In a perfect market with zero cost, perfect fit and equal numbers on both sides, 36.79% of possible matches still fail. That is the floor, and it is 1 over e. Making it free to reach anybody did not remove that friction, it relocated it onto the receiver, whose attention is fixed, and the receiver answered by filtering instead of matching. If your programme is an annual event plus a note to the base, you are present for roughly 1 of the 20 quarters in which a five-year decision gets made, and 95% of that base is not in the market this quarter regardless. Producing more content cannot fix it, which is why 95% of enterprise AI returned nothing measurable. The only thing that moves the number is removing a cost the other side is paying.

Two parties want the same transaction. One holds a budget, a date and a requirement. The other holds the inventory, the capacity and a price they would happily accept. They never meet. Both walk away concluding they are bad at their jobs.

Neither is. What they have run into is a modelled phenomenon with a Nobel Prize attached to it, an exact failure rate that can be derived on paper, and a physics analogy that is not an analogy at all but the same equation. Once you can see the machinery, most of what passes for marketing advice reads like superstition, and the reason the current wave of marketing technology is not working becomes arithmetic rather than opinion.

It also arrives at a decision a great many people are making this quarter, with a spreadsheet of productivity tools open and a column estimating the headcount each one offsets. The model has something specific to say about that column, and it is not the argument either side is usually having.

This is the long version. Math, physics, statistics, the theories, who proved them, and what to do on Monday.

For anyone running a marketing programme, the arithmetic in the section on the annual event is the reason it can feel like it is working while the pipeline says otherwise.

The prize, and what it was actually about.

In 2010 the Sveriges Riksbank Prize in Economic Sciences went to Peter Diamond of MIT, Dale Mortensen of Northwestern and Christopher Pissarides of the LSE. The citation runs five words: for their analysis of markets with search frictions (anything that costs a buyer or a supplier time or money simply to find the other one, before any deal is discussed).

Start with what they were actually studying, because it matters and it is usually skipped. This is not a marketing theory that happens to mention jobs. It is a theory of people looking for work that turned out to describe every other market too.

The question they set out to answer was why an economy can hold large numbers of unemployed people and large numbers of unfilled vacancies at the same instant. Both sides want the same outcome. The match does not occur. The Academy is explicit that the framework generalises past labour into housing, monetary theory, public economics and regional economics, because every market where two parties must locate each other has the same skeleton.

Their opening move dismantles the model almost every marketing plan is quietly built on. In the classical picture, buyers and suppliers find one another instantly, at no cost, with perfect knowledge of every price. That market clears. Nobody holds idle capacity. Nobody goes unserved.

That market has never existed. Search consumes time and money, and the moment it does, some buyers go unserved and some suppliers cannot move what they would gladly move. Not through incompetence. Through structure.

Hold on to the original framing as you read the rest, because everything here runs on the same equation whether the thing being matched is a person to a job, a freelancer to a commission, a supplier to a contract, or a buyer to a product. Same variables. Same failure rate.

In plain terms: In 2010 three economists won the Nobel Prize for working out why a country can have huge numbers of people looking for work and huge numbers of unfilled jobs at the same moment. Their answer was that finding each other costs time and money, and that cost alone is enough to leave people unemployed and jobs unfilled. The same explanation turned out to work for products, services and suppliers, not just jobs.

The 36.79% floor: what fails when nothing is wrong.

Here is the part that tends to end arguments.

Strip out every excuse. Assume zero search cost. Assume perfect product fit, so any buyer would be delighted with any supplier. Assume the numbers are balanced: 100 buyers, 100 suppliers. Remove price, quality, reputation and budget from the problem entirely. The only thing you leave in is the one thing you cannot remove, which is that buyers choose independently and cannot see each other.

Each buyer picks one supplier at random. Economists call this the urn-ball model (the same problem as throwing a hundred balls into a hundred buckets and asking how many buckets stay empty). It is the standard microfoundation (the piece of arithmetic sitting underneath a bigger model) for every serious account of how two sides of a market find each other.

What fraction of suppliers gets nobody?

The chance one specific buyer skips one specific supplier is 1 minus 1 over 100. All 100 buyers must skip for that supplier to get nothing, so the probability is (1 minus 1 over 100) raised to the 100th power. As the market grows, that expression converges on 1 over e, the reciprocal of Euler's number.

1 over e is 0.3679.

36.79% of suppliers get nobody. Not because they were worse. Because two buyers walked into the same door and one door stayed shut.

The same arithmetic runs the other way. Total matches come to 63.21 out of a possible 100, so 36.79% of buyers also fail, standing behind someone else in a queue they could not see, while a perfectly good supplier sits idle across the street.

You can go further and ask not just how many get nobody, but how the whole hundred land. That spread is the Poisson distribution (the standard way of describing how rare independent events scatter across many slots). The share of suppliers receiving exactly k buyers is e to the minus one, divided by k factorial (k factorial just means k multiplied by every whole number below it, so 3 factorial is 3 times 2 times 1). With these numbers it comes out like this:

36.79% receive nobody 36.79% receive exactly one 18.39% receive exactly two 6.13% receive exactly three 1.90% receive four or more

Read that column again. In a market with identical numbers on both sides, no costs, no mismatch and no competition on merit, more than a quarter of suppliers are overwhelmed while more than a third are empty, and roughly 26% of all demand piles onto suppliers who can only serve one of it.

That 36.79% is the floor. It is what remains after you fix everything that can be fixed. Every real friction stacks on top of it.

The practical consequence is blunt. Before you rate anyone's performance, subtract the floor. More than a third of the misses in any market were never anybody's fault, which means a team judged purely on hit rate is being judged partly on arithmetic. It also means reach was never the binding constraint, because the floor exists at unlimited reach and zero cost.

In plain terms: Put 100 buyers and 100 suppliers in a room, make everything else perfect, and let each buyer walk up to one supplier at random. About 37 suppliers end up with nobody, because several buyers happened to pick the same supplier and left others with none. About 37 buyers miss out too. No costs, no mistakes, no difference in quality, and it still happens.

One assumption is doing all of that work. Every supplier in that room was identical, and every buyer chose at random. That is what keeps the floor clean, and it is also the one assumption a brand exists to break. Weight the choosing instead, and the same equation says a supplier holding 5% of a market's attention ends a period empty 62 times less often than one holding a fair share, while everybody else gets measurably worse. That is worked through separately in The Biggest Brands in Media Do Not Spend Less. They Spend Earlier., together with what Procter and Gamble's $9.2 billion advertising line actually buys, and which half of it costs nothing to copy.

The physics, which is not a metaphor.

