Network Effects: When Each User Makes the Product Better

Almost everything written about this subject is written with adjectives. A service gets stronger, a position gets deeper, an effect kicks in. Two counts and one division do the same job, and the reason to prefer them is not tidiness. The adjectives all point the same way and the arithmetic does not.

The argument rests on two published counts and on nothing else. Setu Bazaar, an invented online marketplaceA business that brings buyers and sellers together and never takes title to the goods that pass between them. The marketplace carries the listing and the order, and somebody else holds the stock., has 50,000 buyers and 2,000 merchants. Every figure that follows is one of those two counts, a doubling of one of them, or the number produced by dividing one by the other. A network effect is the idea in company analysis most often argued with adjectives instead of arithmetic, and two published counts are enough to settle the argument.

The second thing the argument rests on is a published result for the year, and that result is what keeps the arithmetic honest. Setu Bazaar closed its published year at minus Rs 2,50,00,000/-. A network effect describes what one participantAnybody on either side of a marketplace, whether they are buying or selling. A buyer is a participant and so is a seller. gets. A result describes whether the year worked. Setu Bazaar carries the effect and carries the loss, in the same published year, and that pairing is the cheapest proof available anywhere that the two are different objects and not two readings of one.

The third thing the argument rests on is arithmetic so ordinary that stating it feels unnecessary. A ratio rises when the top of it rises and falls when the bottom of it rises. Every reader knows that. Almost nobody attaches that rule to the word better, and the moment it is attached, the sentence a marketplace got better splits into two sentences that point in opposite directions.

What is a network effect, and what exactly is getting better?

The whole argument sits inside the definition, so the definition repays reading slowly. A network effect means that the value of a service to one participant depends on how many other participants are there. Not on how good the service is. Not on what it charges. On how many other people showed up.

A network effect with no participant named is not yet a claim anybody can check. Better is a two place word: something is better than something else, for somebody. A sentence that leaves out for whom has dropped the half that could be tested, and what remains sounds like a finding while behaving like a mood.

Take a picture off the street first. A shopping arcade has ten shops in it and an eleventh opens. For a shopper walking in off the road, that eleventh shop is one more reason to come, one more window to look in, one more chance of finding the thing they came for. For the ten shopkeepers already paying rent, the eleventh shop is one more competitor for the same footfall walking through the same doors. One event. Two readings. The two readings are about two different people, and nobody said which one they meant. No contradiction anywhere.

The arcade is the entire structure of a network effect on a marketplace, and it is why every claim about one has to name a side. A marketplace has at least two kinds of people standing in it, and an event that adds people to one kind is not the same event when read from the other kind. Once the split has been seen in an arcade it can be seen everywhere. Performed on two published counts, the same observation produces a number rather than a feeling.

What is the only number on Setu Bazaar that a claim like this can be checked against?

Setu Bazaar has 50,000 buyers and 2,000 merchantsA seller listed on a marketplace, who holds the goods and takes the order, rather than the marketplace itself. On Setu Bazaar there are 2,000 of them.. The first divided by the second gives the number a claim about Setu Bazaar can be checked against. Fifty thousand divided by two thousand is 25.00, and the unit matters as much as the figure: 25.00 buyers a merchant.

The 25.00 measures one merchant's reach from standing in the marketplace, not the marketplace itself. Read it the wrong way round and everything that follows goes wrong, so it is worth saying what the ratio does and does not carry. The ratio counts heads on both sides and it counts nothing else. It says nothing about how much anybody bought or how often. The ratio is not a rate, not a share and not a growth figure. The ratio is a head count divided by a head count, so anybody holding the two published counts can recompute it.

One line of housekeeping saves confusion later. Wherever the figure 25.00 appears below it means buyers a merchant, and it never means anything else. The same two digits can describe a completely different quantity about the same business, and a reader who carries the figure across without carrying its unit has carried nothing at all.

There is one more discipline to apply before moving on, and it is the one every competitive advantage is measured by. A gap has to be measured against something named. So say what the 25.00 is measured against. The 25.00 is measured against Setu Bazaar's own two counts as they stand today. No rival's buyer count or merchant count is published, so it is measured against no rival marketplace anywhere. A figure measured against its own counts is checkable, and that is worth a great deal. The same figure is modest, and that is worth stating out loud rather than quietly hoping nobody asks.

