Your best customer can hide your next market.

Two sheets of paper lie on a conference table. The first is next year's product roadmap for a firm that writes and maintains compliance policies and staff training for large employers. Every line on it has an account name beside it: an audit trail by user for one, single sign-on for four thousand seats for another, a custom reporting module for the largest. The ten biggest accounts pay most of the bills, and each request arrived with a reason and, often, a renewal date. The second sheet is blank except for a heading written by hand: People who never asked for a proposal. The firm and its roadmap are invented. The two sheets are not, in the sense that every established business has both, and only one of them fills itself in.

The first sheet is commercially valuable, and nothing in this essay argues against it. The second sheet is not a weaker version of the first. It is a different research question, and a harder one, because nobody on it has ever called. Clayton Christensen's 1997 book The Innovator's Dilemma is the classic account of why that sheet stays empty in companies that are run well.¹

My argument is that careful listening to existing customers produces improvement for existing buyers, and that it is close to useless for finding access for previous nonbuyers: people who buy nothing in the category today because something about the current offer keeps them out. AI may lower the cost of serving some of those people, which makes the second sheet worth filling in now. But that is a hypothesis about a business model, to be tested, not a forecast. A cheaper product is not automatically disruptive, and a successful incumbent is not automatically doomed.

Why the first sheet fills itself

Serving demanding buyers is rational, and it is worth saying why before criticizing it. Large customers pay, renew and describe their needs concretely. A request for single sign-on comes with a seat count, a deadline and a budget. A product team that builds it can measure the result in retained revenue. Very few investments inside a company are that legible.

Christensen's central case was the disk drive industry, and his point was that the leading firms in it were not badly managed. As he told it, when drives shrank to 5.25 inches, their capacities of 5 and 10 megabytes were of no interest to the minicomputer makers who bought the established drives and were asking for 40 and 60. The smaller drives found their first buyers in a market that was only just emerging: desktop personal computers.¹ The established drive makers had listened to their customers, and their customers had told them, accurately, that they did not need the new product. The book describes a company's resources as following its current customers, and small new markets as too small to move a large company's numbers.¹ Both observations describe sound management, not its absence.

The vocabulary that grew out of the book draws a line between two kinds of improvement. A sustaining innovation makes a product better for the people who already buy it, on the measures they already care about. That is the gold sheet. A disruptive innovation, in the theory's narrower sense, is something else. The Christensen Institute's own summary is blunt: disruptive innovations "are NOT breakthrough technologies that make good products better."² They take root in simple applications at the bottom of a market and move up from there.

The theory names two places to take root. A low-end foothold is a low-cost business model that enters at the bottom of an existing market, serving customers who are offered more than they want to pay for. A new-market foothold creates a new segment, and its first customers are people who were not buying at all.² A 2018 paper from the institute puts the second case plainly: new-market disruptions target nonconsumers, and they compete on different measures of performance from the traditional offer.⁴ Both routes matter. This essay keeps to the second, because it is the one an incumbent's customer feedback cannot see.

What keeps the second sheet blank

If nonbuyers are invisible to customer feedback, the useful question is what keeps them out. The answers are usually ordinary. Some lack money: an eight-person dental practice needs written privacy and safety policies, but it cannot pay for a custom engagement priced for a company with a compliance department. Some lack time or expertise: even a free policy template asks the practice manager to know which rules apply. Some lack access: nobody sells to them, because a sale costs more to make than it would bring in. And some lack a use case that justifies the package as it is sold today, because they need one part of it and the firm sells only the whole.

Not every nonbuyer is a customer waiting to be served. A two-person design studio may genuinely need nothing more than a template and an afternoon. Some people do without because doing without is the right choice. The research question is therefore not "who doesn't buy?" but "who doesn't buy because of a barrier we could lower, and would they change what they do if we lowered it?" Figure 1 shows what the answer looks like once orders start to arrive, and why it has to be sorted with care.

A cheaper draft is not a cheaper service

Now suppose the compliance firm builds a second offer: an AI-assisted service for small organizations that could never afford its custom work. The practice answers a guided questionnaire; the system drafts a standard set of policies and training; a person reviews the set before it goes out; and the set is updated when the rules change. The draft that once took a consultant days now costs very little to generate.

It is tempting to stop the arithmetic there, and that is the most common mistake in this kind of plan. Generation is one line in the cost of delivering a useful service, and often not the largest. Someone has to get the questionnaire answered correctly, which is onboarding. Someone has to check that a dental practice did not receive a policy written for a warehouse, which is quality assurance. Someone has to answer the phone when an inspector arrives, which is support. And someone has to stand behind the result when a policy turns out to be wrong, which is liability: insurance, contract limits and the reviewer's time. The premium engagement bundles all of these into one fee and a relationship. A small-organization service has to pay for each of them out of a much smaller price.

None of this means the service cannot work. It means the claim that it can is a claim about the whole delivery model, not about the cost of producing text. There is a reasonable economic reason to expect some demand. OpenAI's April 2026 framework for thinking about AI and jobs asks, of each occupation, whether demand will grow enough to absorb productivity gains if AI lowers the cost of the service, and it gives the example that more affordable tutoring could lead more families to buy it.⁹ That is a mechanism worth taking seriously. It is not evidence that a particular firm's particular service has a market, and its authors are clear that their categories are not predictions.⁹ Figure 2 puts the mechanism and its main hazard on one sheet.

With the support line at $4,000 a year, the simpler service costs $11,000 to deliver, and the sheet shows a band of organizations that could pay that much but never the $40,000 of the premium engagement. That band is the blue page of Figure 1, as a model. Now look at the faint pencil line near the bottom. It is what the plan looks like if generation is the only cost anyone counted: almost every organization on the sheet appears reachable. That line is the optimistic slide in the board deck, and it is wrong.