So the floor exists. The next question is why, and the answer is that this is a collision problem, in the strict sense that chemists use the word.

The equation economists use to count how many matches a market produces in a period is called the matching function (a formula that takes the number of searchers on each side and returns how many pairings actually happen), and it was lifted directly out of chemical kinetics.

The law of mass action (the rule that a chemical reaction runs faster the more densely its ingredients are packed together) says the rate of a reaction is proportional to the concentrations of the reagents. Two substances do not combine because they share a flask. They combine when they collide, and the collision rate is set by how densely each is present. Nothing about wanting to react enters the equation.

The market version says the same thing with different letters, and it is usually written in what is called Cobb-Douglas form (a standard shape of equation in which each input is raised to a power and the results multiplied together). Matches equal A times searchers raised to the power alpha, times openings raised to one minus alpha, where alpha sits between zero and one and simply sets how much each side of the market counts toward the result.

The term to hold on to is A, called match efficiency: how good this particular market is at introducing two people who ought to meet. Everything else being equal, a market with a higher A produces more matches out of the same two crowds.

And like the chemistry, it scales cleanly. The equation is homogeneous of degree one (double every input and the output doubles too), so double both sides of the market and you double the matches, exactly as doubling both reagents doubles the reaction rate.

That single letter A is the one to hold on to. It is the difference between a market where a good match happens easily and one where the same two people never meet. Almost nothing anybody does in marketing touches it.

Two consequences fall straight out, and they are the two that matter to you.

First, how crowded your side is has a direct and negative effect on you. Economists measure the crowding with market tightness (openings divided by searchers, so a single number for how much demand there is per person chasing it). The rate at which any one supplier fills its capacity works out as A times tightness raised to the power minus alpha. Note the sign. Not a small effect, not a tendency. A minus sign. As more suppliers chase the same buyers, your individual chance of filling the work falls, mechanically, with no reference whatsoever to how good you are.

Second, there is a way to see A from the outside. Plot unfilled capacity against unmet demand, year after year, and you get a downward-sloping line called the Beveridge curve (a standing picture of how much unfilled work sits alongside how many unmatched searchers). Where that line sits is set by A.

That gives a genuine diagnostic. If the whole line drifts outward over time, it means the market is carrying more unfilled work and more people looking for work than it used to, at the same time. That is not weak demand. Demand is right there, unmet, on the other axis. It means the market has got worse at introducing people to each other.

Almost every operator diagnoses a demand problem. Most of them have a matching problem, and the two call for opposite responses.

In plain terms: Economists count matches using the same formula chemists use for how fast two substances react. It has three parts: how many buyers there are, how many suppliers there are, and how good that market is at introducing them, which they call A. The crowding part carries a minus sign, so every extra competitor lowers your odds automatically. And if you ever see a lot of unfilled work sitting next to a lot of people looking for work, that is not weak demand. That is A falling, meaning the market has got worse at introductions.

Why working harder makes it worse.

There is a second-hand effect in all this that nobody experiences directly and everybody pays for. Economists call it a congestion externality (a cost your activity imposes on everyone else that you never see on your own books). Anybody who has sat in traffic already understands it: each additional car is not stuck in the jam, it is the jam, for everybody behind it.

The Academy states the market version plainly. Search activity produces effects that no individual participant takes into account. When one searcher raises their effort, it becomes harder for other searchers on the same side, and easier for the other side of the market.

Point that at a crowded category. Every competitor who starts publishing daily makes it harder for you to be found and easier for the buyer to find somebody. Your visibility is not a function of your effort. It is a function of your effort divided by everyone else's, and everyone else is also increasing theirs.

You are not falling behind because you got worse. You are falling behind because the denominator grew. The model predicted that decades before you felt it.

Which means the usual benchmark is the wrong one. Stop measuring this year against last year. The correct comparison is you against the current denominator, and the denominator is the only thing in the equation that has reliably gone up every year of your career.

And the market does not correct itself. Across the work of the 1980s the Laureates showed that in general an unregulated search market does not produce an efficient outcome. Resource use can be too low, and under some conditions too high. What it is not, reliably, is right. Efficiency requires a knife-edge condition, known in the literature as the Hosios condition (each side has to capture exactly the share of the value that it contributed to making the match happen, no more and no less). There is no mechanism that makes that happen on its own, and no reason to expect your category satisfies it.

There is no invisible hand coming. A market with search costs stays broken until somebody removes a search cost.

In plain terms: Your odds depend on your effort compared with everyone else's, not on your effort. So when everybody in your industry starts posting and pitching more, you all finish roughly where you started, and each of you has made it harder for the others. Economists also showed the market will not correct this on its own.

What happened when the cost went to zero.

Here is the part that explains why this feels worse than it used to, and why that feeling is correct rather than nostalgic.

For most of the twentieth century, every attempt at a match was expensive for the person making it. You walked into the building. You typed the letter and paid the postage. You bought the classified inch. You drove to the showroom. You rang a switchboard during business hours. Each individual attempt cost real time and real money.

Expensive attempts meant low volume. An opening drew a handful of applications. A procurement officer phoned three suppliers because phoning thirty was not physically possible. A shopper compared four options because comparing four hundred meant four hundred miles.

Now look at what the math actually reads. The matching equations do not contain a term for how much effort somebody spent. They contain the ratio. Low volume per opening is a favourable ratio, and a favourable ratio produces a high match rate. That world was slow, it was geographically unfair and it was closed to anybody outside the network, and none of that is worth defending. But on the narrow question of whether two willing parties found each other, it worked, and it worked because attempts were scarce.

Then the cost per attempt went to zero. One click applies for a job. One search returns seventy thousand products. One send reaches two thousand contacts at three in the morning.

And here is the inversion, which almost nobody names correctly.

The cost did not disappear. It relocated. It fell for the person initiating and rose for the person receiving. Every application that costs the applicant nothing costs an employer a slice of attention. Every pitch that costs a vendor nothing costs a buyer a slice of attention. Every upload that costs a creator nothing costs an audience a slice of attention.

Total search cost in the market did not fall. It moved to the one side that cannot scale, because attention is fixed. A person can read forty things carefully in a day and that number has not changed since 1970.

What does a receiver do when volume exceeds the attention available? They stop matching and start filtering. And a filter is not a matching process. A filter is optimised to reject cheaply, which is a completely different objective. It discards on keyword, on format, on year, on postcode, on whatever is machine-readable, because reading properly is the thing there is no longer time for.

In the language of the floor, this drives A, the match efficiency term, down. Which means the Beveridge curve moves outward: more unfilled openings and more unmatched searchers at the same time. That is precisely the pathology the 2010 prize was awarded for explaining. It is not the only thing dragging A down, and the sections that follow deal with the others, but this one we built deliberately, by making it free to try.