What happens to a merchant when the marketplace doubles its merchants?

Hold the 50,000 buyers exactly where they are. Do not move them, do not add one, do not lose one. Now take the merchants from 2,000 to 4,000, twice the published count. The same division runs again: fifty thousand divided by four thousand is 12.50 buyers a merchant.

Nothing was taken away from any merchant and every merchant is worse off. That sentence is uncomfortable on first reading and it is exactly right. No buyer left the marketplace. No merchant lost an account. The same 50,000 people are simply spread across twice as many sellers, so the reading falls to half of what it was. Look at the denominatorThe bottom number in a division. Here it is the merchant count, and a ratio falls when its bottom number rises even though the top number has not moved. rather than at anybody's behaviour and the whole thing is obvious.

Say it in a picture. One ferry crosses a river and carries everybody who wants to cross. A second ferry starts the same route on the same river with the same passengers walking down to the same jetty. Nobody took a passenger off the first boat. Nobody boycotted it. The first boat now carries half as many people. The arithmetic did all of the work on its own, and no explanation involving service quality, pricing or effort is required.

The reading is not new. The account of platform businesses with two dependent sides works this exact event through from both sides: twice the choice for a buyer, half the reach for a merchant. The question nobody attaches to it is the one that matters for durability. If an effect can run backwards for a side that is already present, is it the kind of gap that survives, or is it just a description of why people turned up?

What happens to a merchant when the marketplace doubles its buyers?

Same business, same starting pair, opposite end of the division. Hold the 2,000 merchants exactly where they are and take the buyers from 50,000 to 1,00,000. One lakh divided by two thousand is 50.00 buyers a merchant, and that figure came from doubling the buyers.

The same ratio, moved from the other end, does the opposite thing. Not a smaller version of the same thing. Not a weaker version. The opposite direction, from one control that looks, in a headline, exactly as much like growth as the other one did. The pair of results yields one sentence. More buyers make the marketplace better for a merchant, but more merchants help a buyer only if they bring rangeHow much genuinely different choice a buyer gets, as opposed to how many sellers are counted. Two sellers of the same thing are two sellers and one kind of choice..

The second half of that sentence needs unpacking. No figure for range is published, and inventing one would be worse than saying nothing, so the unpacking has to be done without a figure. Think about what a buyer actually wanted. Somebody who came looking for one particular kind of thing, and who could already pick between two thousand sellers of it, gains close to nothing from the two thousand and first seller of the same thing. The same buyer gains a great deal from the first seller of something they could not get on the marketplace at all. A count of sellers is a proxyA stand-in figure used because the thing actually wanted is not counted. A proxy moves roughly with the real thing and is not the real thing. for range and it is not range.

Walk down a street with twenty tailors on it. The twenty first tailor changes almost nothing for a customer walking down that street. The choice they had was already wide in the only direction it went. The first shoe shop on that street changes everything, and it changes it for a customer who was never going to buy a shirt. One new shop, counted the same way in both cases, and two completely different effects on the person walking past.

Why is a network effect not symmetric?

Put the two directions in one place and read them together. Separately each of them looks unremarkable and together they are the finding. From the published pair of 50,000 buyers and 2,000 merchants: doubling the merchants takes the merchant reading from 25.00 to 12.50, and doubling the buyers takes the same reading from 25.00 to 50.00. Two events. One ratio. Opposite outcomes.

A network effect is not symmetric, so growth on one side and growth on the other are not the same event and must never be summarised together as growth. Both events add participants. Both would be reported as the marketplace getting bigger. Only one of them improves the reading for the side already standing there.

Now the difficult part, said plainly. Almost every short description of a network effect is written as though every extra participant helps everybody, and that description is not a rough simplification of the arithmetic. On one of the two sides it is the reverse of the arithmetic. A merchant reading that a marketplace has strong network effects, and understanding it to mean that every new arrival helps them, has been handed a sentence that runs the wrong way for one of the marketplace's two sides.

The fix costs one clause. Before any effect is called an advantage, the question is which side it was measured on. Naming the side is the whole discipline, and it is cheap enough to apply to every claim a reader meets.

What happens when a marketplace recruits sellers faster than buyers?

