Solving the Cold Start Problem in 2025: Advanced Strategies for Igniting Multi‑Sided Marketplaces
The playbook for marketplaces has changed. Yes, you still need to seed both sides, find an “atomic network,” and march city‑by‑city or niche‑by‑niche. But in 2025, new capabilities-and new constraints-are reshaping how the cold start problem gets solved. AI agents can now stand in for real users (or help you test incentive designs before launch). Regulators are sharpening rules about authenticity and transparency. And distribution increasingly hinges on landing a few “whales” who can tilt supply–demand dynamics overnight.
Below is a practitioner’s guide to advanced-not basic-strategies that work now.
Start with liquidity, not “users”
As Lenny Rachitsky puts it, “The prime directive of a marketplace is to generate liquidity where none existed before.”Liquidity-how efficiently you match buyers and sellers-is the product. Optimize for fill rate, time‑to‑match, and time‑to‑first‑transaction long before you worry about top‑line users. (Lenny’s Newsletter)
Andrew Chen popularized the idea of atomic networks: the smallest self‑sustaining cluster with enough density to stand on its own. Win one atomic unit (a campus, a neighborhood, a craft category), then replicate adjacently. You can’t skip to “global” until your first atoms hum. (Lenny’s Newsletter)
1) Whale acquisition: land the “anchor tenants” of your market
In retail real estate, an anchor tenant pulls foot traffic and de‑risks everything around it. The same thing happens in digital marketplaces: a small number of high‑leverage participants (enterprise buyers, power sellers, fleet operators, or institutional suppliers) can instantaneously change the market geometry. (Point Acquisitions)
Case in point: Uber’s partnership to integrate ~14,000 New York City yellow taxis put a massive block of supply into the app in one shot-classic “anchor supply” acquisition that compresses time to liquidity. (The Verge)
Why this works: many marketplaces are “Pareto‑shaped.” Borrowing from gaming, a tiny fraction of “whales” often drive a huge share of outcomes. One widely cited analysis found that ~0.19% of mobile gamers generated ~48% of revenue-a reminder of how outsize certain actors can be. (You’re not a game, but the math of heavy‑tail distributions still applies.) (WIRED)
How to run a whale program (safely and fast):
Map the value graph. Which single contract (or creator, or shop network, or enterprise buyer) instantly raises the baseline match rate?
Craft a bespoke value prop: SLAs, reporting, co‑marketing, integrations, or even custom tooling that embeds you in their workflow.
Offer tailored economics: Early‑access pricing, temporary negative rake on their first $X of GMV, or volume‑based rebates.
Design PR for proof: A public integration (e.g., “taxis now on our app”) doubles as demand marketing.
2) Use AI to simulate or stand in for one side-with disclosure and guardrails
In 2023, researchers demonstrated generative agents that “simulate believable human behavior.” Their agents planned, remembered, coordinated, and even organized a Valentine’s Day party-without human micromanagement. That’s no longer science fiction for prototyping marketplace dynamics. (arXiv)
What this unlocks:
Synthetic supply or demand for QA: Before you have live users, populate “shadow demand” (queries, carts, bids) to stress‑test search, pricing, and ranking; similarly use agent “sellers” to test listing flows and fulfillment logic. See how your algorithms behave under load and edge cases.
24/7 baseline liquidity: For categories where a lightweight version of the service can be automated (e.g., first‑response Q&A before a human pro engages), AI can reduce time‑to‑first‑contact and lift conversion.
Agent‑to‑agent commerce: The Financial Times recently noted the rise of AI shopping agents from OpenAI, Google, Microsoft and others that can browse, compare, and even complete purchases-meaning some “demand” will soon originate from bots acting on behalf of humans. Design your marketplace to be legible to them. (Financial Times)
Model in simulation before you subsidize: Recent work on agent‑based modeling of two‑sided markets shows how platform dynamics can be explored before you deploy cash. Use agent sims to preview how a change in fees or incentives impacts liquidity and quality. (IFAAMAS)
Ethics & compliance matter. The FTC’s 2024 rule prohibits selling or purchasing fake reviews or testimonials-and covers AI‑generated content. If agents participate, disclose it; never pass bots off as independent human reviewers or counterparties. In the EU, the DSA raises transparency and risk‑mitigation expectations for marketplaces. Build in labeling, audit trails, and appeal mechanisms from day one. (Federal Trade Commission)
3) Design novel incentive structures that create liquidity without blowback
Subsidies still work-but they must be precise, adaptive, and trustworthy. Three patterns:
A) Threshold‑based and dynamic incentives (quests, streaks, zones)
Batched incentives can nudge supply to the right place at the right time and reduce wait times. Academic work documents these threshold‑based incentives (e.g., Uber “Quest” bonuses), and companies publish the mechanics: Uber Quest pays extra for hitting targets; Lyft Bonus Zones nudge drivers to busy areas with upfront bonuses. (ScienceDirect)
At the algorithmic frontier, Lyft’s Personal Power Zones system uses an escrow‑like budgeting mechanism to steer drivers, and the rollout delivered a ~0.5% increase in incremental bookings-a small percentage on huge volume equals real money. (arXiv)
Caution: Incentive design can backfire if it’s opaque or cross‑subsidized in misleading ways. DoorDash learned this the hard way, agreeing in 2025 to pay $16.75M to New York delivery workers over how tips interacted with guaranteed pay in 2017–2019 (the company replaced that model years ago). Design for clarity and auditability. (The Guardian)