Diamond 1971 then lands the last blow. A receiver facing a high search cost does not compare carefully and pick the best. They take the first acceptable option and stop. Which hands the outcome to whoever was easiest to find, not to whoever was best. Everyone in the pile behind that first acceptable option was never evaluated at all.

That is the honest comparison between then and now. It used to be hard to reach anybody and reasonably likely that reaching them worked. It is now trivial to reach everybody and very unlikely that reaching them works. The technology optimised the wrong side of the equation, and the math says so plainly, because the math never cared how easy it was to try.

Which condemns almost every plan whose central idea is to contact more people. It is optimising the side that is already free. The scarce resource is on the other side of the transaction, and it belongs to somebody who is defending it with a filter.

In plain terms: It used to cost real money to apply for a job, pitch a client or place an advert, so people did less of it and the few attempts that were made got read properly. Now it is free, so everyone does it constantly. The cost did not disappear, it landed on whoever has to read it all, and they cannot, so they use rules to discard most of it unread. That is why you now get more unfilled jobs and more people looking for work at the same time.

What happens when your side fills up.

Balanced markets are a fiction. Categories run heavy on the supply side, and the math is unforgiving about it.

With B buyers and S suppliers, the probability that a given supplier receives nobody is e raised to the minus B over S. That is an exponential, which means crowding does not hurt you gradually. It hurts you on a curve.

One buyer per supplier: 36.79% of suppliers get nothing One buyer per two suppliers: 60.65% get nothing One buyer per three suppliers: 71.65% get nothing One buyer per five suppliers: 81.87% get nothing One buyer per ten suppliers: 90.48% get nothing

This is the honest answer to why a competent operator with real capability cannot find the buyers who are actively looking for exactly what they do. In a category running five suppliers per live buyer, 82% of suppliers end a cycle empty, and the sorting has almost nothing to do with who was best. Nobody in that 82% did anything wrong. The distribution simply had to put them somewhere.

In plain terms: One buyer for every supplier and about 37% of suppliers get nobody. Two suppliers chasing each buyer and it is 61%. Five suppliers and it is 82%. Ten and it is 90%. The more crowded your industry gets, the more of you finish with nothing, and which ones is mostly luck.

The actual mathematics of matching a need to a product.

Everything so far explains why matching fails. This is the arithmetic of how it succeeds, because a match is not one event. It is five operations in sequence, and each one has its own math.

In plain terms: For a buyer to find you, five things have to go right. You both have to describe things in the same units, or no comparison is even possible. Your offer has to clear every requirement at once rather than most of them, which is far harder than it sounds: six conditions at 80% each leaves 26%. Word-based search cannot handle a requirement like under forty dollars, because that is a limit, not a word. Anything sensible has to filter on the hard limits first and only then rank what survives. And filters throw away good answers whenever good answers are rare. Out of a thousand organisations who could genuinely use you, about eight are still standing before anyone judges whether you are any good.

One. Nothing can be matched until both sides are in the same coordinate space.

This part is borrowed from graphics rather than economics, and it is the operation everybody skips.

When Adobe software draws a shape in the right place on a page, it is not because the shape knows where it goes. Every object is defined in its own local coordinates. The software multiplies it by a transformation matrix that maps those local coordinates into one shared page space. In PostScript and PDF that is the current transformation matrix, six numbers written a, b, c, d, e and f, and applying it puts the object exactly where it belongs. Skip it, or apply the wrong one, and the object does not land slightly off. It lands somewhere else entirely, or off the page, where nobody sees it at all.

A market has precisely this problem and almost never solves it.

The buyer states the need in their own coordinates: two hundred people, mid-March, under forty a head, discreet, nothing branded on the outer carton. The supplier states the offer in theirs: item 4471, five-ounce cotton, minimum order two hundred and fifty, three-week lead, decoration extra. Both descriptions are complete. Both are accurate. They are written in different bases, and until something transforms them into a shared space there is no operation defined between them at all. Not a weak match. No match function.

That transform is the entire job. Everything after it, the filtering, the ranking, the distance, is geometry, and geometry is the easy part. It is the transform that is missing.

Which is what a structured, machine-readable catalogue actually is. Not a nicer website. A coordinate transform. Publishing your price, minimum, lead time, capacity and finish as values rather than as prose is the literal act of mapping your offer into the space the buyer's question is asked in. Do it and you are placed on the page. Leave it in prose and you are not ranked low, you are off the page, for exactly the same reason and by exactly the same arithmetic as a shape drawn without a transform.

And once both sides are in one frame, something changes that is worth stating precisely, because it is the most useful idea in this entire piece.

In physics, displacement is the straight-line change in position from one point to another. It is a vector, it has a magnitude and a direction, and it depends only on the two endpoints. Not on the route. Distance travelled is path-dependent and can be enormous. Displacement between the same two points is the same whether you walked there directly or went around the world first.

Apply that to a market in a shared coordinate space and finding each other stops being a search. It becomes a subtraction. The need minus the offer is the gap, and if that gap is zero on every hard constraint, it is a match, and it was a match the whole time.

Which means the route stops mattering. Geography stops mattering. So does the number of intermediaries, the introductions you never had, the conference you did not attend, and the thousands of miles and millions of records between the two of you. A requirement stated properly can reconnect with a supplier that no chain of introductions would ever have reached, not because anybody searched harder, but because the operation became defined.

And here is the sharp end of it. Displacement is undefined without a shared origin. Two points sitting in different frames do not have a large displacement between them. They have none, in the strict sense that the quantity cannot be computed at all. That is not a description of a difficult market. It is a description of most markets, right now.

It also puts the earlier point about referrals in its proper place. A referral is a path. It works because it physically walks information from one frame into another, one human at a time. A shared coordinate space does the same work without needing the path, which is the only mechanism that has ever scaled.

Two. Fit is multiplicative, not additive. A real requirement is never a single thing. It is a set of conditions that must all hold at once: the budget, the quantity, the date, the compliance, the shipping, the finish. If each condition independently has a probability p of being satisfied and there are N of them, the chance a given item clears every one is p to the power N.

Six conditions at 80% each is not 80%. It is 26.21%. Ten at 80% is 10.74%. Even at a generous 90% each, six conditions leaves 53.14% and ten leaves 34.87%.

This is the exact reason a catalogue of seventy thousand items does not solve a buyer's problem. Breadth raises the count of candidates; it does nothing to the exponent. Fit collapses geometrically while range grows linearly, and the exponent always wins.