Then raise the support control. Small customers often need more help per dollar than large ones, because they have no compliance staff of their own. As support rises, the blue line climbs and the band narrows. Somewhere above $30,000 a year it meets the gold line, and the apparent new market disappears entirely: the simpler service now costs as much as the premium one and reaches nobody new. Nothing about the generation cost changed. The opportunity was erased by the part of the service the plan had treated as an afterthought.

Moved, downgraded or new

Suppose the firm launches anyway, and orders arrive. The next mistake is to count all of them as the new market. Each sale belongs on one of the three pages of Figure 1. A buyer who left a competitor for a cheaper version of the same thing is substitution: real revenue, but a share shift in an existing market. A buyer who moved down from the firm's own premium service is cannibalization, and it may cost more margin than it brings in. Only a buyer who had never bought anything like this before is new consumption, the evidence the second sheet was looking for.

These cannot be told apart from the revenue line. They have to be asked about, at the point of sale: Who did this for you before? What did you spend on it last year? What would you have done if this did not exist? The answers will be imperfect, and some buyers will not know. But a launch report that says "two hundred new customers" without saying how many of them were previously buying nothing has not yet said whether access created anything.

A bounded exploration

The practical proposal is modest: treat the small-organization service as an experiment with its own economics and its own success criteria, written down before launch. The criteria should include the share of buyers who are first-time buyers, the full cost to serve a customer by the sixth month, including the support that arrives after the honeymoon, retention at renewal, and the rate at which premium accounts move down. A threshold for each, set in advance, turns "promising" into a decision.

Separate evaluation matters because of the arithmetic Christensen described. A new service that might bring in a small fraction of the firm's revenue in its first years will lose almost any comparison against the roadmap's next enterprise feature, judged by the mature business's margin thresholds. Christensen's own remedy was to place the new venture in an autonomous organization that can be rewarded by small wins and small customers.¹ That is an option, not a command. A separate unit loses shared knowledge, the brand and the sales channel, and sometimes a team inside the business with its own scorecard and protected budget is enough. What the exploration needs is not a new building. It is a scorecard that the main business's scorecard cannot overrule by default.

How far to trust the theory

A framework this influential deserves a fair reading of its critics. In 2015 Andrew King and Baljir Baatartogtokh surveyed 79 industry experts about 77 examples of disruption drawn from two of Christensen's books, asking whether each case showed four elements the theory requires: incumbents improving along a trajectory, that improvement overshooting what customers needed, incumbents able to respond, and incumbents then faltering.⁶ By their count only seven of the 77 cases, about 9 percent, matched all four.⁷⁸ The element most often missing was overshooting, present in 17 cases.⁸ Their conclusion was that the theory may not fit as many situations as is often assumed.⁶

Christensen and his colleagues have themselves argued that the term is widely misapplied. In a 2015 article with Michael Raynor and Rory McDonald he argued that Uber, commonly hailed as a disrupter, did not fit the theory.³ A 2018 review in the Journal of Management Studies, with McDonald, Elizabeth Altman and Jonathan Palmer, describes a concept that gained wide currency despite widespread misunderstanding of its core principles, and traces its development from a framework about technology change toward a broader causal theory, with open research questions still ahead.⁵ Read together, the critique and the defense point the same way. The theory is a developing body of ideas with boundary conditions, not a law, and a list of famous cases chosen after the fact is not a forecast for the next entrant.

That is exactly how this essay uses it. Christensen's book is a good source of questions about who is excluded and why. It does not tell anyone which company will win.

Objections, considered

Nonbuyers may have no meaningful demand. Often true. Many people do without a service because it is not worth much to them at any price. That is why the second sheet is a research question rather than a target list. The test in Figure 2 is whether a buyer can pay the full cost of a useful service, and the test in Figure 1 is whether buyers are actually new. A market that fails both tests should be dropped, quickly and without regret.

Premium customers finance everything else. They do, and they also reveal valuable new needs; some of the best products began as one demanding customer's request. Nothing here argues for neglecting them. The argument is narrower: their requests describe one kind of opportunity, and a roadmap made only of their requests will describe only that kind.

Entrants usually fail, and incumbents can respond. Both are true, and the critics' evidence supports them. Most cheaper offers do not grow into anything, and incumbents that see a threat clearly can often meet it. That is a reason for incumbents to explore deliberately, not a reason for entrants to expect victory. A startup cannot borrow destiny from a theory, and weak unit economics are not excused by calling them disruptive.

A cheaper offer may cannibalize profitable work. It may, and the graphite page exists to catch it. If most early buyers turn out to be premium accounts moving down, the service is a discount, not a new market, and the experiment's criteria should say so before anyone has to argue about it.

The people who cannot buy from you do not appear in your customer feedback.

Before the next roadmap review

• Who does without? Name the organizations or people who need the outcome your product delivers but buy nothing in the category today.
• What prevents purchase? Separate money, time, expertise, access and fit, and say which one a new offer would actually lower.
• What does it cost to serve them, all in? Price onboarding, support, quality assurance and liability, not only generation, and find the support level at which the opportunity disappears.
• What would prove new demand? Decide before launch how you will tell first-time buyers from switchers and downgrades, and what share of first-time buyers counts as success.

Back to the conference table. The roadmap stays as it is; the accounts on it are real money, and their requests are good ones. The second sheet gets something written on it at last, though not a market forecast. It gets four questions, a cost line that includes support, and a note at the bottom about how the first hundred buyers will be sorted. That is not a prediction that anyone on it will buy. It is the start of finding out.