Haute Lumière · The reading

You do not know a method until you have watched it meet one stubborn case

On the case study as the form in which a body of principle submits to the world, and the ten decisions that make one worth reading.

The frame is five questions on one page, fixed before the first case is read. What makes it rigorous is that it was written down first.

THE PARTICULAR CASE

A canon is a body of principle that a field has agreed to carry forward. It is compact by design, and compression is both its virtue and the whole of its trouble. Compression is what makes a principle portable enough to teach in an afternoon. Compression is also what strips out every condition under which the principle was true, every awkward exception its author had in mind, every room where it was first tested and nearly failed. Read a canon on its own and it reads as a set of sentences nobody could reasonably object to. Apply one, and the objections arrive in person.

The distance between those two experiences is where the case study lives. Everyone in practice has met the person who can recite the principle perfectly and still misread the room in front of them, and everyone has met the one who cannot name the principle at all and gets it right every time. Neither of them is the goal. The first has the map and no terrain; the second has terrain and cannot hand it to anybody else. The case study is the instrument that joins the two, because it holds a general claim and a particular situation in the same frame long enough for the reader to see where they touch and where they come apart.

This is why a case study is not an illustration. An illustration is chosen after the conclusion, to make the conclusion easier to hold, and its selection is governed by how well it flatters the point. A case study is chosen before the conclusion, and it is allowed to return a verdict the author did not want. That difference is not a matter of degree. It changes what the document is for, what counts as a good one, and who is permitted to write it. An illustration is a piece of rhetoric with a photograph attached. A study is a trial the principle can lose.

The claim in the phrase, then, is exact. A canon in action is a canon at risk. It has been taken out of the seminar and put into a situation with dates, constraints, budgets, people who did not read the book, and an outcome that will be what it is regardless of what anybody hoped. Whatever comes back from that situation is the only thing the canon actually knows. The rest is an assertion in good standing, waiting.

What follows is the method that makes such a trial worth running, set out as ten decisions rather than ten instructions. Each of them is a place where a study is usually lost. Each of them can be made well by a careful practitioner with no special resources, which is the point: the case study is the cheapest serious instrument any field has, and it is the one most often used badly because it looks like it needs no preparation at all.

THE OBJECT FIRST

Most case studies are lost before a single piece of evidence is gathered, in the moment when somebody decides to write one up without deciding what it is for. The work then proceeds cheerfully. Material accumulates. Interviews are held, documents are pulled, a timeline is assembled, and at the end there is a long readable account of something that happened, which settles nothing, because nothing was ever put at stake.

The distinction worth keeping is between a chronicle and a study. A chronicle answers the question what happened, and a good one has real value: it is the record, and records are the raw material of everything after. A study answers a narrower and harder question, which is what does this instance settle. Chronicles can be written from curiosity. Studies require a question that could have come out the other way, and the discipline of naming it before the answer is known.

A question worth a study is one you could lose.

The practical form of the decision is to write the objective as a single declarative sentence that is capable of being false, and to write it down where the reader will see it. Not an area of interest. Not a topic. A sentence with a claim in it, such that a reader finishing the study can say whether the sentence held. This is more uncomfortable than it sounds, which is the reason it gets skipped, and the discomfort is the instrument working before any evidence has arrived.

The objective then does a second job that is easy to miss: it governs scope. A study with a clear object knows what to leave out, and leaving things out is most of what makes a document readable. The account of the vendor negotiation, the reorganisation that happened in the same quarter, the personality that everyone remembers — each of these belongs in the study only if it bears on the sentence. Without the sentence there is no principle of exclusion, and a document with no principle of exclusion grows until nobody finishes it.

The test at the end of this decision is simple and it is worth applying literally. Hand the objective to somebody who was not in the room and ask them what would have to be true for it to be wrong. If they can answer, the study has an object. If they can only restate the topic in different words, what exists so far is an interest, and interests are where studies come from rather than what they are.

