- 1. Stop trying to challenge your assumptions
- 2. Find the beliefs nobody defends
- 3. Identify which of the four kinds you have
- 4. Test whether the belief has a cost
- 5. Find out where it came from
- 6. Run the smallest possible experiment
- 7. Design the test so it can fail
- 8. Publish the result before you argue about it
- 9. Expect the belief to survive the evidence
- 10. Change the process, not the minds
- 11. Know which ones to leave alone
The standard advice is to challenge your assumptions. It appears in every strategy book and it is close to useless as an instruction, because the defining property of an assumption is that you cannot see it. Telling somebody to challenge what they cannot see produces either a workshop full of things everybody already doubted, or nothing at all.
What follows is a procedure instead. Eleven steps, in order, for finding a belief your whole industry holds, working out whether it is costing you anything, and testing it cheaply enough that being wrong is survivable. It is written for someone who has never done this deliberately, because most operators have not.
1. Stop trying to challenge your assumptions
Assumptions you can name are not the problem. Anything a leadership team can list in a workshop is already contested somewhere in the business. The beliefs that cost you money are the ones that never come up, because they are not experienced as beliefs at all. They are experienced as facts about the world.
The practical consequence is that the search has to be indirect. You cannot ask people what they believe and get a useful answer. You have to look at behavior and work backward to the belief that would have to be true for that behavior to make sense.
This is why the exercise is usually done badly. A facilitated session produces a list of stated assumptions, everyone agrees they are worth revisiting, and nothing changes, because the list contains only the beliefs that were already visible. The valuable ones survive the workshop untouched precisely because nobody thought to say them out loud.
There is a second reason the direct approach fails, which is that stating a belief out loud in a room changes its status. An unexamined rule is simply how things are done. The moment somebody names it as an assumption, it becomes a position, and positions acquire defenders. You will find that a rule nobody had ever thought about is passionately held about four minutes after you ask whether it is true, and the passion is a response to the framing rather than to the rule.
So the sequence matters. Find it quietly, price it quietly, test it small, and let the result be the first time most people hear the belief stated as a belief. Every one of the eleven steps below is arranged around that principle.
For a contrarian bet against an entire industry’s assumption, see my podcast episode the Charif Souki Cheniere LNG conviction audit.
2. Find the beliefs nobody defends
Look for rules that everybody follows and nobody can source. The signature of an industry orthodoxy is that when you ask why, you get an answer about consequences rather than about origins: not “because we tested it,” but “because that is how it works.
Three places to look. Look at what you never quote on, because a category of business you decline automatically is a belief with a boundary. Look at your minimum order quantities, minimum volumes, and minimum margins, since every one of those is a number somebody chose once. And look at the sentence that starts “customers won’t,” which is the single most productive phrase in this exercise.
Then apply the source test. Ask three people why the rule exists and see whether the answers converge on a reason or on each other. Converging on a reason means the belief has a basis, whether or not it still holds. Converging on nothing, or on the fact that it has always been so, means you have found something worth the rest of this procedure.
The other useful signal is emotional temperature. A belief that gets a slightly irritated response when questioned is more promising than one that gets a considered answer, because irritation is what people produce when they are asked to justify something they have never had to justify.
A fourth place to look, and the most productive one in businesses with a long history, is the exception list. Every operation has two or three customers or products that are handled outside the normal rules, usually for reasons that predate everyone currently working there. Those exceptions are proof that the rule is not physical. Somebody, at some point, made it not apply, and the business survived. Ask why the exception exists and whether it could be generalized, and you frequently find that the orthodoxy has already been broken and nobody noticed.
What you are assembling from all four sources is a shortlist rather than a target. Six to ten candidate beliefs is the right size. Narrowing happens in the next two steps and it happens on evidence rather than on which one irritates you most.
3. Identify which of the four kinds you have
Industry orthodoxies come in four kinds and they fail differently. Inherited constraints, customer proxies, economic orthodoxies, and identity orthodoxies. Naming which one you are holding tells you what kind of evidence would move it, which is most of the work.
An inherited constraint was genuinely true when it was formed and has been made false by something outside your control: a change in tooling cost, in setup time, in shipping, in software, in what a machine can hold. These are the easiest to break and the most common, because nobody revisits a rule that was correct when they learned it.
A customer proxy is a claim about what customers want that has been carried forward from a small number of conversations, often with one large account, often years ago. The tell is that it is stated as a general truth about the market and evidenced by a specific memory.
