Put a Network on Top of Waste
Find a fragmented industry full of slack, lay a network over it, and build liquidity city by city.
- Difficulty
- Advanced
- Time to result
- ~ongoing to results
- Steps
- 6
- Confidence
- 91%
Gurley traces the whole lineage to a 1996 Harvard Business Review article by Brian Arthur of the Santa Fe Institute, Increasing Returns and the Two Worlds of Business, which he calls the first piece to describe network effects. That gave Benchmark the impetus for OpenTable, and OpenTable in turn gave Gurley the template he applied to Uber: he was trying to think of other industries where putting a network on top would absorb waste and make it more efficient and more usable — which pointed at black cars rather than at the taxi framework. The mechanism has two halves. First, the flywheel bet must be statable in one sentence: if enough restaurants get on this thing then the consumers will come, and if the consumers come then people will have to get on. Second, you have to build liquidity city by city rather than going everywhere at once; OpenTable played a local game starting in San Francisco. The payoff proof was empirical. The sales model required each rep to close four restaurants a month and they were running at 7.7, until a board meeting revealed one rep who had closed 35 in a month. That rep was the only one left in San Francisco, where penetration had reached 90 percent — the last ten percent was signing itself up. That is what a network effect looks like from the inside.
Origin
The concept came to Gurley from Brian Arthur's 1996 HBR piece on increasing returns, encountered through the Santa Fe Institute. Benchmark used it to justify OpenTable despite terrible unit deployment economics, and Gurley then reused the pattern deliberately to search for what became Uber.
Core principles
- 01Network effects produce winner-take-most structures, which is precisely why they produce outlier outcomes.
- 02The reusable pattern is a template, not a one-off: OpenTable's success generated the search that found Uber.
- 03The target is an industry where a network would absorb waste and make the thing more efficient and more usable.
- 04Marketplaces should be built city by city to reach real liquidity, not spread thin across many markets at once.
- 05Belief in the network effect is what lets you accept an ugly early rule set — hardware in small businesses, broadband you have to provision yourself.
- 06Local dominance produces observable proof: sales productivity collapses into order-taking when penetration approaches saturation.
How to run it
- 1
Hunt for industries carrying visible waste
Scan for sectors with idle capacity, poor matching or manual coordination, and ask what a network laid on top would absorb. Gurley did this explicitly after OpenTable and it pointed at black cars.
Pro tip Look past the obvious incumbent framing — Uber came from looking at black cars, not at taxis.
- 2
Test for winner-take-most structure
Confirm the category actually locks in: more success creates more lock-in, switching becomes costly, and the user experience is a function of everyone being on it.
Pro tip Instagram is Gurley's canonical example — you can clone the product but not the graph.
Watch out Many marketplaces have weak network effects and stay permanently fragmented; those never produce outlier returns.
- 3
State the flywheel as an explicit two-clause bet
Write the causal loop in one sentence and commit to it as the thing you must believe. For OpenTable: get enough restaurants on and consumers come; once consumers come, the remaining restaurants have to get on.
Pro tip If you cannot write the loop in one sentence, you do not have a network business.
Watch out Everything downstream depends on believing this, so be honest about whether you do.
- 4
Accept an ugly rule set if the loop holds
Network conviction lets you take on deployment work you would otherwise reject. OpenTable put PCs into small restaurants that had no connectivity, so they had to partner to get broadband installed.
Watch out Your normal rule set says do not put hardware into cash-poor SMBs and do not provision their broadband. Overriding a rule set requires the network thesis to be genuinely load-bearing, not a rationalisation.
- 5
Saturate one city before opening the next
Play a local game. Build liquidity market by market rather than going everywhere at once, so at least one market crosses the density threshold where the effect kicks in.
Pro tip Reject the pressure to show national logos early; density is the only metric that matters pre-liquidity.
Watch out Spreading thin across many cities can leave every market below the liquidity threshold simultaneously.
- 6
Read sales productivity as your liquidity gauge
Track closes per rep per market. A sudden collapse of sales effort into order-taking in one market is direct evidence the network effect has taken hold there.
Pro tip Ask who the outlier rep is and which market they are in before you celebrate the number.
In the wild
OpenTable's scaled model required each salesperson to close four restaurants a month; the team was running at 7.7 and feeling good. A board meeting listed one rep who had closed 35 in a single month. Gurley asked who it was: the only rep left in San Francisco, where OpenTable had 90 percent penetration and the last ten percent were effectively signing themselves up.
→ Gurley took it as his favourite proof point that the network effect was genuinely working, validating the city-by-city liquidity strategy.
Having seen OpenTable work, Gurley deliberately searched for other industries where putting a network on top would absorb waste and make the service more efficient and usable. That search pointed at black cars rather than at working within the taxi framework.
→ Benchmark backed Uber, and the same network logic later explained why the market grew rather than being capped by the existing taxi fleet.
Common mistakes
Going everywhere at once
Expanding across many cities before any one reaches density leaves every market below the liquidity threshold, so the network effect never fires anywhere and the business looks like a plain sales operation.
Letting the deployment rule set veto the thesis
OpenTable violated every sensible rule — hardware into small businesses with no money, broadband you had to provision. Applying the normal rule set would have killed a business whose network thesis was sound.
Assuming any marketplace has network effects
Network effects require structural lock-in where the user experience is a function of everyone being on it. Without that, the category stays fragmented and produces ordinary rather than outlier returns.
Is it for you?
Best for
Founders and investors evaluating two-sided marketplaces in fragmented, inefficient, physically local industries.
Not ideal for
Categories with genuinely weak network effects, where scale confers no advantage and the market stays permanently fragmented.
From the transcript
“I was trying to think of other Industries where if you put a network on top of it it would absorb waste and make it…”
“if we get enough if we get enough restaurants on this thing then the consumers will come and if the consumers come then people will…”
“that sales person was the one sales person left in San Francisco where we had 90 penetration”
From the episode
#651: Legendary Investor Bill Gurley on Investing Rules, Finding Outliers, Insights from Jeff Bezos and Howard Marks, Must-Read Books, Creating True Competitive Advantages, Open-Source Strategies, Adapting Mental Models to New Realities, and More