CHOOSING THE CASE

Selection is not a preliminary to the argument. Selection is the argument, made silently, before a word is written. A set of cases chosen entirely from successes produces a document that will be read as promotional no matter how carefully its prose is weighed, and it will be read that way correctly, because the conclusion was fixed at the moment the shortlist was drawn. The reader who senses this cannot usually name what is wrong. They simply stop believing, which is worse, because there is nothing to answer.

Variety in a case set is structural rather than decorative. What a study needs is not a pleasing spread of industries or sizes but a spread of the conditions that actually determine outcomes: different constraints, different failure modes, different degrees of buy-in, different amounts of time. Two cases from opposite sides of the world that failed for exactly the same reason are one case. Two cases in the same building that failed differently are two, and the second one is where the learning is.

This is the argument for the near-miss, which is the most under-collected case in every field. The instance that almost worked contains more information than the clean success, because in a clean success every factor was present and none can be isolated. In a near-miss the factors separate. Something specific was absent, and the people involved usually know exactly what it was, and they will say so, because the account costs them less than a failure would.

The clean success teaches least, because nothing in it had to be load-bearing.

Then there is the case that went the other way. A study that includes one is doing something a study without one cannot: it is admitting the conditions under which its own principle does not hold, which is the only way a principle becomes usable rather than merely believed. A reader who can see the boundary of a claim can decide whether their own situation falls inside it. A reader given only successes has to guess, and most of them will guess that their situation is the exception, and they will be right often enough that the study will quietly lose its authority.

Whatever rule governed the selection belongs on the page, stated plainly: how many instances were available, how many were examined, how the ones written up were chosen, and what was excluded. This is the study's denominator, and a study that hides its denominator is asking for a trust it has not earned. A study that states one — even an awkward one, even three cases out of four with the fourth unavailable — has told the reader precisely how much weight the conclusion will carry, which is a service and not a weakness.

Most of the work is finding the figure somebody wrote down before anyone knew how it would end.

THE PEOPLE IN IT

A case seen from one desk is a case seen from one desk. This is not a moral point about inclusion; it is a point about evidence. The person who ran the project knows the decisions and not the daily friction. The person who did the work knows the friction and often not the reason. The person who paid knows the constraint that shaped everything and was never told what it cost downstream. Each of them will give a coherent account, and the accounts will not match, and the mismatch is the most valuable material the study will collect.

Departments hold different halves of the same event, and they hold them in different vocabularies, which is why a single interviewer asking a single set of questions gets three versions of one answer rather than three answers. The correction is to ask each person about what they actually touched, in their own terms, and to resist translating too early. Translation is the writer's job at the end. Done at the start, it flattens the very differences that were worth collecting.

The standard to hold is an account that every participant would sign. Not an account they all like — a study that offends nobody has usually been sanded smooth — but one in which each person can recognise what they did and find it described fairly. Where they genuinely disagree about what happened, the disagreement is recorded as a disagreement, with both versions attributed, rather than averaged into a single sentence that nobody said and nobody would defend.

Where two honest accounts differ, the difference is the finding.

There is a rule about language here that is worth stating on its own, because studies break on it more often than on any question of method. Describe the file, never the person. A decision that did not work out is a decision that did not work out. A projection that missed is a projection that missed. A reference that cannot be traced is an unresolvable reference. None of these require a judgement about anybody's intent, and inserting one converts a study into an accusation, at which point every future participant in every future study has learned exactly how much candour is safe.

The last piece is unglamorous and it is what makes the next study possible. Tell people what the document is for, where it will be read, and what will carry their name. Show them the passages in which they appear before it is finished. Nobody is owed a veto on the conclusion, and a study that grants one has stopped being a study. Everybody is owed accuracy about themselves, and the ten minutes it takes to confirm it will buy access to rooms that would otherwise close.