Three. Keyword search cannot evaluate a constraint. Classic retrieval scores a document by how many of the query's words it contains, weighted by how rare those words are across the whole collection. That is the family of formulas behind every search box you have ever used.

Now state a real requirement out loud: two hundred people, mid-March, under forty dollars a head, shipped blind, no visible branding on the outer carton. Almost none of those words appear on a product page, and the ones that do carry no meaning as words. Forty is not a term to be matched, it is a ceiling. Word-overlap scoring has no way to represent a ceiling, a date or a quantity, so it returns things that read similar and fails the things that must be true.

Four. What actually works is filter, then rank. The correct order is not negotiable. First eliminate on hard constraints, the ones with a right answer: price ceiling, minimum order, stock on hand, lead time against the date, certification. That is a set operation, and it either runs or it does not, depending entirely on whether those values exist as machine-readable fields.

Only then rank what survives, by similarity to the softer parts of the brief, the style, the audience, the feel. Similarity scoring compares two descriptions as lists of numbers and measures the angle between them, which is a genuinely useful way to capture resemblance. But resemblance is not satisfaction. A vector can tell you two things are alike. It cannot tell you one of them is over budget.

Get the order wrong, rank before you filter, and you produce a beautifully relevant list of things the buyer cannot buy. Most catalogue search does exactly this, and it is the second failure that traces straight back to operation one: you cannot filter on a constraint that was never expressed as a value.

Five. Filters at low base rates are brutal, and this is where the good answer dies. Take a thousand candidates of which 5% genuinely fit, so fifty. Run a filter that is 90% accurate in both directions.

It passes 140 items. Only 45 of them are real. Precision is 32.14%, meaning two out of three things that clear your filter should not have. Tighten specificity to 95% and precision only reaches 48.65%. You need 99% to get to 82.57%.

And every version of that filter also throws away five of the fifty genuinely good answers, permanently and silently, because a rejected candidate never gets a second look. That is the mathematical form of the Harvard finding: 88% of employers know their software vets out qualified people, and the base-rate arithmetic explains why it is nearly impossible for it not to.

Then ranking concentrates whatever survives. Attention across a ranked list is not even, it follows a power law. On a list of a hundred, the top result takes about 19.28% of attention, third place 6.43%, tenth place 1.93%, fiftieth 0.39%. The top ten take 56.46% between them. Ranking eleventh instead of tenth is not one place worse. It is most of the way to invisible.

Put the whole chain together. Start with a thousand organisations that could genuinely use what you do.

5% are in-market this quarter, leaving 50. 26.21% clear all six hard requirements, leaving 13.11. 63.21% survive the coordination floor, leaving 8.29.

Eight. Out of a thousand who could genuinely use you, before a single person forms an opinion about whether you are any good. Quality is applied to the survivors of that chain, not to the thousand.

Which puts the practical conclusion in one line. You cannot raise the thousand, and you cannot argue with the exponent. What you can do is the transform. Publish your hard constraints as values a filter can read, so that you are in the coordinate space where the question is being asked, and you survive to be ranked at all rather than being eliminated by a field you left blank.

Four lives, one equation.

The abstraction is the point, so here it is with the abstraction taken off. Four people, one equation.

The young applicant. Take an opening that draws 250 applications. Reported averages for corporate postings cluster around 242 to 250. Your chance is not one in 250 adjusted for how good you are. It is the coordination lottery, multiplied by a filter you will never see and cannot appeal, running on a document that a person may never open. Roughly 0.4% per application.

Now the cruel part. The standard advice is to apply to more openings. Do the algebra on that. Raising your applications raises your own numerator, and because every other applicant received the identical advice, it raises the denominator for everyone by the same factor. This is the congestion externality with a human face on it. Individually rational, collectively self-defeating, and the reason the process degrades a little more every year while everybody involved is working harder than the year before. Nobody is doing anything wrong. The equation simply does not reward it.

The older worker. Here it stops being a lottery and becomes something else, and this is where the numbers are worth reading twice.

In a controlled field experiment, researchers sent roughly 14,400 applications in response to 3,607 genuine retail postings. The applications were built to be equivalent in every respect that should matter. The variable was age.

Men aged 29 to 31 were called back 23.08% of the time. Men aged 64 to 66, 16.97%. Women aged 29 to 31 were called back 25.51% of the time. Women aged 64 to 66, 17.45%.

That is a relative decline of about 26% for men and about 32% for women, on applications that were otherwise the same. So an older applicant is not only paying the 36.79% coordination floor and the 250-to-1 crowding. There is a systematic filter stacked on top of both, it is measurable, and it falls hardest on older women. When an experienced person says the process feels rigged, the honest answer from the data is that a measurable part of it is, and the rest is a distribution that would have failed them anyway.

The freelancer, the creator, the person offering their own work. You are a supply side of one in a category where supply is effectively unbounded, because the cost of announcing that you are available fell to zero and the number of people announcing it did what the math said it would. Run e to the minus B over S with S climbing and no ceiling, and your probability of finishing a cycle with nobody heads toward 1. Working harder cannot fix it, because your effort is also an input to the denominator that everybody else is fighting. What fixes it is being in the one specific place the commission is actually decided, at the moment it is decided, which is a different activity entirely from being visible.

The buyer, drowning. The last life is the person with the money, who is usually assumed to be fine.

Iyengar and Lepper set up a tasting table with either 24 varieties of jam or 6. The larger display drew more people over. Shoppers were 10 times more likely to actually buy from the display of 6. In fairness, this finding has been argued about since: a 2010 meta-analysis found the link between choice and purchase far from consistent, and a later re-analysis of 99 studies set out the conditions under which it holds rather than treating it as a universal law.

The part that survives every version of that argument is the part that matters here. More options is not more matching. Past some point, an additional option raises the cost of evaluating the set and lowers the chance anything is chosen at all. A catalogue of seventy thousand items is not three hundred and fifty times better than a catalogue of two hundred. Somewhere between those two numbers it stops being range and starts being a tax on the buyer, paid in the only currency they cannot make more of.

Which reframes curation completely. Cutting a catalogue down to the twelve things that fit this buyer is not a merchandising preference or a nice touch. It is a direct reduction of the buyer's search cost, which is A, which is the one term in the entire model that describes a market getting better at introducing people. A curated room is a search-cost intervention wearing a nicer jacket.

Four lives, four situations that have nothing in common socially, and one equation underneath all of them.

Two conclusions follow, and which one applies depends on which side of the filter you are standing on. If you are being filtered, volume is the one lever that provably does not work, because it is also everybody else's lever. And if you are the one doing the filtering, understand what you have built: a machine tuned to reject cheaply, which will keep discarding the right answer for reasons nobody would defend out loud.