An economic orthodoxy is a rule about where you can and cannot make money: a volume floor, a margin threshold, a lot size. These are usually artifacts of a costing method rather than of the economics themselves, which is why they are worth testing.
An identity orthodoxy is a claim about what kind of company you are. We do not do consumer work. We do not sell direct. We are not a service business. These are the hardest to break because the evidence does not touch them, and I will say plainly that this procedure works less well on them than on the other three.
The reason the taxonomy matters practically rather than academically is that it tells you what a successful test would even look like. An inherited constraint is broken by a physical demonstration: you run the thing and it works. A customer proxy is broken by asking customers, which sounds obvious and is almost never done because everybody believes they already know the answer. An economic orthodoxy is broken by recosting rather than by producing anything, and the test is analytical rather than operational.
Mismatching the test to the kind is the most common way this work fails. Running a production trial to disprove a belief about what customers want proves nothing, because the belief was never about whether you could make it. Recosting a product to disprove an inherited machine constraint has the same problem in reverse. Name the kind first, then design the test.
4. Test whether the belief has a cost
Most orthodoxies are harmless and you should leave them alone. Before spending anything on a test, establish that the belief is currently costing you something specific, because a belief with no cost is simply a convention and breaking conventions has its own price.
The cost question is concrete. What business did we decline in the last two years because of this rule, and what would it have been worth. If you can construct a number from actual declined opportunities, the belief has a cost. If the answer is that it might be limiting us in principle, it does not, and you should go find a different one.
Quote logs are the best source here. Every business keeps a record of what it declined to quote or quoted deliberately high to avoid winning, and that record is a direct measurement of what your beliefs cost. Almost nobody reads it for this purpose.
Be careful about one thing when you price it. The cost of a belief is what you declined, not what you would have won. Not every quote you did not submit would have converted, and a number built on the assumption that all of them would is a number your CFO will dismantle in the first meeting, taking your credibility with it. Apply your normal win rate to the declined volume and present the smaller figure. It will still be larger than anyone expects.
There is a second cost worth counting, which is the compounding one. A belief that removes a category of business also removes the learning that would have come from being in it, so a business that has not quoted a segment in eight years is not only missing that revenue, it no longer knows how to price it, make it, or serve it. That cost does not appear in any log and it is usually the larger of the two.
One more filter before you spend anything. Ask whether you control the test, meaning whether you can run it without anybody outside your own authority agreeing to participate. A belief that can only be tested with a regulator’s blessing or a customer’s active cooperation is not a candidate for this procedure, however expensive it is. It is a negotiation, it should be resourced as one, and it belongs on a different list with a different owner and a longer horizon.
Applied honestly, the four filters usually leave a business with one or two live candidates. That is the correct number. Operators who finish this stage with six have not applied the cost filter strictly, and the tell is that several of the six are justified by what they might be limiting rather than by what they demonstrably cost.
5. Find out where it came from
Trace the origin before you design the test, because the origin tells you what changed. An inherited constraint that was true in 2011 became false because of something specific, and knowing what that something is turns a general challenge into a precise argument.
The people who know are usually not in the leadership team. They are the long-tenured operators, the retired engineer somebody still calls occasionally, the customer service person who has been there through three owners. Ask them directly when the rule started and what the situation was.
Two things come out of this reliably. First, a date, which lets you ask what has changed since. Second, and more useful, the discovery that the original rule had a qualifier that has since been dropped. A rule that started as “we cannot run that grade on the old line” becomes “we cannot run that grade,” and the old line was replaced six years ago.
Rules lose their qualifiers over time. What began as “we cannot run that on the old line” becomes “we cannot run that,” and the old line was replaced six years ago.
Origins have a second use, which is political rather than analytical. A belief with a traceable origin can be discussed as a decision somebody made under conditions that have changed. A belief with no origin can only be discussed as a claim about reality, and arguing with someone about reality goes badly regardless of who is right.
So when you find the date, use it. Saying that a rule was set when the plant ran a different line is a completely different conversation from saying the rule is wrong. The first invites the room to reason with you about what has changed since. The second invites them to defend a position they did not know they held until you attacked it.
6. Run the smallest possible experiment
The test should be small enough that failing it costs less than the meeting you would otherwise hold about it. One customer, one order, one run, one quarter. The instinct to run a proper pilot is the main reason these tests never happen.
Small tests have a property that large ones do not: they can be authorized by one person. A test requiring three functions to agree will not survive its own approval process, because every function has a legitimate reason to prefer the current state and only you have a reason to prefer the test.