THE SAME FRAME

Structured analysis means one thing in practice: the same frame is applied to every case in the set, in the same order, whether or not it flatters the material. This sounds like bureaucracy and it is the opposite. Comparison requires a constant. Without one, each case is analysed in whatever terms it happens to suggest, the terms drift from case to case, and the conclusion at the end is a comparison of three differently shaped descriptions, which is not a comparison at all.

The frame in this book is the canon itself, and that is what gives the title its second meaning. The canon supplies the questions; the case supplies the verdict. Each instance is read through the same set of principles, in the same sequence, and what comes back is a row in a table that can be set beside every other row. The reader can then see something no individual case could show: which principles held everywhere, which held only under particular conditions, and which turned out to be doing no work at all.

A frame is an instrument, and every instrument has a field of view and therefore a blind spot. The honest move is to name the blind spot in the document rather than discover it later in somebody's rebuttal. A frame built around decisions will not see drift. A frame built around outcomes will not see the case that was abandoned before it produced one. Naming this costs a paragraph and buys the study its credibility, because a reader who can see the edges of the lens can allow for them.

A frame that fits every case perfectly has stopped measuring and started decorating.

The interesting thing happens when the frame will not fit. A case arrives that has no equivalent of the third movement, or in which two of the principles collapse into one, or in which the sequence runs backwards. The reflex is to force it — to nominate something as the missing element and move on, because the table wants filling. The discipline is to record the misfit exactly as it is. A frame that cannot accommodate a real instance has just told you something about the frame, and that finding is usually worth more than the case that produced it.

None of this requires elaborate apparatus. A frame can be five questions on a single page, as long as they are the same five questions every time and they are written down before the first case is read. What makes structured analysis rigorous is not the sophistication of the structure. It is that the structure was fixed in advance, applied uniformly, and reported even where it embarrassed the person applying it.

WHAT ACTUALLY CHANGED

The question every reader of a case study is actually holding is what changed, and the number of studies that answer it is smaller than the number that appear to. An account can describe a great deal of activity, list a great many decisions, and quote a great many people who felt the work was worthwhile, without ever establishing that anything is different now than it was before. Activity is not outcome. It is the most common substitute for one.

An outcome needs two measurements, and the first is the one usually missing. Somebody has to have written down the before figure, at the time, when nobody knew how the story would end. Reconstructed baselines are not worthless but they are soft, because memory reliably adjusts the starting point in whichever direction makes the ending look like an achievement. A study that begins by finding the contemporaneous record — the old dashboard, the quarterly report, the original ticket count — has done the single most valuable piece of work in the whole exercise.

The second discipline is to take the measurement a sceptical reader would take, rather than the one that shows best. If three metrics were available and one moved, the study reports all three and says which moved. If the improvement is concentrated in a segment, the study says which segment. If the effect appeared in month two and faded by month six, that shape is the finding, and a study that reports only the peak has not made an error of emphasis, it has reported a different result from the one that occurred.

A figure rounded upward for rhythm is a small lie that costs the page everything.

Attribution is the hardest part and the part most quietly skipped. Something changed; the question is whether this changed it. Other things were moving in the same period — a market, a season, a reorganisation, a departure — and a study that does not name them has left its central claim resting on coincidence and hoping nobody checks. Naming them does not weaken the finding. It is what converts an anecdote with a chart into an argument a professional can use, and it also protects the practitioner from being asked, two years later, to reproduce an effect that was never theirs.

Finally, the denominator again, in its measurement form: say what was not measured. Every study leaves something unexamined, usually for good reasons of cost and access, and the reader needs to know where the instrument was not pointed. A study that states its coverage honestly can be believed inside its bounds. A study that implies complete coverage is one discovered gap away from losing everything it established, including the parts that were entirely sound.

The near-miss holds more than the clean success, because in a near-miss the factors come apart and can be named.

THE ROOM AFTER

A case study that is read and not argued about has done roughly half of its work. The document is the artefact; the conversation it provokes is the mechanism. This is the part of the method that most organisations skip, because reading is schedulable and argument is not, and it is the part where the transfer from one situation to another actually happens.