In plain terms: The same arithmetic explains four situations that look unrelated: the graduate sending 250 applications, the 64-year-old getting fewer callbacks for an identical CV, the freelancer nobody can locate, and the buyer with too many options to choose between.

Diamond 1971, and why nobody publishes a price.

Diamond published a result in 1971 that remains, fifty-five years on, the most uncomfortable finding in this literature. He modelled buyers searching for the best price while suppliers set prices knowing that buyers must search.

A very small search cost does not produce a slightly less competitive market. It produces the monopoly price (the price a supplier with no competition at all would charge). The Academy puts it without hedging: equilibrium prices equal the price a monopolist would have set in the same market with no search costs at all.

That is not a gradient. It is a cliff. A trivial amount of friction, the cost of one more phone call, collapses the outcome all the way to monopoly. A buyer who cannot cheaply compare does not overpay slightly. They overpay by the entire amount an unchallenged supplier could have extracted.

Which explains, precisely, why every quote in this industry arrives as a range and why nobody posts a number. The opacity is not disorganisation. It is worth the whole margin, and the people holding it know that.

In plain terms: In 1971 Peter Diamond proved that if it costs a buyer even a tiny amount of effort to compare prices, suppliers can charge what they would charge with no competition at all. Not a bit more. The full amount. That is the real reason quotes in most industries arrive as a range instead of a number.

Why the annual event and the note to the base is a 5% strategy.

A team whose programme is an annual event plus periodic contact with the existing base is not underperforming. It is running a structurally losing game, and the size of the loss can be worked out in advance.

Put real numbers through it.

Professor John Dawes of the Ehrenberg-Bass Institute, working with the LinkedIn B2B Institute, established what is now called the 95:5 rule (at any given moment only about five in every hundred of your potential buyers are actually in the market). Because organisations change providers of things like banking, legal advice, software or telecoms roughly every five years, only 20% are in the market in a given year and just 5% in a given quarter. The other 95% are not in the market at all.

Run that against a base of 2,000 contacts.

100 of them are in a live buying window this quarter. Everyone else is unreachable in the only sense that matters, because no message converts somebody who has no requirement.

Now add timing. A five-year replacement cycle is 20 quarters. An annual event plus the weeks of salience after it makes you present for roughly one of those 20. You are structurally absent for 95% of your own base's buying cycle. The 95:5 rule is not really a targeting rule. It is a timing rule, and going through the motions on an annual rhythm means you are silent during nineteen of the twenty quarters in which the decision can be made.

Then stack the 36.79% coordination floor on the residue. Even among the small group who are in-market, awake, and looking at you, 36.79% of the possible matches fail for no reason at all.

There is worse. The CMO Survey found that more than half of companies have increased the number of channels they use, adding digital, social, retail media and face-to-face in parallel. In a collision model that is precisely backwards. Match probability depends on your density in the specific place the buyer actually looks, not on your total presence across all places. Holding budget flat and spreading it over seven channels divides your concentration in every one of them. You have not widened the net. You have thinned the reagent.

This is why the base-nurture and annual-event motion feels like it should work and does not. It is not badly executed. It is executed against a model of the market that the 2010 Nobel Prize was awarded for disproving.

In plain terms: A list of 2,000 contacts, worked with one event a year plus occasional emails, produces under one real opportunity per quarter, roughly three a year. Change nothing except being present all year instead of once, and the same list produces about 66 a year. The difference is the timing, not the effort, the copy, the venue or the budget.

Now close the arithmetic, because this is the actual proof.

Take the base of 2,000 and put every stage through it in order. Each assumption is stated, and each one is generous.

2,000 contacts. 5% are in a live buying window this quarter, per the 95:5 rule, leaving 100. Your annual rhythm means you are genuinely present in 1 of the 20 quarters in the cycle, so the chance your quarter of presence lines up with a given buyer's quarter of need is 1 in 20, leaving 5. 26.21% of those clear all six hard requirements, leaving 1.31. 63.21% survive the coordination floor, leaving 0.83.

Zero point eight three. A base of two thousand, worked diligently on an annual rhythm, produces less than one live, qualified, actually-connected opportunity per quarter. About 3.3 a year. That is the whole return on the programme, and it was determined before anyone wrote a word of copy.

Now change one variable and nothing else. Keep the same base, the same fit, the same coordination floor, and make the presence continuous instead of annual. The timing term goes from 1 in 20 to 1.

16.57 per quarter. About 66 a year. Twenty times the same programme.

That is the proof, and note what it is not. It is not a bigger list, a better event, sharper copy or more budget. Every one of those leaves the timing term untouched, which is why doubling any of them produces almost nothing. The 20x is sitting entirely in the rhythm, and rhythm is close to free. It costs a decision about how the year is organised.

One more term can zero the whole line. If your prices, minimums and lead times are not published as values a filter can read, you are eliminated at the filter stage before the ranking ever runs, and every number above becomes zero regardless of how good the rest of it was.

None of that is an execution problem. The programme is not failing because the event was mediocre or the copy was flat. Fix both and the arithmetic is unchanged. A 5% duty cycle (the share of the total decision window during which you are actually present) against a base that is 95% out of market, spread thinner every time a channel is added, cannot be rescued by execution quality. It has to be re-architected around presence at the moment of the question, and that is a different job with a different budget line.

The AI pitch, and the actual numbers.

Every operator is being pitched software that promises to fix this. Here is what the record shows.

MIT Media Lab's Project NANDA published The GenAI Divide: State of AI in Business 2025, built between January and June 2025 from a review of more than 300 publicly disclosed AI initiatives, 52 structured interviews and 153 survey responses from senior leaders. Against $30 to $40 billion of enterprise generative AI spending, the finding was that 95% of organisations were seeing no measurable business return. Only 5% of integrated pilots were extracting real value. Only 5% of custom enterprise tools reached production at all.

The CMO Survey, 35th edition, directed by Professor Christine Moorman at Duke University's Fuqua School of Business, fielded 7 to 29 January 2026 across 308 marketing leaders, 97% of them VP-level or above, lands in the same place from the other direction. Adoption is climbing hard. Usage has more than doubled in two years. Generative Engine Optimization (the practice of shaping your material so an AI assistant surfaces and cites it, the way search engine optimisation did for Google) is already running at 4 in 10 companies. Companies expect AI to power the majority of marketing activity within three years.

And on execution, this: across a wide range of marketing technology activities, no capability scores above 5 on a 7-point performance scale, and performance has not improved in two years.

Adoption has more than doubled. Performance is flat. Both statements come from the same instrument.