Pick the customer deliberately. You want one that is representative enough for the result to mean something and forgiving enough that a failure does not cost you the account. That is usually a mid-sized long-standing customer rather than either your largest or a new one.
Set a date to stop. An experiment without an end date becomes a permanent exception, which is worse than either outcome because it produces no evidence and permanently complicates your operation.
One practical note on scope. Keep the test inside a single function wherever possible, because a test that requires two functions to coordinate acquires a schedule, and a schedule acquires a status meeting, and at that point you are running a project rather than an experiment. The whole advantage of the small test is that it can happen faster than the organization can form an opinion about it.
And do not describe it as a test of the orthodoxy while you are running it. Describe it as what it literally is: an order, a run, a quote. The framing comes afterward, once you have a result. Announcing in advance that you are testing a widely held belief guarantees that everyone with a stake in the belief will be watching for reasons the test does not count, and they will find some, because every real test has them.
Choose the smallest unit that still produces a real number. A single order at a real price with a real customer beats a simulated run every time, because a simulation can be argued with on its assumptions and an invoice cannot. If the only version you can authorize is a simulation, run it anyway and treat the result as a reason to authorize the real one rather than as the finding itself.
7. Design the test so it can fail
Write down in advance what result would confirm the orthodoxy, and treat that as the primary outcome. A test designed only to demonstrate that the belief is wrong will demonstrate exactly that, and nobody will believe it, correctly.
State the number before you run. If the belief is that a product cannot be made profitably below a certain volume, name the margin you expect at the test volume and name the margin at which you would concede the point. Both numbers, in writing, before the run.
This is also the step that protects you personally. An operator who breaks an orthodoxy and is right becomes someone with good instincts. An operator who runs a documented test with pre-registered outcomes and is right becomes someone whose method can be repeated, which is worth considerably more to your organization and to you.
Pre-registering the confirming outcome does something else that is worth understanding. It converts the exercise from an argument into a shared instrument, which means that people who disagree with you can participate in designing it. Somebody who thinks the orthodoxy is correct and helps you set the threshold at which they would concede has, in the act of setting it, agreed to be moved by the result.
Invite that participation deliberately. The strongest defender of the belief is the best person to help you write the conditions, and their involvement is worth more than a cleaner experimental design would be. A test the skeptic helped build is a test the skeptic cannot dismiss.
Write down who decided to run it, on what authority, and on what date. This sounds bureaucratic and it is protective. Tests that produce inconvenient results occasionally attract questions about whether they were properly sanctioned, and a one-line record written at the start settles that in a sentence rather than in a meeting.
8. Publish the result before you argue about it
Send the numbers first, with no recommendation attached. The moment a result arrives packaged with a proposal, everyone evaluates the proposal and the evidence becomes a supporting argument rather than a finding.
Give it a few days to sit. What happens during those days is that people who would have opposed a proposal instead find themselves absorbing a fact, and a meaningful share of them arrive at the implication on their own, which is a considerably more durable outcome than being argued into it.
Then bring the proposal separately, and make it smaller than the evidence would justify. Asking for less than the result supports is the fastest route to getting it, and the second test is much easier to authorize than the first.
How you write it matters as much as when. Report what happened rather than what it means: the volume, the margin, the lead time, the customer’s response. Resist adjectives entirely. A result written in plain numbers is difficult to argue with, while the same result described as surprising or significant hands people something to disagree with that is not the finding.
Send it to a wider list than you would send a proposal to. This is the one moment in the procedure where breadth helps, because you are distributing a fact rather than asking for something, and facts distributed widely are harder to quietly contain than proposals are.
Timing matters more than most people expect. Send the result when there is nothing else competing for attention, which in most businesses means not in the week of a month end, a budget review, or a board meeting. A finding that lands in a busy week gets read by three people and referenced by none, and you only get one first distribution.
9. Expect the belief to survive the evidence
A single test rarely changes an industry orthodoxy and you should plan for that rather than being surprised by it. The most common response to a successful test is that the case was special: that customer, that grade, that quarter, that salesperson.
That response is not obstruction and treating it as obstruction is a mistake. It is a reasonable epistemic position, because one result genuinely is one result. The answer is a second test in a deliberately different configuration, which costs you another quarter and removes the objection entirely.
Budget two tests before you expect anything to move. Operators who plan for one and get the special-case response usually read it as defeat and stop, which is how a correct finding gets buried by its own author.
