Who Sets the Price? Designing Pricing Control in a Two-Sided Market

Pricing control is a deliberate design choice, not a default setting: sellers keep the price lever when supply is scarce and trust matters most, platforms take it when demand needs predictable, comparable prices, and algorithms take it when liquidity has to flex in real time. That choice, more than take rate, quietly shapes retention on both sides.

Quick answer: Seller-set pricing protects supply trust; platform-set protects demand trust; algorithmic/dynamic optimizes liquidity but strains fairness perception. Most marketplaces start seller-set and migrate toward managed pricing once supply concentrates and price dispersion starts costing conversion.

Every marketplace PM eventually gets asked some version of "why can't we just let sellers charge whatever they want?" or its mirror, "why can't we just set the price for them?" Both questions treat pricing control as a binary, when it's really a spectrum with real mechanics attached to each position. Get the position wrong and you don't just lose margin — you lose the trust of one side of the market, quietly, over months.

This is a distinct decision from the marketplace PM role question of take rate. Take rate is how much of the transaction the platform keeps. Pricing control is who decides the number the take rate gets applied to. Confusing the two leads teams to "solve" a liquidity problem by adjusting commission when the real lever — who holds the price — was never touched.

The Three Levers of Pricing Control

A marketplace's pricing-control model falls into one of three buckets: seller-set (the supplier picks the number), platform-set (the platform picks the number and sellers opt in or out), and algorithmic/dynamic (a pricing engine adjusts the number continuously based on live supply and demand signals). Each trades liquidity for trust in a different direction, and most marketplaces use more than one model across their catalog.

Three real archetypes make the differences concrete:

1. Airbnb (seller-set, algorithmically assisted). Hosts set the nightly rate; Smart Pricing suggests a range based on comparable listings, but the host can override it entirely. Airbnb keeps host trust by never overriding the number itself.
2. Uber (platform-set, algorithmically dynamic). Riders never negotiate; Uber sets the base fare and applies a surge multiplier when demand outstrips available drivers. Drivers can accept or decline a trip, but not counter-offer.
3. Etsy (seller-set, no dynamic layer). Sellers price handmade and vintage goods entirely on their own judgment. Etsy's role is discovery and payments, not pricing — which is consistent with a market where value is subjective and comparison shopping is limited.

The point isn't that one archetype is "correct." Airbnb, Uber, and Etsy are optimizing for different goods: differentiated experiential inventory, fungible time-boxed capacity, and one-of-a-kind creative goods, respectively. The pricing-control model should follow from what kind of good is being sold, not from what's easiest to build.

Pricing Control Is Not Take Rate

Take rate and pricing control are independent variables, and treating them as one is the single most common category error marketplace teams make. A platform can run a low take rate with total price control (Uber's roughly-quarter cut with algorithmic fares), or a high take rate with zero price control (many creative-services marketplaces charge 15-20% on fully seller-set prices).

The distinction matters because they solve different problems:

• Take rate is a monetization decision — how much value the platform captures per transaction.
• Pricing control is a market-design decision — how much price risk and responsibility the platform absorbs on behalf of either side.

A platform can raise or lower take rate without touching liquidity dynamics much, because sellers and buyers usually don't see the take rate directly — it's baked into the number they experience. Changing who sets the price, by contrast, changes the actual number a buyer sees and the actual payout a seller receives, which is why it moves liquidity far more than a take-rate change does.

Amazon's Buy Box is worth a closer look because it's a hybrid most marketplace PMs underrate: sellers technically set their own price, but Amazon's algorithm decides which seller's listing wins the default "Add to Cart" button — heavily weighted toward the lowest landed price. Sellers keep the pricing pen, but the platform controls who gets to use it profitably, which functionally pushes prices down without ever setting one directly. It's algorithmic pricing pressure without algorithmic pricing authority.

This is also where the jobs customers are hiring your marketplace for matters. A buyer hiring Etsy is often buying uniqueness, so price dispersion is expected and even desirable. A buyer hiring Uber is hiring predictable, fast transportation — dispersion there reads as untrustworthy, not diverse.

The Liquidity and Trust Tradeoffs of Each Model

Every pricing-control model buys liquidity on one side of the market by spending trust on the other, and the size of that trade depends entirely on how commoditized the underlying good is. Understanding which side you're spending trust on is the real design decision — the model itself is just the mechanism.

Two economists who've written extensively on multi-sided platform pricing, David S. Evans and Richard Schmalensee (authors of Matchmakers), frame this as a subsidize-one-side problem: platforms routinely favor whichever side is more price-sensitive, and pricing control is one of the tools used to manage that asymmetry without collapsing participation on either side.

Seller-set: high supply trust, fragile demand experience

Sellers who set their own price feel ownership over their margin, which is why seller-set marketplaces are the default choice for solving the cold-start problem — you need supply to show up before you've earned the right to tell them what to charge. The cost shows up on the demand side: buyers face price dispersion, comparison friction, and inconsistent perceived value across otherwise-similar listings.

• Works well when goods are genuinely differentiated (handmade items, freelance skill level, unique real estate).
• Breaks down when goods are fungible — buyers start price-shopping across nearly identical listings, and the marketplace becomes a race-to-the-bottom auction it never designed to be.