The search model explains why, and the explanation is not that the technology is bad.

Look again at why working harder makes it worse. Your match rate depends on your effort relative to the rest of the market. A tool that raises your output raises the numerator. But this tool went out to your entire category in the same eighteen months, so it raised the denominator by the same factor. Tightness does not move. Your fill rate, A times tightness to the minus alpha, does not move either. Output per person went up across the board and the number of matches stayed where it was, because match rates were never a function of how much anybody produced.

Worse, it feeds the congestion externality directly. Everyone can now generate more, faster, into the same finite attention, which raises the cost to the buyer of evaluating anything. Higher buyer search costs, per Diamond, push the market further toward opacity and monopoly pricing, not away from it. The average participant is not standing still. They are paying for a tool that degrades the commons they depend on.

So what did the 5% do differently? On MIT's account the divide is not model quality, it is integration into an actual workflow that learns. In search terms that is the same finding in different clothing. The 5% used it to reduce a cost somebody else was paying, most often the buyer's cost of evaluating and deciding. The 95% used it to increase their own output. Only one of those two things appears anywhere in the matching function.

There is a supporting detail worth sitting with. In the same survey, training budgets have fallen to 3.8% of marketing spend and headcount growth has dropped by 50% in a year, while the most cited capability shortfall is not a missing skill but a shortage of people, time and budget to make existing capability work. Companies are buying the reagent and cutting the flask.

In plain terms: MIT examined more than 300 company AI projects: against 30 to 40 billion dollars spent, 95% produced nothing measurable. Duke found AI use more than doubled in two years while performance stayed flat. The reason is that everybody bought the same tools in the same period, so everybody produces more and nobody became easier to find.

The tool you are being asked to approve this quarter.

All of which arrives at the decision actually being made right now, in most organisations, by somebody with a spreadsheet. A list of productivity tools. A monthly cost against each one. And, in the column that is doing the real work, an estimate of the headcount each one offsets.

Here is what the model says about that spreadsheet, and it is not what either side of the argument usually claims.

The savings are frequently real. Somebody genuinely does produce in an afternoon what used to take a week. That part is not in dispute and pretending otherwise is how a person loses the room.

The match rate still does not move. Output was never the binding constraint. Nothing in the matching function reads how much anybody produced, and every competitor is running the same upgrade in the same eighteen months. You will get a cheaper cost per unit of output in a market that was never short of output.

Then there is the harder part, and it comes from the two studies already cited rather than from an opinion.

MIT's account of why 95% of enterprise deployments returned nothing is explicitly not that the models were weak. The divide was organisational: whether the thing was integrated into a workflow that actually learns. And the CMO Survey, asking a different question of a different sample, lands on the same obstacle from the other direction. The most cited capability shortfall is not a missing skill. It is a shortage of people, time and budget to make existing capability work at all.

Read those two findings next to the funding plan. The tool fails without integration. Integration is people and time. The purchase is being funded by removing people and time. In the same survey, training has fallen to 3.8% of marketing spend and headcount growth has dropped by 50% in a year.

That is not a moral objection to cutting costs. It is a sequencing error with a measurable failure rate attached, and the failure rate is 95%.

Three questions worth putting to any vendor. They come straight out of the equation, and they are more useful than a demo.

Does this lower a cost that somebody else pays, or a cost that we pay? Only the first one appears in the matching function. A tool that makes your team faster is a margin story. A tool that makes it cheaper for a buyer to work out whether you can do the job is a distribution story, and those are not the same purchase.

When it produces a number, is that number retrieved from a record or generated? If nobody can answer that in one sentence, the answer is generated.

Who integrates it, and out of whose week? If that person does not exist in the plan, the plan is the 95%.

And increasingly the reviewer is also a machine. This is the part missing from most of these conversations. Buyers are no longer only reading pages. They are asking a model to shortlist, compare and summarise, and Generative Engine Optimization is already in use at 4 in 10 companies. Two things follow, and they cut in opposite directions.

If the model doing the evaluating is not wired to a source, it will compose the comparison, including the numbers in it. A procurement decision then inherits the whole hallucination problem, one layer up, with a purchase order attached to the output.

And on the other side: if your own prices, stock positions, minimums and lead times are not machine-readable, an agent cannot surface you. Not ranked low. Absent. For most of thirty years, distribution meant being findable by a person. It is becoming being retrievable by a machine, and the organisations treating that as a marketing question rather than a data question are going to discover the difference from the wrong side of it.

Which is the same conclusion the whole model keeps producing. The thing that moves is never how much you can make. It is what it costs the other side to find out you exist.

In plain terms: A tool that makes your own team faster usually does save money, and it does nothing to help customers find you. MIT found these projects fail because nobody builds them into how the work actually runs, and doing that takes people and time, which is normally exactly what was cut to pay for the tool.

Why the machine needs a pipe to the real numbers.

There is a specific reason so much of this technology returns confident nonsense, and it is not a defect in the reasoning. It is the input.

A language model asked for a figure it has not been handed will produce a figure. The output is shaped to be plausible, and plausible is precisely what a wrong number looks like. It arrives with the same syntax, the same confidence and the same decimal places as a correct one. That is why it survives review.

This is not hypothetical, and it is already settled law in at least one jurisdiction. In Moffatt v. Air Canada, decided by the British Columbia Civil Resolution Tribunal in February 2024, the airline's website chatbot told a customer that a bereavement discount could be applied retroactively to a ticket he had already bought. Air Canada's actual policy required the fare to be requested before booking. The airline argued that the chatbot was a separate entity responsible for its own answers. The tribunal rejected that outright, holding that a chatbot is still part of Air Canada's website, that it makes no difference whether information comes from a static page or a chatbot, and that Air Canada is responsible for all the information on its website. The airline paid the difference.

Read that as a pricing problem rather than a legal curiosity. A composed number becomes a commitment the moment a customer relies on it, and every quote, capacity claim, lead time and minimum that a model produces from memory rather than from a record is that same exposure, repeated at machine speed.

Now scale that. Every quote, every capacity claim, every lead time, every minimum and every price that a model produces from memory rather than from a record is that same coin flip, and the ones that land wrong land in front of a client.

The fix is architectural, not a better prompt. The model has to be able to read the system of record at the moment it answers. That is what the Model Context Protocol exists for (an open standard for wiring a model into live tools and data). In plain terms it is a socket. The model plugs into your systems and reads the real answer: the catalogue, the stock position, the price, the minimum, the lead time, the analytics. With that pipe, a number is retrieved. Without it, a number is composed.

And this is exactly the intervention the matching function says works, which is the part almost nobody notices.