The most common response to a successful test is that the case was special. That is not obstruction, it is a reasonable position, and the answer is a second test rather than an argument.
The second test should differ in the dimension the objection named. If the response was that the customer was unusual, run it with an ordinary customer. If the response was that the quarter was slow, run it in a busy one. Choosing the dimension yourself is tempting and wastes the quarter, because you will pick a variable you already believe is irrelevant and the objection will simply be restated.
Ask the objector to name the condition. Two things follow from that. You get a test that addresses the actual doubt, and you get an implicit commitment, because a person who names the condition under which they would be convinced has said something they will be held to by everyone else in the room.
Prepare for the second response as well, which is quieter and more common than the special-case objection. It is agreement without adoption: the result is accepted, nobody disputes it, and nothing changes because no artifact was touched. That is not resistance, it is the default state of any organization, and it is what step ten exists to defeat.
10. Change the process, not the minds
Once the evidence holds, change the rule in the system rather than trying to change what people think. Minutes, quoting rules, approval thresholds, the standard cost file, the configurator. The belief lives in those artifacts far more durably than it lives in anybody’s head.
This is the step that determines whether the work survives you. A team that has been persuaded reverts under pressure, and reverts completely when the person who persuaded them leaves. A quoting rule that no longer contains the constraint does not revert, because reverting it would require somebody to make a decision and put their name on it.
Find every place the old rule is encoded and change all of them in the same week. Orthodoxies regenerate from whichever artifact you missed, and the one you miss is nearly always a spreadsheet somebody maintains privately.
Watch for the artifact that regenerates the rule quietly. In most manufacturing businesses it is the quoting tool or the configurator, because those encode constraints as hard limits rather than as guidance, and a hard limit is invisible to the person using it. They do not experience it as a rule they are following. They experience it as a thing the system will not let them do, which is the most durable form an orthodoxy can take.
Ask whoever maintains that system for a list of every constraint it enforces. The list will be longer than anyone expects and several items on it will have no owner. That document is the closest thing your business has to a written record of its own orthodoxies, and almost nobody has ever read it.
Do the artifact change while the result is still recent. There is a window of perhaps six weeks after a test lands during which changing a rule in a system is uncontroversial, because the evidence is fresh and nobody has yet constructed a reason it should not apply generally. Outside that window the same change requires its own justification, its own meeting, and frequently its own approval, and the effort roughly triples for no additional benefit.
11. Know which ones to leave alone
Leave alone anything where the belief protects a real risk, where you do not control the test, and where the belief is about identity rather than economics. Those three categories account for most of the orthodoxies in any industry and breaking them is not brave, it is expensive.
Safety and regulatory conventions are the clearest case. A rule that looks like an inherited constraint and turns out to encode a hard-won lesson about what goes wrong is exactly the kind of thing a newcomer is most likely to test and least likely to understand. Trace the origin properly before touching anything in that category, and when the origin is an incident, stop.
The second category is beliefs whose test requires somebody else’s cooperation. If you cannot run the experiment without a customer, a regulator, or a supplier agreeing to participate, you do not have a test, you have a negotiation, and it should be sequenced and resourced as one.
And be honest about identity orthodoxies. Evidence does not move them, because they are not empirical claims. They are decisions about what the business is for, and they should be argued as decisions, in the open, by whoever has the standing to make them, rather than smuggled in through a pilot.
Two final tests before you commit a year to this. Ask whether the belief, if broken, changes what the business can sell or what it can make. If the answer is neither, you have found a convention rather than an orthodoxy and the return will not justify the political cost. And ask whether you will still be here in eighteen months, because the process changes in step ten take that long to become permanent and an orthodoxy broken by somebody who then leaves reassembles within two years.
That last point is not a reason to avoid the work. It is a reason to do step ten properly and to do it early rather than saving it for the end, which is where most people put it.
About the Stagnation Assassin
Todd Hagopian is a Fortune 500 transformation executive who has generated $3B in shareholder value across Berkshire Hathaway, Illinois Tool Works, Whirlpool, and JBT Marel, where he serves as Global VP of Product Strategy and Marketing. Known as The Stagnation Assassin, he is the author of two published books: The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn.
The reason most orthodoxies survive is that testing one requires a decision nobody is authorized to make. Book a Decision Velocity Audit and we will find out whether anyone in your business could authorize a one-customer experiment without a committee.
See also: Why Do Profitable Companies Stop Growing?, Why Your 30-60-90 Plan Should Be Shorter