Recruiting one side faster than the other is where the asymmetry stops being a curiosity and turns into a question about whether an advantage lasts. Signing sellers is visible work. Somebody can be sent to do it, the count goes up, and the count can be put in a report. Signing buyers is the same work performed on the side that is harder to reach and harder to count. So of the two counts that make up the ratio, the easier one to move is the one sitting underneath.

A marketplace that recruits sellers faster than buyers has made itself worse for the side it already had. Follow that through for the gap rather than for the quarter. A merchant stayed on Setu Bazaar for the buyers it could reach from there. A marketplace that spreads those buyers thinner has spent the reach that was holding that merchant, and it spent it in order to report a bigger number.

Say the other half plainly. The first half on its own is unbalanced. The asymmetry is not an argument against signing sellers. A marketplace with too few sellers is a marketplace where buyers find nothing, and nobody is suggesting that a seller count should sit still. The argument is narrower and harder: the two counts have to move together, or the advantage narrows while the headline widens, and the two things happen in the same quarter and get reported in the same sentence.

There is a weekly vegetable market version of this that anybody who has stood in one has watched happen. The market doubles its stalls without doubling its shoppers. From the road it looks twice the size and twice as busy. Every stallholder goes home with more unsold stock than the week before, and none of them can point at the person who did it to them.

Does a network effect make a business profitable?

No. An account that skips this answer has taught a reader something true and left them holding a conclusion that does not follow from it, and a true fact with a false conclusion attached is worse than teaching nothing.

Setu Bazaar's published year is a result for the yearWhat is left after every cost is taken off. The result can be a positive figure or a negative one, and Setu Bazaar's is negative. of minus Rs 2,50,00,000/-, which is two croreAn Indian unit of counting equal to one hundred lakh, being ten million. Rs 2,50,00,000/- is two crore fifty lakh rupees. fifty lakh rupees of loss. Its network effect, meanwhile, is entirely real: a merchant standing on that marketplace reaches 25.00 buyers it would otherwise have to find one at a time.

Setu Bazaar carries the effect and carries the loss, in the same published year, and neither fact argues with the other. The rule holds in both directions: a network effect is about what a participant gets, and it says nothing at all about whether a year worked. The effect and the result are answers to two different questions, and only one of them is a question about the accounts.

Why is this confusion so common and so expensive? Because Setu Bazaar's effect is genuine, and a reader who has just been shown a real advantage is in the most receptive state they will be in all day. Almost any conclusion offered immediately afterwards gets accepted, and the conclusion most often offered is the one about the result. The joins are where it happens. The 25.00 and the minus Rs 2,50,00,000/- linked by so or therefore. A sentence beginning with as the marketplace grows. A caption pairing a ratio and a result inside one clause. None of those joins is earned by the two counts or by the result.

When is a network effect an advantage that survives, and when is it only growth?

An advantage is a gap that survives, and a gap survives only where a rival cannot put the other side together. Why assembling a second side is genuinely hard is set out separately under Platform Businesses: Why Two-Sided Markets Behave Differently.

The test that falls out of the asymmetry has two steps. The first asks which side the effect protects. The second asks what that side would lose by leaving. For a merchant on Setu Bazaar, leaving means giving up the buyers it reaches, and that loss is exactly the 25.00 buyers a merchant computed from the two published counts. An effect that only describes why participants arrived is not yet a reason they cannot leave.

The second question is what disposes of the adjective. Strong network effects is not an answer to what would this side lose by leaving. A ratio is. The second question is the practical reason to carry the number and drop the adjective, and it is also why the number keeps its unit attached: a reader can take 25.00 buyers a merchant, hold it against the counts, and disagree in arithmetic rather than in tone.

Two questions sit just outside the arithmetic. Why a participant who could leave stays anyway is covered under Switching Costs: Why Customers Stay Even When They Could Leave. Why a business is able to raise the fraction it keeps of what crosses it is covered under The Sources of Competitive Advantage, and Whether Any of Them Lasts. On that second point, the sources that let a business hold its charge without losing volume are four: brands, switching costs, scarcity and rivalry. A network effect is not one of the four. The arithmetic supports something narrower: a marketplace can hold what it charges a side only while that side is still getting more there than it would elsewhere, and the ratio measures the more.