Most organizations, in other words, are sitting just below the bend.

A simple test helps. For any AI success story you hear, ask: if the three keenest users left tomorrow, would the process still run the new way? If not, the AI still lives in people. Whether the time saved then turns into value is a separate question, taught in Productivity vs Realized Capacity in Module 3.

What the organization adds

If personal use spreads by itself, what does the organization contribute? The Danish study gives an unusually clear answer. Even in workplaces that neither encouraged AI use nor provided tools or training, about 40 percent of workers had used AI chatbots at work. Where employers took active steps, take-up almost doubled. And where employers combined encouragement with enterprise tools and training, 93 percent of workers had used them⁵.

The lesson is not that leaders should chase usage numbers; the board page already has plenty of those. It is that spread at the bottom of the curve happens with or without leadership, while depth does not. Bounded tools, training and permission are what turn scattered personal use into something a process can depend on.

Training is also no longer optional in every jurisdiction.

Several points at once

Few organizations sit at one point on the curve. The Census survey shows how concentrated use is: among firms using AI, the most common functions are sales and marketing (52 percent), strategy and business development (45 percent) and IT (41 percent)². McKinsey’s surveys report a similar concentration in marketing and sales, product development, service operations and software engineering⁶. Inside one company, that unevenness is the normal state. The grid below is an illustration, not survey data: a hypothetical organization whose functions sit at different points, of the kind most executives will recognize.

The mistake is to average this into one enterprise score. One number creates false comfort (“we have AI in fraud”) and false despair (“we have not transformed”). A map by function and by process is more honest and more useful, because it shows where the next move is: usually a process in a function that is already close to the bend.

What moves the curve

Tools do not move an organization along the curve. Leaders do, and the moves are mostly unglamorous.

The first move is choosing one real business problem rather than a showcase. The second is settling what the AI may touch, so people are not left to guess. The third is redesigning the workflow and its checks so that the process, not the enthusiast, carries the AI. The fourth is preparing managers for a different kind of supervision: reviewing prepared work, coaching exceptions and watching quality, not just activity. Alongside these sits a quieter discipline, stopping the pilots that consume attention but never reach a decision.

None of this is a roadmap; sequencing belongs to Building the 90-Day AI Plan in Module 9. The point is narrower. You cannot buy a higher position on the curve, and the goal is not AI everywhere. It is deliberate movement, one process at a time.

Story: one chemicals maker, two teams

The company in this story is a composite, built from patterns common in specialty chemicals. It illustrates the argument; the evidence is in the sections before it.

A specialty chemicals maker, selling coatings and additives to industrial customers, holds its quarterly AI review. Two teams present.

The technical service team goes first. Its chemists took to the AI assistant faster than anyone in the company. They use it to summarize customer complaints, draft advice on how a product should be applied and write up test results. Its usage figures are the highest in the firm, and three chemists have become known as the experts everyone calls. But the work still moves the old way. Customer requests wait in a shared mailbox. Anything that needs a lab test waits for the weekly scheduling meeting. Managers still count tickets closed. Six months later, two of the three experts leave for a competitor, and much of the gain leaves with them. Customers wait as long for an answer as they did before.

The product safety team presents second, and its slides are quieter. Every time a supplier changes a raw material, a change notice arrives: a new formulation, a new test certificate, sometimes a new safety data sheet. Each one can affect dozens of the company’s own products and their safety documents. The team leader chose that intake step and built the AI into it. Each notice is read on arrival. The affected products and the relevant sections of their safety data sheets are attached, missing test data is requested from the supplier, and unusual substances are flagged. A specialist checks the summary against the documents and makes every compliance decision. A senior specialist samples files every week.

When a new specialist joins the product safety team, she produces the same files in her first week. Other parts of the team’s work are still experimental, and the team leader says so in the review.

Which team would look better on the board page from the start of this chapter? Technical service, almost certainly, because the page rewards what is visible. Run the test from the bend instead: if the keenest users left tomorrow, which work would still run the new way? Only the product safety team moved the AI into the process. The chemists did nothing wrong. Enthusiasm without redesign is simply the most common position on the curve, and one enterprise score would have hidden both stories.

What this means for leaders

The curve gives leaders a vocabulary for honesty. It lets you credit real progress at the bottom of the curve without mistaking it for transformation, and it points to the one move that matters most: the bend from person to process.

Three habits follow. Describe your position by function and by process, not with one enterprise number. Treat personal productivity as the starting material for redesign, not as the result. And judge every claim of progress by where the AI lives, which is a question about the work, not about the tools.

Check yourself

1. If every employee has an AI license, the organization has moved up the adoption curve.MythFact
2. Among US firms that use AI, most use it in only a few business functions.MythFact
3. A faster employee means a faster business.MythFact
4. Personal AI use only spreads when leaders actively promote it.MythFact
5. Different functions in one organization can sit at different points on the curve.MythFact
6. Once a process carries the AI, the organization can stop experimenting.MythFact

Reflection: find the bend in your function

What comes next

The curve gives a shared language for where an organization stands. It does not, by itself, create an advantage. Two organizations can sit at the same point, with the same access to the same AI, and still pull apart. The next chapter, AI Leaders vs AI Followers, asks why some organizations move ahead while others, with the same tools, fall behind.