Go back to the physics. Generating more output raises your numerator, and because every competitor bought the same tool in the same eighteen months, it raises the denominator by an identical factor. Tightness does not move. Your fill rate does not move. Nothing happens.

Grounding a model in real inventory and real pricing does something structurally different. It lowers the cost to the buyer of establishing whether you can actually do the thing. That is A, the match efficiency term, and raising A moves the entire Beveridge curve inward. A is the only variable in the whole model that describes a market getting better at introducing people to each other, as opposed to one participant shouting louder inside it.

So the distinction is not a preference. Generation moves the denominator. Retrieval moves A. Only one of those two terms appears anywhere in the equation that decides whether a buyer finds you.

In plain terms: If you ask an AI for a figure it has not been given, it will produce one that looks right, and your company owns the consequences. In February 2024 a tribunal in British Columbia decided exactly that case. Air Canada's website chatbot told a customer he could claim a bereavement discount after booking. That was not the airline's policy. Air Canada argued the chatbot was responsible for its own answers, and the tribunal rejected it, ruling that a chatbot is still part of the company's website and the company is responsible for everything on it. The airline had to pay. The fix is to connect the system to your real records so it looks a number up instead of composing one.

The measurement void, which is also a search cost.

One more number, because it closes the loop.

Huggg's 2026 UK employee gifting benchmarks, a survey of 85 HR professionals, found that 1.6% formally track the return on their gifting programme, while 65.9% believe it aids retention and 2.1% have actually measured a retention improvement. Belief runs roughly thirty times ahead of evidence.

That is a search friction wearing different clothes. A buyer cannot compare suppliers on outcome, because almost nobody in the category produces an outcome number. So they compare on the only visible axis, price, and the price arrives as a range. Diamond then does the rest.

Meanwhile the money is impatient. Marketing budgets are down to 9.0% of company revenues and 9.6% of overall budgets, with spending growth at 1.7%, the weakest in several years, and more than 70% of leaders prioritising immediate results over long-term gains. Acquisition budgets remain 26% larger than retention budgets even though retention is now the stronger performance driver, and the median duration of marketing impact has lengthened to six months, which is longer than the window most of these programmes are judged in.

Buyers who cannot compare. Suppliers who cannot be compared. Shortening patience on the money side. A category-wide inability to produce evidence. That is not a marketing problem. That is a textbook search market behaving exactly as specified.

In plain terms: In one survey only 1.6% of buyers could actually prove their programme worked, while 65.9% believed it did. When nobody can show a result, buyers fall back on comparing price, and meanwhile the budgets are being cut by people who want proof.

What this means for you, plainly.

If you are looking for a job. Harvard Business School and Accenture asked employers directly, and 88% agreed that qualified high-skilled candidates get vetted out by their own hiring software. For middle-skilled roles it was 94%. The same authors put the number of hidden workers in the United States at roughly 27 million. Meanwhile LinkedIn is taking 11,000 applications a minute, up 45% in a year, and the average UK graduate opening draws 140 replies. Do this instead: read that 88% again. The employers running the system already know it rejects people who can do the job. Going around it is not a trick, it is the documented workaround. One person inside the company who hands your name to the hiring manager is worth more than another hundred forms, because it is the only route that does not pass through the filter everybody admits is broken.

If you are over fifty and looking. Same experiment, same application, only the age changed. Men aged 64 to 66 were called back 16.97% of the time against 23.08% for men aged 29 to 31. For women it was 17.45% against 25.51%. A quarter to a third fewer callbacks for identical applications. Do this instead: the online form is exactly where that filter runs, so it is the worst channel you have and should be the smallest part of your week. Everything goes into people who can walk your name in, because a conversation does not have a field for your birth year.

If you are trying to find a supplier. In one benchmark, 1.6% of buyers could actually prove what their programme returned. When nobody publishes an outcome you end up comparing on price, and the price comes back as a range. Do this instead: before you accept a proposal, ask two questions. What does this cost, and show me what it did for someone my size. Whoever cannot answer both has just shortened your list for free.

If you run marketing. Companies change providers roughly every five years, which is 20 quarters. An annual event plus a periodic email puts you in front of a buyer for about one of those twenty. More than half of companies added channels last year without adding budget, which thins the same money across every one of them. And there is a newer problem on top: buyers increasingly build the shortlist with an AI assistant, and an assistant cannot read a brochure. If your prices, minimums and lead times exist only as sentences on a page or inside a PDF, you are not ranked low. You are absent. Do this instead: three things, in this order. First, put your actual numbers on the page as numbers: price or price band, minimum order, lead time, capacity, turnaround, what you will and will not do. That is the coordinate transform, and it is the difference between being on the page and being off it. Nothing else on this list works until it is done. Second, pick the two places your buyers genuinely ask the question and turn up every week, instead of everywhere once a year. Third, stop reporting reach. Nobody ever bought anything because they saw you.

If you are a creator or freelancer. Announcing you are available costs nothing, which is why everybody does it, which is why the feed is not where work gets handed out. Do this instead: put your rate, your turnaround and what you have actually delivered somewhere a client can reach without having to message you first. That is the whole transform in one move: your terms as values, in the open, so a search or an assistant can place you without a conversation happening first. Most of your competitors will not do it, and that gap is the entire opening.

If you are buying AI tools. MIT looked at more than 300 enterprise AI programmes. Against $30 to $40 billion spent, 95% produced no measurable return, and the reason was not weak models. It was that nobody built them into how the work actually runs. Do this instead: ask the vendor one question. Does this reduce work for us, or reduce work for our customer? A tool that only does the first makes you faster at something that was never the problem. Then ask a second one: when it gives you a number, is that number read from your records or written by the model? If nobody can answer in a sentence, it is being written. Then ask who integrates it, out of whose week. If that person is not named, you are buying the 95%.

Two sentences for everybody. Doing more of what you are already doing is the only response the numbers say does not work. And before anything else on this list can help you, your side of the deal has to exist as values somebody else's system can read, because you cannot be matched to from a frame nobody can compute in.

The patterns that show where the demand actually is.

None of this says the demand cannot be found. It says it cannot be found the way it is usually looked for. There are patterns in the data that get you close, across people, services, products and networks, and they look almost nothing like what is marketed to the general public as marketing.

The difference is simple to state. Consumer-grade tools offer you an audience: reach, impressions, followers, lookalike segments, a bigger list. Every one of those is a move on your own numerator, and the numerator is the term the matching function ignores. What actually locates demand is a signal, and signals have four recognisable shapes.