B) Reverse/low rake to win the market, then expand the pie
Bill Gurley’s classic line still holds: “High rakes are a form of friction.” Early on, keep rake low or negative where it matters (e.g., on first transactions, or for your anchor suppliers), then monetize via premium tools, financing, or adjacent services rather than taxing the core match prematurely. (Above the Crowd)
C) Balance‑sheet innovations for B2B liquidity
B2B marketplaces win when they reduce working‑capital friction. Faire famously popularized Net 60 terms, letting retailers buy inventory now and pay later-greasing initial matches. Pair that with sell‑through protection and you de‑risk experimentation for the buyer side. (Faire)
D) Tokenized “proof‑of‑work” incentives-used judiciously
In physical‑world networks (coverage, maps, sensors), “Proof‑of‑Physical‑Work/DePIN” token designs have been used to bootstrap supply. Helium rewarded hotspot operators via Proof‑of‑Coverage and onboarded hundreds of thousands of IoT hotspots globally-an existence proof that designed incentives can coordinate resource deployment at scale. Use lessons (low‑friction onboarding, anti‑gaming) even if you don’t use tokens. (whitepaper.helium.com)
4) Don’t just “list”-manage: increase quality with a managed marketplace layer
“Managed marketplaces” take on parts of service delivery-qualification, standardization, trust/safety-so that early users experience reliably good matches, not just any matches. Intervening on quality speeds the march to liquidity and retention. (Andreessen Horowitz)
Airbnb’s AirCover (e.g., $3M host damage protection, ID verification) lowered perceived risk and made the early network more usable for both sides-proof that platform guarantees are growth features. (Airbnb)
5) Build single‑player value for the hard side
If you’re supply‑constrained, give suppliers software that’s valuable even with zero demand-inventory tools, scheduling, or analytics-then flip on marketplace demand. OpenTable solved this brilliantly with its Electronic Reservation Book (ERB)-hardware + software that digitized the host stand; once enough restaurants ran ERB, diner liquidity followed. (SEC)
6) Instrument what matters: operational definitions of liquidity
Agree on crisp, leading indicators:
Fill rate (demand side): % of requests fulfilled within your SLA.
Sell‑through (supply side): probability that a listing sells in X days.
Time‑to‑first‑transaction (both sides): median time from signup to first match.
Matching latency: time from request/search to confirmed match.
Treat these as the north star until you surpass a clear liquidity threshold in each atomic network. (For a deep dive on definitions and examples, see Lenny’s marketplace metric guide.) (Lenny’s Newsletter)
7) Sequencing: a 90‑day advanced plan
Weeks 1–2: Value‑graph and atom selection
Pick one atomic network where you can credibly reach >70%+ fill rate quickly.
Map the whales (anchor suppliers or buyers) whose participation collapses time‑to‑liquidity. (Lenny’s Newsletter)
Weeks 2–6: Whale acquisition + single‑player
Develop whale‑specific SLAs, reporting, and integrations. Offer temporary negative rake or tiered rebates to land them. (Above the Crowd)
Ship “single‑player” tools for the hard side (e.g., CMS/inventory, scheduling, payouts). OpenTable‑style embedded tooling beats cold emails. (SEC)
Weeks 3–8: AI‑assisted ignition & sandboxing
Use agent simulations to tune supply allocations, fees, and ranking without burning cash. Inject synthetic test traffic to measure matching latency and search relevance under load. (IFAAMAS)
Where appropriate, deploy AI first‑response or triage so demand always gets an instant touch; if agents transact, label them clearly to comply with FTC/DSA norms. (Federal Trade Commission)
Weeks 5–12: Incentive precision
Roll out threshold‑based incentives (quests/streaks/bonus zones) to shape availability during critical windows; publish transparent rules and in‑app ledgers so participants trust the math. (Uber)
For B2B categories, pilot Net 30/60 with credit underwriting to reduce working‑capital friction. (Faire)
Always on: Quality & trust
Add managed‑layer interventions (screening, standardized pricing, escrows/guarantees) that turn early matches into repeatable matches. Consider escrow‑ or SLA‑backed guarantees to reduce perceived risk. (Andreessen Horowitz)
8) Pricing: be patient on take rate
Take rate is a lagging lever. Early on, behave like a payments rail + quality guarantee, not a tollbooth. Gurley again: “High rakes are a form of friction.” When you win liquidity and trust, layer on SaaS‑like tools, financing, logistics, or insurance where you create real incremental value-and price those, not the raw match, whenever possible. (Above the Crowd)
9) Reality check: matching incentives to behavior
Not all incentives are created equal. Research shows that threshold‑based incentives (e.g., do X jobs to unlock $Y) change where and when sellers work; but they can also create cliffs and churn if goals feel unattainable. Monitor marginaleffects (supply hours shifted per $1 of incentive) and spillovers (post‑incentive engagement) to ensure you’re buying durable behavior, not just screenshots. (ScienceDirect)
Bringing it all together
Ignition in 2025 looks like this:
Win whales (anchor tenants) to compress time to liquidity,
Use AI-ethically-to simulate, prototype, and even stand in for one side of the market,
Engineer incentives that are surgical and transparent (and test them in simulation first),
Manage quality, not just listings,
Instrument liquidity metrics and scale atom‑by‑atom.
Do those, and your marketplace stops feeling like a cold start and starts behaving like a flywheel.