There Is No Such Thing as a Good Roadmap.

You spent three weeks listening. Sales wanted the integration that would close the quarter. Engineering wanted a month to pay down the platform debt that was quietly slowing everything else. The head of product wanted the feature that had been on every slide since January. The churn dashboard wanted something different again. You took all of it, weighed it honestly, and produced a roadmap. The reaction in the room, when you presented it, was not an argument. It was a flat, shared disappointment. Everyone had got a little of what they asked for and nobody had got the thing they came in for. You went home fairly sure you were bad at your job.

You were not. What happened in that room has a proof, and the proof is older than the profession.

The part that was never your fault

Take the smallest version of it, away from roadmaps entirely, because the shape is easier to see when nothing is riding on it. Three friends are choosing where to eat. Ana would take Thai over Italian, and Italian over sushi. Ben would take Italian over sushi, and sushi over Thai. Carla would take sushi over Thai, and Thai over Italian. Three sensible orders, nobody being difficult. Now vote on it, a pair at a time, two against one. Thai beats Italian. Italian beats sushi. So Thai surely beats sushi. It does not: sushi beats Thai. There is no winner. The group’s preference runs in a circle.

It is rock, paper, scissors. Rock beats scissors, scissors beat paper, paper beats rock, and asking which is strongest misreads the game, because the beating runs in a circle and never stops. Three rational people produced a group preference that eats its own tail, and not one of them did a thing wrong.

This has a name. It is the Condorcet paradox, after the Marquis de Condorcet, who set it out in 1785, though a Catalan monk named Ramon Llull had found the same loop five hundred years earlier and the world managed to forget. Three voters and three options are all it takes, and no amount of skilled facilitation dissolves it, because it is a property of adding preferences together, not of the people doing the adding.

Now put the stakes back, and it gets worse rather than better. A roadmap looks like the three friends, three teams and three things worth building, and it is tempting to say the loop is simply why nobody was happy. But real roadmaps rarely loop as cleanly as that, and the reason is worth being exact about. Take tech debt. Only Engineering ever ranks it high, because only Engineering can see it. It is the professional chef wiping down between every step while the gifted amateur leaves the pans where they fall: the cost of the mess stays invisible right up until the kitchen seizes, and it is invisible to anyone who has never had to cook the next order standing in it. So Sales ranks the deal it needs first and the debt nowhere, Product ranks its feature first and the debt nowhere, and the thing keeping the platform alive loses every vote it is in. There is a winner. It is just never the thing that mattered most to the one team that could see the whole board.

And that is still the tidy version, because a real roadmap is never three items. It is which deals out of a pipeline of dozens, which features, each touching corners of the product some customers live in and others have never opened, which fixes, wanted fiercely by the few they affect and unnoticed by everyone else. Hundreds of options, a dozen constituencies, and here is the part that turns a bad afternoon into a theorem: piling all of that on does not average out into a clearer answer. It makes the group’s preference less stable, not more. There is a proof of exactly this, the chaos theorem, shown by Richard McKelvey in 1976 and extended by Norman Schofield in 1978: once the choices spread across several dimensions, majority rule has no stable resting place at all. Most of the time there is no winner, any option can be reached from any other by some sequence of votes, and whoever gets to set the order of the voting can steer the result almost anywhere. That is not a metaphor for why the roadmap is reopened every quarter. It is the machinery of it. The loop between three friends was the smallest visible piece of an instability that only grows with the stakes.

So much for counting your way out. The other exit, a cleverer method that is simply fairer, was shut from the opposite direction decades ago. In 1951, in a book called Social Choice and Individual Values, an economist named Kenneth Arrow wrote down four things you would want from any fair way of combining preferences. It should cope with whatever opinions people actually hold. If everyone prefers A to B, it should too. It should not just copy one person and ignore the rest. And its choice between two options should not flip because of some third, unrelated one. Four conditions, each the least you would ask. Arrow proved that the moment there are three or more options, no method keeps all four at once. Not one that exists, not one anyone will build. It won him a Nobel Prize in 1972, and it means the perfect roadmap process you have been half-hoping to find has been proven not to exist.

The picture I keep for it is a blanket that is too short. Cover your feet or your shoulders; no amount of pulling covers both. Every way of building a roadmap leaves something uncovered, and the question was never how to cover everything. It is which part you leave in the cold.

The thing you are actually arguing about

Here is what Arrow lets you see that the argument in the room does not. Hold everyone’s preferences fixed, change nothing about what anyone wants, swap only the method you use to combine them, and the roadmap changes. Different counting, same people, different plan. Which means a great many roadmap fights are not about priorities at all. They are about the counting method, run by people who have not noticed that is what they disagree on.

Watch the same three teams under three different methods.

Count it first past the post, loudest voice takes the quarter, and you get the roadmap of whoever has the most senior champion or the angriest customer. Decisive, fast, and it ships the thing with a plurality of fans and a quiet majority who would have taken almost anything else.

Count it by compromise, knock out the most-hated option and redistribute until one thing is left standing, and you get the beige result: the item nobody objects to and nobody actually wanted. It is the restaurant everyone in the car agrees on, chosen because it is the only one no one will veto.

Count it proportionally, everyone gets a slice sized to their pull, and you get eleven things sixty per cent built and nothing a customer can use. A quarter of broad, fair, useless progress.