Timing beats identity. The 95:5 arithmetic says the binding question is not who but when, which makes demographic targeting close to worthless on its own, because 95% of a perfectly-defined segment is out of market anyway. What carries information is anything marking entry into a buying window: a funding event, a hire into a role that owns the budget, a venue booked, a contract approaching renewal, a job posting describing the problem you handle. One dated trigger outperforms a large well-profiled list, and the difference is not a matter of degree.

Questions beat keywords. Matching is collision, so what matters is the specific room where the requirement gets stated out loud, in the week it gets stated. That is a place, not an audience. Find the handful of rooms where your category's question is actually asked and the location problem is largely solved.

Read what the filter reads. Receivers no longer evaluate, they filter, and filters key on whatever is machine-readable. So the practical move is not more persuasion, it is being correct and complete in the exact fields the filter parses: the price, the minimum, the lead time, the certification, the availability. Persuasion is aimed at a person who has already stopped reading.

Networks are not a soft channel, they are a different distribution. This is the one worth understanding properly. The 36.79% floor exists because buyers choose independently and blind to one another. That is the whole cause. Which means any mechanism that coordinates the choice, an introduction, a referral, a curated shortlist, a trusted intermediary, is not a nicer version of the same process. It is a different draw, and the floor does not apply to it. That is the actual mathematical reason a warm introduction converts unlike anything else, and why curation is worth more than range.

Then the fifth pattern, which is the one almost nobody runs: your own results are a signal no competitor holds. When 1.6% of a category can prove an outcome, your own record of what worked, for whom, at what size, is both the best predictor of the next match and the only evidence available to a buyer who cannot otherwise compare anyone.

None of these are reach. All of them are legible demand. The tools marketed to everybody are built to help you talk more, and talking more is the single response the equation is indifferent to.

In plain terms: The signals worth chasing are: when somebody enters a buying window, where they ask the question out loud, what your information looks like to an automated filter, who can introduce you directly, and what your own past results prove. None of those are reach, and reach is what most tools are aimed at.

What actually moves the number.

If friction is the mechanism, louder signalling is not the lever. Signalling harder feeds the congestion externality and makes the commons worse for everyone including you. Five things work, and each one attacks a specific term in the model.

Publish a real price. This attacks Diamond 1971 head on. Opacity is worth monopoly margin to whoever holds it, which means the first participant to make comparison cheap collects the buyers who were stuck. A published number is not a discount. It is a structural attack on the friction your competitors depend on. This is the reason the estimate engine gives a costed answer on the page instead of asking for an email.

Concentrate, do not spread. The collision model rewards density in the one place the buyer looks, not presence in seven. Adding channels on a flat budget lowers your match probability in every one of them. Pick the room where the question actually gets asked and be overwhelming in it.

Be present continuously, not annually. A five-year cycle is 20 quarters and you cannot know which one is live. Presence during 1 of 20 is a 5% duty cycle. The 95:5 rule is answered by always-on, not by a bigger event.

Make the outcome measurable. When 1.6% can prove a return, proof is not a hygiene factor, it is a position. Instrumentation moves you out of the pile that cannot be compared and into the very short list that can.

If you are the one being filtered, change the room, not the volume. This is the individual version of the same instruction, and it is the opposite of the standard advice. More applications, more pitches and more posts raise the denominator you are already losing to. What moves your odds is being present where the decision is actually made, before it is made, in a place where the filter is a person rather than a keyword. One conversation inside the room beats two hundred documents outside it, and that is arithmetic, not networking folklore.

Ground the machine in the record. Any model that answers a buyer on your behalf must be reading live inventory, live pricing and live minimums, not recalling them. A composed number is a coin flip in front of a client, and a retrieved number is the thing that raises A. This is the only one of the five that gets better rather than worse as the rest of the market adopts the same technology, because everyone else is pointing it at output.

In plain terms: Six things work: publish a real price, be in fewer places more often, stay present all year rather than once, prove your results, put your details where a machine can read them, and if you are the one being filtered out, change the room rather than the volume.

One term was deliberately held constant throughout all of this: how much of a buyer's attention you already hold before any searching starts. Let that vary and the arithmetic moves hard in both directions at once. The Biggest Brands in Media Do Not Spend Less. They Spend Earlier. runs it, and finds the distance between invisible and a coin flip is a factor of 3.5, not a factor of four hundred.

The summary.

Everybody caught in a search market believes the problem is reach. The mathematics disagrees, and it disagrees the same way whether you are a person or a company.

36.79% of matches fail at zero cost, perfect fit and equal numbers. Crowd one side five to one and 81.87% of that side ends the cycle empty. An opening drawing 250 applications gives any one of them about 0.4%, and an identical application from a 64 year old is called back 16.97% of the time against 23.08% for a 29 year old, worse again for women. 95% of the buyers you can reach are not in the market this quarter, and an annual rhythm leaves you absent for 95% of the cycle in which they decide. 95% of enterprise generative AI produced no measurable return, because it raised output in a market where output was never the constraint.

And there is no single cause to point at, which is the part that makes this hard to argue with. The coordination floor, the crowding curve, the congestion externality, price opacity, the timing window, choice overload, systematic filtering and the gap between output and match efficiency are separate mechanisms with separate proofs. Any one of them is enough to explain a market where two willing parties never meet. They are all running at once, and they multiply.

And the tool that was supposed to fix it will keep returning plausible wrong numbers until it is wired to a system of record, because a model without a pipe does not look up a figure, it composes one.

Diamond, Mortensen and Pissarides won a Nobel Prize for describing this in 2010. Most of the industry is still operating as though the classical market exists, where everyone finds everyone instantly and price is settled by supply meeting demand.

It does not exist. It never did. The friction is the market.

The house that was built for this.

Every lever above is a cost removed from somebody else. That is the whole design brief here, and it is why the infrastructure looks the way it does.

70,000+ products across 200+ brands in an MCP-ready catalogue, queryable by an agent in real time, so the answer is retrieved from live records rather than composed from memory. A costed answer on the page, with the working shown, instead of a range and a phone call. Blind-ship fulfilment with NDA-standard discretion. Single-source accountability, strategy through manufacturing through delivery. One operator, one line, one number to call when it matters.

We do not add noise to a saturated market. We answer the question at the moment it is asked, from the record.

If you are tired of shouting into a market that stopped listening, the mathematics says the problem is structural rather than personal. The same mathematics says it is fixable, and names exactly which term to attack.

If you are building from what is left, you are not finished.

Jenny Huang Goodman MPA, MSc, MHSA Principal Huang Goodman | Hako Shikin | POPS4 Virginia Beach, since 1997 ASI #217876 | DUNS 18-204-6339 (757) 390-0424 | jenny@huanggoodman.com