Disruptive Innovation: Why Good Companies Lose

Disruption is not a company being beaten. Disruption is a business serving its most profitable buyers better every year, correctly, on arithmetic that comes out the same way each time, and finding that the run of correct answers left it with no position in a set of buyers it was right to decline. The mistake is not in any one year, and that is what makes it hard to see.

Everybody uses this word. What does it actually name?

Read enough notes and the word arrives attached to any competitor anybody in the room finds alarming. A rival that grew quickly is called disruptive. A rival that cut its price is called disruptive. A rival nobody saw coming is called disruptive, and so, six months later, is one everybody saw coming and did nothing about. When a single word covers all of that, it has stopped separating anything, and a word that separates nothing cannot be checked by the person reading it.

So the first job is to take work away from the word rather than to add to it. Most losses are not this. A business beaten on price, on service, on speed, or on the plain execution of the same job by somebody doing that job better, has lost to competition. Losing to competition happens constantly, in every trade, to careful people, and it needs no special word at all. Saying so plainly matters more than it looks. The ordinary case is the common case, and treating the common case as the rare one is where most of the trouble on this subject begins.

One neighbouring case is worth putting aside at the outset. When a business's own newer line takes sales from its own older line, that is buyers moving between two things the same business sells, and it is tested by looking at what the business's own sales did on both lines. Sales moving between a business's own two lines are covered separately under Cannibalisation: When Your New Product Eats Your Old One. The mechanism worked through below is about buyers the business never served being served by somebody else. Two different events, two different tests, and the arithmetic below fits only the second of them.

Clayton M. Christensen set out the account worked through here in The Innovator's Dilemma, published in 1997. The frame is borrowed and no figure is taken from it. A book is not a source for a number. Everything numerical below is arithmetic, worked from first principles so that it can be checked rather than accepted.

What has to be true before any of this can happen?

Four things, and none of them requires anybody to be bad at their job. The first is that a business has a limited ability to serve and must decide where it goes. Hours, floor space, attention, and people who know the work are all finite, and every period something has to settle where they are pointed. Settling where that ability goes is resource allocationHow a business settles where its limited ability to do work actually goes. Hours, space, people and attention are finite, so something has to decide, and in most businesses that something is a set of figures rather than a person., and in a well run business it is done on figures rather than on preference. The process that produces the decision described here is the same process that produced every good period the business ever had.

The second is that a comparison of two returns points the same way every time it is asked, however small the difference between them is. If one set of buyers gives back more for each unit of ability spent on it than another does, this period's answer is the first set, and it is still the first set next period, and the period after that. Nothing inside the comparison remembers how many times it has been asked. The mechanism does not need a large difference between the two. The mechanism needs a consistent one, and consistency is the least intuitive part of the whole account.

The third is that a run of correct answers is not the same thing as a correct position. Each decision is judged against that period's figures and each one passes. No single decision is ever asked what the whole run of them adds up to. The question belongs to no period and therefore appears on no period's papers. The error has nowhere to live, and that is exactly why it survives review.

The fourth is the part that keeps this a conditional rather than a prediction. Nothing in the arithmetic says the declined set of buyers ever becomes worth serving. If what they need is never met by what they are already buying, then the run of correct answers was correct in the sum as well as in each step, and the business did precisely the right thing throughout. The mechanism is a shape of risk rather than an account of what happens, and writing it as an inevitability would turn a conditional into a prophecy and invent the only thing that mattered.

Why would a well run business turn down a set of buyers?

Start with the part that reads backwards. The change that matters in this account is worse, not better, on the measure the best paying buyers judge by. The change is slower, or coarser, or it does less, or it fails at something the paying buyers regard as basic. Being worse on that measure is the whole reason the change is safe to decline, and the whole reason declining is a trap.

Three parts and no more. First, a product that underperforms on the established measure. Second, a set of buyers for whom that underperformance is good enough, usually because they were being served badly or not served at all, so their requirementWhat a set of buyers needs a thing to do before it is usable to them at all. It is a threshold rather than a preference, and below it nothing else about the thing counts for very much. is a lower one and the cheaper, cruder thing clears it. Third, an established business, the incumbentThe business already settled in a trade and already serving the buyers who pay best in it. The word records only who was there first and is there still; it claims nothing about size, skill or quality., that can see the whole arrangement perfectly well and correctly wants nothing to do with it. The established business is not blind. The business is looking straight at the arrangement and doing the arithmetic.

Picture a food stall outside one office building. The regulars take a full plate at lunch, they come every working day, and they pay for it. A trickle of small orders also comes in from the flats behind, a spoon of this and a scoop of that, each one taking the same handling time and paying a fraction of a plate. The stall turns them down. Turning them down is the correct call at every single lunchtime, on figures anybody can check, and nobody at that stall is being short sighted or lazy. The mechanism below is that stall, written out with the arithmetic attached.