The reason is structural. A written case can only describe the conditions the author noticed. The reader's own conditions are different in ways neither of them can predict in advance, and the only reliable way to surface the difference is to put the case in front of people whose situation is not the one described and let them say why it would not work where they are. That objection is not resistance. It is a condition, arriving in the only form conditions ever arrive in.

The sentence that would never work here is the most useful thing anyone says about a case.

Running such a room well takes one main act of restraint on the part of whoever brought the case: not defending it. The instinct when a carefully built study meets its first sceptic is to explain why the sceptic has misread it, and that instinct, indulged twice, ends the usefulness of the session permanently. Everyone present learns what kind of comment is welcome, and thereafter only that kind is offered. The alternative discipline is to write the objection down verbatim and ask what would have to be true for it to be decisive.

It also helps to give the room something to do other than react. A case read cold produces opinions; a case read against a question produces findings. Hand people the frame from the fifth decision and ask them to run their own situation through it before the discussion begins. What comes back is not commentary on somebody else's case but a first draft of their own, which is precisely the output the whole exercise exists to produce.

What the room finds then goes back into the document, which is the part that turns a discussion into a method rather than an event. A dated note of what was objected to, by whom, and on what grounds, appended to the study, is worth more over time than the original conclusions, because it is a record of the boundary being mapped by people who were standing on it. Studies that accumulate these notes become steadily more useful. Studies that do not are as good on their last day as they were on their first.

KEEPING IT CURRENT

A case study is dated on the day it is finished, and two separate clocks start running. The first belongs to the instance: the organisation described keeps moving, the people described move on, and the outcome that looked settled at twelve months may look quite different at thirty-six. The second belongs to the field: the conditions that made the approach sensible — the tools available, the costs, the surrounding practice — shift underneath it, and at some point the study is describing a solution to a problem in a form that no longer occurs.

Neither clock is a reason to write fewer studies. Both are reasons to date them prominently and to revisit them deliberately. A study with a visible date is an honest instrument at any age, because the reader can allow for the interval. A study with no date is a claim about the present made by a document that may be a decade old, and the reader has no way to tell, which means the only safe response is to discount it entirely.

The revisit is often the better study. What happened two years later is a question the original could not answer, and the answers tend to be structurally interesting: the practice that held because somebody made it their own, the one that lapsed the week its champion left, the improvement that survived but for reasons nobody predicted. Follow-ups are cheap — the access already exists, the baseline is already recorded, the frame is already written — and they are rare, which makes a body of work that contains them unusual in a way readers notice.

Any study old enough to be embarrassing is old enough to be worth checking.

There is a rule about how the update is made, and it is worth holding firmly: nothing overwrites the original. A revision is a new stamped edition set beside the first, with its own date, and the first remains readable. This costs a little storage and it buys something that cannot be bought any other way, which is the ability to see what was believed at the time and how it changed. A study quietly corrected in place has erased the most interesting thing about itself, and a reader who discovers the practice will not trust anything else in the collection.

The practical rhythm that survives contact with actual work is modest. Put a review date on each study when it is published. When the date arrives, spend an hour: confirm the figures still stand, confirm the people described still describe it that way, and note what has changed in the field. Most reviews end in a dated line saying that everything held, and that line is worth writing, because a finding confirmed by a second look at a later date is a stronger claim than the original ever was.

THE DIAGRAM'S JOB

A diagram earns its place by carrying something a sentence cannot, and the test is unforgiving because there are only a few such things. Structure, where the relations between parts matter more than the parts. Sequence, where what followed what is the point. Proportion, where the reader needs to feel a difference in size rather than read one. Distribution, where a shape in the data says something no summary statistic preserves. A picture doing one of those jobs is worth a page of prose. A picture restating a sentence already on the page is a decoration that has been given the authority of evidence.

