TL;DR: Google research on self-replicating programs found shorter ones succeed. Longer ones don't. Marketplaces work similarly. Craigslist: post, contact, buy — three steps. Works fine for a $50 bike. But a $30,000 car needs financing, inspection, warranties. More steps, because buyers need protection at that price. Most marketplaces have several different loops running. Uber has one for riders (quick), another for drivers (slow, needs background checks). Success depends on matching your process length to what you're selling and how much trust people need.
Google recently published research on self-replicating computer programs. The study wasn't about marketplaces, but the findings, in my experience, apply directly to how marketplaces grow.
Short programs replicated successfully. Long ones never did.
This matches what happens in real marketplaces. Craigslist works with three steps: post, contact seller, transact. OpenTable needed restaurants to sign contracts, install hardware, train staff, update inventory, then wait for diners to discover them, create accounts, and book. One grew organically. The other needed $100 million to reach profitability.
Marketplaces Run Multiple Loops
A marketplace isn't one thing. It's several feedback loops operating simultaneously.
Take Uber. There's a driver loop: sign up, get background checked, upload documents, get approved, start driving, earn money. That's already six or seven steps before earning anything.
The rider loop is different: request ride, get picked up, pay automatically. Three steps.
Then there's geographic density. More drivers in an area means faster pickups. Faster pickups bring more riders. More riders attract more drivers.
And surge pricing, which everyone complains about but actually balances supply and demand during peak times.
Each loop has different speeds and friction points. The rider loop naturally spreads because it's short. The driver loop needs subsidies and guarantees because it's long.
Rightmove operates similarly:
- Estate agents compete on listing quality to get buyer leads
- Buyers browse and save properties
- Performance metrics show agents which listings work
- Agents adjust their approach based on data
Different loops. All reinforcing each other. But each with different dynamics.
Transaction Value Determines Complexity
A $50 Facebook Marketplace purchase doesn't need the same process as buying a car.
But Carvana can't sell $30,000 cars without financing options, warranties, and inspection reports. Those aren't friction. They're requirements.
A simplistic pattern:
- Under $100 transactions: Remove everything possible
- $100-$1,000: Basic ratings and maybe verification
- $1,000-$10,000: Identity checks, payment protection
- Above $10,000: Full diligence process
SaaS marketplaces selling to enterprises learn this fast. Average deal size of $50K needs vendor verification, security audits, legal review. You can't skip these for enterprise sales.
Geographic Constraints Aren't Binary
Marketplaces aren't simply "local" or "global." Different parts need different geographic strategies.
LinkedIn handles this well. Job postings are local - nobody commutes from New York to London. But professional reputation is global. So they built both: local job matching by city and commute distance, plus a global network for professional connections.
DoorDash seems hyperlocal. The delivery is. But driver utilization works across neighboring zones. Restaurant partnerships span cities. Menu data gets standardized regionally.
Etsy sells globally (handmade jewelry from Thailand ships worldwide) but also built local seller communities for materials and support. Both dynamics work together.
The question isn't whether you're local or global. It's which parts of your value creation need geographic density versus which benefit from broader reach.
Platform Investment Amplifies User Value
YouTube without recommendation algorithms is just video hosting. The algorithm makes user content valuable. Same with TikTok's For You page or Instagram's feed.
The pattern isn't users versus platform. It's platform investment making user contributions easy:
- YouTube spends billions on CDN infrastructure
- Amazon built fulfillment centers so sellers can focus on products
- Uber's routing algorithm makes driver time efficient
The successful model: Platform invests in infrastructure → Users create value more easily → Value compounds → Platform takes percentage.
Not: Platform creates everything → Users consume → Hope for subscription revenue.
Interaction Density Needs Context
Raw interaction count misleads. A B2B marketplace might celebrate "10,000 messages sent" but if 9,000 are tire-kickers and serious buyers get lost in noise, those interactions hurt more than help.
Uber discovered this in smaller cities. Having drivers available at 2pm Tuesday doesn't help if demand comes Friday night. You need the right interactions at the right time.
Better approach:
- Identify when value actually gets created
- Focus resources on those moments
- For Uber: Friday nights, airport runs, commute hours
- For B2B: Industry-specific budget seasons
Density at valuable moments beats overall volume.
Behavioral Diversity Predicts Resilience
Count how many different ways users succeed on your platform.
eBay's problems started when they went from supporting collectors, vintage dealers, people selling random stuff, to basically becoming "Amazon with auctions." Revenue went up. Resilience went down.
Etsy maintains diversity:
- Digital downloads for instant gratification
- Vintage collectors who browse patiently
- Wedding supplies for time-sensitive, high-value purchases
- Craft materials as hidden B2B transactions
If wedding trends change, they survive. If digital downloads face regulation, they're fine.
Most marketplaces optimize toward one behavior because it's easier to operate. Then that behavior shifts and the platform struggles.
Flat Growth Periods Are Normal
The research showed programs doing nothing for thousands of cycles, then suddenly replicating everywhere.
Airbnb was flat for two years. Nearly failed. Then growth exploded. Could've been the professional photography program. Could've been the 2008 financial crisis pushing people to rent spare rooms. Hard to know exactly.
This pattern appears repeatedly. Long flat period, then sudden growth. It's not failure followed by success. It's pre-emergence followed by emergence.
The problem: You can't know if you're in pre-emergence or actual failure. If the fundamentals work (short paths where possible, users creating value for others, sustainable economics), flat periods might be normal.
Practical Evaluation Framework
Map all loops separately. Write down each feedback loop in your marketplace. Driver acquisition. Rider retention. Geographic density. Pricing dynamics. Each one distinct.
Score each loop:
- Steps from start to value creation
- Can users improve this for other users?
- Natural frequency of the loop
- Does value compound or just repeat?
Check economics against friction:
- Does your take rate support this complexity?
- If you need seven steps, is LTV high enough?
- Where does monetization fit naturally?
Identify geographic requirements:
- Which loops need local density?
- Which benefit from global scale?
- Where do these conflict?
Track pattern diversity:
- List all successful user behaviors today
- Are new patterns emerging or consolidating?
- What would break each pattern?
The Core Point
You can't force these patterns. You can create conditions where emergence becomes more likely, but you can't guarantee it.
OpenTable spent $100 million pushing adoption. Worked eventually. Craigslist spent almost nothing and patterns emerged naturally. Both succeeded through different paths because their underlying dynamics differed.
Success comes from understanding your specific constraints — transaction value, trust requirements, geographic dynamics — then finding the shortest sustainable path within those constraints.
The question is: given what we're selling, who's buying, and what trust they need, what's the minimum viable process for users to create value for each other?
Answer that, and emergence becomes possible. Not certain. Just possible.
In marketplaces, possible is what you get.
Framework drawn from "Computational Life: How Well-formed, Self-replicating Programs Emerge from Simple Interaction" (Google Research, 2024)