What does that decision look like as arithmetic?

Read the arithmetic below as machinery rather than as anybody's accounts. The construction sets out a property, it describes no business at all, and it has been stripped deliberately of a name, of a trade, of a country and of any period. Where a real case would name the work being done, the construction names only a unit of ability to serve.

One unit of that ability. Two sets of buyers it could be spent on. The first set returns Rs 46.20/- of contribution a unitThe money one sale leaves once every cost that moves with it has been met. What stands there is available to carry everything that does not move with volume, and to leave a profit after that., and that figure is published in these notes rather than made up here. The second set returns Rs 12.00/-, a demonstration figureA number invented so a piece of arithmetic can be watched working. It measures nothing, and it is marked as invented every time it appears., put here so the property can be seen. The difference is Rs 34.20/- and the ratio is 3.85 to one.

The decision is not close, and the fact that it is not close is what makes the mechanism work. If it were a near thing, somebody would argue about it, the argument would be recorded, and a record is something a later reader can find. A ratio of 3.85 to one produces no argument at all. The ratio produces a line in a note and a signature under it.

Ask what would have to be true for the answer to come out the other way, and the question turns out to have no home. The answer would need a claim about a period nobody has named: that the smaller set will one day be worth more than the larger one. No set of period figures anywhere supplies such a claim. Period figures record what happened and never what was declined. So anybody arguing for the second set has to argue against numbers, using something that is not a number.

What happens when the same decision is taken five times?

The same question comes up five times over. Five is a count chosen so the arithmetic can be checked by hand, not a number of years. No period is named anywhere in the construction. Each time the question is asked, the answer is Rs 46.20/- rather than Rs 12.00/-, and each time it is correct on the figures in front of whoever asked it.

Five times Rs 46.20/- is Rs 231.00/-. Five times Rs 12.00/- is Rs 60.00/-. The run is therefore Rs 171.00/- ahead. One route to a figure without the other is not a full showing of the working, so the second route confirms the first: five times the gap of Rs 34.20/- is also Rs 171.00/-, and the two routes agree to the paisa. Five correct answers in a row can add up to one position nobody chose, and no single answer in the run was the mistake.

The empty frame on the papers is the part worth sitting with. A business reviewing itself period by period is doing exactly what careful businesses do, and it cannot reach the question the run raises, however carefully it reviews. The question is not hard. The question is homeless. Every paper that exists asks whether this period's decision was right, and every one of those papers answers yes.

Two settings are worth reaching for deliberately. Push the control down to Rs 2.00/- and the ratio reads 23.10 to one. A gap that wide is the case everybody imagines when they hear the mechanism described. Now push it up to Rs 46.15/-, where the difference is five paise, the ratio rounds to 1.00, and the bracket between the two running totals has almost vanished. The five turns beneath the bars have not moved. A difference of five paise points exactly where a difference of Rs 34.20/- points. The comparison is a direct one between two figures, and it turns only where they are equal. Go one step further, to Rs 46.20/- itself, and the two are level. Nothing is being declined at that setting, so no dilemma exists there at all.

So where does the loss come from, if no step was a mistake?

Here is the sentence, and it stays exactly as written. If the declined set's requirement is ever met by what it is already buying, then a business that spent every decision correctly holds no position there, and the Rs 171.00/- it correctly earned does not buy one. The one thing the money cannot buy is standing with buyers who were served by somebody else throughout. Money and standing are not the same asset and one does not convert into the other on demand.

The boundaries of the account matter as much as its content. The mechanism fixes no period, no pace and no path, and none of the three can be read off its shape. Whether the declined set's requirement is ever met, and when, are open questions at every step of the arithmetic. The pace at which anything spreads is settled separately under Technology Adoption and Diffusion: What Sets the Pace.

The silence is not evasion. The entire difficulty facing the established business is that this is a conditional it cannot evaluate from its own figures. Its papers record what it sold, to whom, at what return. Nothing in them can tell it whether a set of buyers it never served is about to become a set of buyers anybody can serve profitably. Resolving the conditional would hand the reader the one thing nobody standing in that position ever has, and real businesses would then be measured against a certainty nobody in that position could ever hold.

If the requirement is never met, the run of correct answers was correct in the sum too, and a business that declined the second set was right from the first turn to the last. Both endings sit inside the same arithmetic, and the arithmetic cannot say which of the two is in force. The silence between the two endings is not a weakness of the account. The silence is the account.

How Disruption Affects Established Companies

Look inside the established business while all of this is going on, and name each piece of machinery as it actually behaves. Not one of the four items below is broken. Every one is doing the job it was built to do, and doing it well.