The comparison table deserves particular mention, because it is the natural output of the fifth decision and it is the figure most readers will actually use. Cases down the side, frame elements across the top, one cell each. Built honestly it is the whole study at a glance, including the awkward parts: the empty cell where a case had no equivalent, the row where nothing moved, the column where every case came out the same and the principle turns out to be doing no discriminating work. Those visible gaps are the table's best feature and the reason it should not be tidied.

A figure that agrees with every sentence around it is not evidence, it is upholstery.

The prose deserves the same care as the figures, and it is usually given less. A case study set as an article — a real headline, an opening that lands, movements with their own shape, a closing that does not restate what was just read — is finished in a different sense from one delivered as a deck of bullets. The bullets are faster to produce and they transfer almost nothing, because the connective material, which is where the reasoning lives, was removed to make them fit. What remains is a list of conclusions the reader has no way to evaluate.

Where the rigour goes, then, is the back of the document. Methodology, selection rule, the interview schedule, the frame in full, the sources kept whole and never trimmed to make a layout fit. A reader who wants to check can check, and a reader who does not want to check can read an argument that has not been interrupted every third sentence by its own apparatus. Both readers are served, and the study can be audited years later by somebody who was not there.

Everything in this decision resolves to a single question that can be asked of any element on any page: what would be lost if this were removed. Applied to a figure, it removes most figures. Applied to a paragraph, it removes most summary paragraphs. What survives is short, and studies that survive this question tend to be the ones people still cite when the software described in them no longer exists.

THE TRANSFER

The final decision is what happens between the conclusion and Monday morning, and it is where the majority of well-made studies quietly end. The document is sound, the analysis holds, the reader agrees with all of it, and nothing is done, because agreeing with a finding and knowing what to do about it are separate acts and only the first one was supported.

A lesson learned that cannot be acted on is a sentence, however true. Communication is important. Alignment matters. Engage stakeholders early. Nobody disputes these and nobody has ever changed a Tuesday because of one, because they name a direction without naming a move. The conversion is mechanical and it is not difficult: say who does something, say what specifically changes, and say what to watch to find out whether it worked. Three clauses, and the same sentence becomes something a person can carry into a room.

A finding nobody can act on differently by Tuesday is a finding written for the author.

The conditions travel with the action or the action does not travel at all. A move that worked in the case worked under particular circumstances — a size, a deadline, a skill in the room, a tolerance for the thing going wrong once — and an action stated without its conditions will be adopted by the reader whose situation does not meet them and will fail there, discrediting a finding that was correct. Naming the conditions looks like hedging and is the opposite: it is what lets a reader tell whether the finding is addressed to them.

This is where the reader's own case begins, and it is the real destination of every decision described here. The frame that was applied to the instances in the study can be applied to the reader's own situation in an afternoon, and it will be a worse study, with thinner evidence and a smaller denominator, and it will be about the one situation the reader can actually change. That trade is nearly always worth taking. The purpose of a well-made case is not to be admired; it is to be a template somebody else picks up and uses badly at first.

Which returns to the beginning, and to the claim underneath the whole method. A canon is not a set of principles that were established and then applied. It is the accumulated residue of principles that were applied and survived it — every sentence in it a compression of some particular room where it was tested and held, and every sentence steadily losing the conditions that made it true as it is repeated. The case study is how the conditions get put back. A canon that has stopped producing them is not a canon any more. It is a catechism, and the field will go on nodding at it for years after the last person who could put it into action has retired.

Free to read

Free to read, and free to hear. Every chapter of every book in this house, and every narration of it, is open to anybody. No account, no card, nothing to cancel.

What is in it

• Case Studies: The Canon in Actionopen this in the house search

A principle that has never cost anybody anything has never been tested.

The case you would rather not write up is the one worth writing up.

Measurement is how a claim agrees in advance to be wrong in public.

A canon stops being a canon the day it stops being risked.

Keep looking

Every phrase on this page opens into the house search. The shelf holds The Developmental Canon and six other shelves, and the reading is free.