Same wants. Three roadmaps, all different, and the only thing that changed was the arithmetic. The method is a real decision with real consequences, and almost nobody makes it on purpose; it gets inherited from whoever happens to be running the meeting. The single most useful thing you can do is say out loud which method you are using before you start, because half the time the room finds it agreed on the priorities all along and was only ever fighting about the count.

Pick what you break, and say it

If no method keeps all four of Arrow’s conditions, then every roadmap you have ever shipped quietly broke one. You did not notice because the breaking was silent. The skill is to make it loud.

A roadmap set by one decisive leader breaks the no-dictator rule. Early in a company’s life that is often the right trade: when speed matters more than agreement, one person deciding beats six people converging. The failure is never that a founder made the call. It is dressing the call as a fair process it was not, so everyone else spends the quarter feeling cheated by a vote that never happened.

A roadmap set by full consensus breaks decisiveness. It is scrupulously fair and it ships almost nothing, because the price of offending no one is agreeing on next to nothing.

The Dutch built a country on that trade and gave it a name. The polder model, after the low wet land whose dykes only hold if everyone keeps maintaining them, politics or no politics, runs consensus at national scale: government, employers and unions bargain until they agree. It buys a legitimacy and a durability much of the world envies, and it costs a decisiveness that can run to years. There is even a verb for it, poldering, used as praise and as complaint in the same breath. That is the whole move performed in the open: a choice to spend decisiveness on alignment, made deliberately, named plainly, and owned by everyone who lives under it. Most roadmaps make the opposite trade by accident and then hide it. The thing worth taking from the Dutch is not the consensus. It is the naming.

And the one everyone has lived through: a settled roadmap that reshuffles the instant a new request appears, not because anything real changed but because the list of options did. That is Arrow’s fourth condition, independence, being broken in front of you. An unrelated idea drops in and the old rankings rearrange around it. You have watched a single fresh ticket quietly reorder a quarter’s work, and now you know it has a name.

None of these is a mistake to be stamped out. They are the price of aggregating anything, and the theorem says you will pay one of them. What separates the people who are good at this from the people who feel like frauds is not that the good ones found a method that pays nothing. It is that they choose which price to pay, on purpose, and name it in the room. “I am making this call top down because we have to ship, and here is the fairness I am spending to do it” is the same decision as a quiet top-down roadmap. It just arrives as an owned choice instead of a hidden one, and the whole feeling in the room turns on that difference.

The other place it goes wrong

So far this is about how you count. The second failure is about what you count, and it is quieter and does more damage.

Roadmap arguments are full of bets that pass for facts. “This market is about to grow.” “This pays for itself in eighteen months.” And the classic, often repeated but equally often not argued with sufficient quantitative evidence: “This one is strategic.” Each is a guess about a future nobody can check, and each gets set on the table next to the churn number and the signed contract as if it had the same standing. It does not. One is a measurement. The other is a hope with a confident voice.

The machine that does the laundering is the scoring framework. RICE, ICE, weighted scorecards: take a guess at reach, multiply by a guess at impact, divide by a guess at effort, and out comes a number to two decimal places. The arithmetic is real. The inputs were hunches, and the hunch is no truer for having been multiplied. What the framework produces is not a measurement at all, and the confidence around it was manufactured by the sum.

The fix is not a better framework. It is a line down the middle of your inputs. On one side, the things anyone can check and nobody can seriously dispute: churn, ticket volume, the contract you have already signed, the dependency that blocks the other work. On the other side, the bets: what the market will do, what the customer will want, what pays back and when. Keep them apart. Argue the bets as bets, out loud, as the judgement calls they are. The sin is not making bets; you cannot build a roadmap without them. The sin is letting a bet borrow the authority of the number sitting beside it.

This is the same error I have written about in other rooms: a finding’s confidence spent as though it were confidence about a different question, a claim helping itself to a certainty it was never owed. It turns up wherever something subjective needs to win an argument and reaches for the voice of something objective to do it.

What to do on Monday

Notice what is deliberately not here: another prioritisation framework. The lane is full of them, and as above, the framework is often where the laundering happens. Three moves instead, none needing a tool.

• Name the method before the fight. Say which way you are going to count, out loud, at the start, so the argument is about priorities and not secretly about arithmetic nobody agreed to.

• Declare the price. Say which of Arrow’s conditions you are spending this quarter and why: we are deciding this top down for speed, or we are taking the slow consensus because trust matters more than pace right now. Own it.

• Split the inputs. Draw the line between the facts and the bets, put both on the table, and defend the bets as bets. Let the checkable things be unarguable and let the judgement calls be visibly judgement calls.

Do these and the roadmap stops being a thing you defend as correct, which it can never be, and becomes a thing you defend as a deliberate choice with named trade-offs, which it always was. Same document, completely different conversation, because now the disappointment in the room has somewhere honest to go.

The roadmap nobody liked was never evidence that you are bad at this. It is a mathematical certainty you have been taking personally for years. The job was never to make everyone happy; there is a proof against it, filed in 1951. The job is to choose your compromise on purpose, pay the price you meant to pay, and say plainly which one it was. There is no such thing as a good roadmap, not if good means the one everyone wanted; there is only an honest one, and that is enough. Do that and you have not solved the impossibility, because it cannot be solved. You have just stopped apologising for a theorem.