One, the allocation process routes ability to its highest return. That is what it exists for, that is what everyone in it is measured on, and a business whose allocation process did anything else would be worse at everything. Ask it to compare Rs 46.20/- with Rs 12.00/- and it will answer correctly, every time, without hesitating and without needing to be told.

Two, the best paying buyers ask for improvements in the direction the business is already going. They want the thing they already buy to be a little faster, a little better finished, a little more reliable. Listening to them is correct, and listening to them leads, by a coincidence nobody arranged, in the same direction the arithmetic points. Two independent-looking signals agree, and neither of them was ever going to point anywhere else.

Three, the declined set fails every test the business uses. The set is smaller for each unit of ability, thinner in marginA share rather than an amount: some line of profit set against sales and written as a proportion of it. The word carries no meaning until somebody says which costs were taken off first., and harder to serve for each unit. The set does not lose narrowly on one measure while doing well on another. The set loses on all of them at once, so there is nothing to weigh and no trade to consider.

Four, anybody inside who argues for the declined set is arguing against the figures. That is a losing position in a well run business, and it is a losing position for good reasons: businesses that let people override figures with enthusiasm make expensive mistakes constantly, and the discipline that prevents those mistakes is the same discipline operating here.

Every one of the four is the machinery working, not failing, and that is why better management does not by itself change the outcome. A sharper allocation process declines the second set faster. Closer listening to the best paying buyers declines the second set with more conviction. The phrase good companies in the title is meant literally rather than ironically. A careless company would have wandered into the declined set by accident, so the account only works if the company is good.

Take it home for a moment. A household running on one salary sits down every month and pays down the most expensive borrowing first. Paying the most expensive borrowing first is the correct order, it is what any careful person would advise, and every month it is right again. The small, unglamorous thing that would have changed what the household can borrow at all never gets to the top of the list. Such a thing never wins against the most expensive borrowing on any single month's arithmetic. Nobody in that household made a mistake in any month.

Why do the business's own measures point the wrong way?

The claim below is about agreement rather than about accuracy, and the difference matters. Rank the two sets by contribution a unit and the first set wins. Rank them by margin and the first set wins. Rank them by what each does to the standing costA cost whose total does not move when the amount of work moves. It is carried in full at any volume, so more volume spreads it thinner and less volume concentrates it. carried for each unit and the first set wins again. Three measures, three agreements, and every one of the three is correctly computed.

Now ask where the three came from. All three are readings of the same period, taken from the same book of figures, by the same method of division. Three measures built from one set of period figures cannot check each other, and their agreement is close to arithmetic rather than close to corroboration. It arrives feeling like three votes and it is nearer to one, and the feeling of three votes is the thing most likely to end the discussion before anybody has said something uncomfortable.

So what would an independent check look like? An independent check would have to come from outside the period's figures altogether, and the honest answer is that no measure supplies one. The check is the conditional set out two sections above, about whether a requirement is ever met, and no period's figures can settle it. A business looking for one more number to break the tie will not find one. There is no tie, and the tie was never the problem.

Can any of it be seen in the accounts?

The question put straight: what would actually have to be observed before calling something in a set of published accounts a disruption? Four things, and they have to come in order. Each one is meaningless without the one before it.

Walk them one at a time against what a reader outside actually holds, and every cell fills with the same answer. A declined request leaves no line in any account. No statement anywhere records the orders a business turned away, the enquiries it did not answer, or the buyers it decided were not worth the handling. Accounts record what happened. The whole of this mechanism lives in what did not.

Apply the same test to the material in these notes and let it come up empty. Coming up empty is the honest result. None of the four items is published about any business used in these notes, so these notes name no disruption in their own evidence, and no case is manufactured to fill the space. An empty result is a finding rather than a gap, and it is the ordinary situation for any reader of any set of accounts. Producing a case would have required manufacturing all four items at once, and that manufacturing is precisely what is at issue.

So what can a reader outside honestly do?

Two questions are worth asking even though neither can be answered from the accounts. Asking them names exactly what is missing instead of covering it over. One, what sets of buyers does this business decline to serve, and on what measure? Two, is there anything serving those buyers that is worse on that measure and getting better?

Neither question is answerable unless somebody inside chooses to answer it. Being unanswerable is not a reason to drop them. A question that can only be answered by disclosure identifies precisely which disclosure is missing, and a reading that names its own gap is worth more than a reading that fills it. Two questions with no answer in the accounts beat one word with no evidence behind it.

So write the positive version when the answer is not disclosed. Write that the business declines sets of buyers, as every business with limited ability to serve must; that the accounts do not say which sets, or on what measure; and that the reading stops there. Then stop. The temptation at that exact point is to carry on into a label. A label reads like a conclusion, and an admission of ignorance reads like a shrug. The label is the weaker document of the two, and the failure below shows what it costs.