Platform-set: high demand trust, fragile supply retention

Platform-set pricing removes buyer decision fatigue entirely — there's one number, it's comparable across every transaction, and it builds the kind of trust that lets a first-time user convert without research. The cost lands on supply: sellers who feel the price was imposed on them, rather than earned by them, churn faster when a competing platform offers even marginally better economics.

• Works well when the good is a commodity from the buyer's perspective (a ride, a delivery, a unit of standardized labor).
• Requires the platform to actively manage supply incentives elsewhere — bonuses, guaranteed minimums, tiered access — since price is no longer available as a supply-side lever.

Algorithmic/dynamic: fastest liquidity, hardest fairness story

Dynamic pricing is the only model built to solve liquidity in real time rather than in aggregate — it can pull supply toward a demand spike within minutes, which fixed pricing structurally cannot do. Uber's surge multiplier exists because a fixed fare has no mechanism to summon more drivers when everyone wants a ride at once; the price itself becomes the recruiting signal.

Research on Uber's own marketplace, including a widely cited NBER working paper by economists Keith Chen and colleagues, found that letting fares rise with demand meaningfully pulls idle drivers back onto the road during spikes — directionally the opposite of what a fixed fare can do.

The tradeoff is perceptual, not mechanical: a price that moves in response to your urgency reads as opportunistic even when it's functioning exactly as designed. Marketplaces that use dynamic pricing without explaining the mechanism absorb a trust cost that has nothing to do with whether the price was actually fair.

This is precisely why liquidity is the actual product a marketplace sells, not the catalog or the app. Bill Gurley, the venture investor whose Above the Crowd essays are among the most cited liquidity frameworks in marketplace circles, has argued that even small amounts of friction — including pricing friction — can suppress transaction volume disproportionately in a thin market.

A pricing-control model that maximizes short-term match rate but erodes long-term trust on either side is trading a durable asset for a temporary one. The trade rarely shows up in a single dashboard metric until retention has already slipped.

When to Migrate From Seller-Set to Managed Pricing

Marketplaces should migrate from seller-set toward platform-set or algorithmic pricing when supply concentrates into professional sellers, when price dispersion measurably suppresses conversion, or when repeat buyers start treating price comparison as a tax on using the marketplace at all. The migration is rarely instant — it usually moves through an assisted-pricing stage first.

Watch for these signals, roughly in the order they tend to appear:

1. Supply professionalizes. Early sellers are hobbyists or side-hustlers who price intuitively; as power sellers emerge, pricing becomes strategic (and sometimes adversarial), which is exactly when unmanaged price wars start.
2. Price dispersion widens for comparable goods. If two nearly identical listings differ by 3x, buyers stop trusting the category, not just the outlier listing.
3. Conversion drops correlate with price research time. Rising time-on-listing-page combined with falling checkout rate is a classic sign buyers are doing manual price comparison your search and ranking should be doing for them.
4. Repeat buyers churn to platform-set competitors. If a platform-set entrant (think Uber entering markets that used to run on negotiated taxi fares) is taking share, that's evidence the market is ready to trade seller autonomy for predictability.
5. Support tickets skew toward pricing disputes rather than quality or fulfillment — a sign the price-setting mechanism itself, not the transaction, is the friction point.

Market-design economist Alvin Roth, a Nobel laureate for his work on matching markets, has written that healthy markets need thickness, low congestion, and safety to function — and unmanaged pricing can quietly damage all three by letting a handful of underpriced or overpriced listings crowd out the ones buyers would actually transact on.

The migration path that preserves the most trust is rarely a single flip:

• Seller-set — the starting point for almost every marketplace.
• Seller-set with algorithmic suggestions — Airbnb's Smart Pricing model.
• Seller-set within a platform-defined band — autonomy with guardrails.
• Fully platform-set or dynamic — the platform owns the number outright.

Each stage lets the platform earn the right to take more control by proving the suggestion or band improves seller outcomes first.

Mapping this migration against the buyer's actual journey is what separates a pricing change that improves trust from one that just moves a number. A buyer who has been burned by inconsistent pricing before doesn't just want a lower price — they want evidence the platform is now standing behind the number, which is a trust signal, not a discount.

Modeling the Pricing Feedback Loop Before You Choose

Pricing control decisions are hard to reason about in isolation because they're reinforcing loops, not one-way switches. A change that looks purely positive on a static spreadsheet can quietly destabilize the loop it's embedded in:

1. The pricing model changes seller behavior (what sellers list, how they price it).
2. Seller behavior changes catalog quality and price dispersion.
3. Catalog quality changes buyer trust and comparison friction.
4. Buyer trust changes demand volume.
5. Demand volume changes seller economics — which loops back to step one.

This is a textbook case for systems thinking applied to product decisions rather than a linear before/after comparison. A move toward platform-set pricing, for instance, might raise short-term conversion (the visible, immediate effect) while slowly reducing the supply of your highest-quality sellers (the delayed, second-order effect) — and by the time churn shows up in a dashboard, the loop has already reinforced itself for months.

It won't tell you which model to pick; it's built to make the tradeoff you're actually making visible, so the choice between seller-set, platform-set, and algorithmic pricing is deliberate rather than accidental.

Key Takeaways

• Pricing control and take rate are separate levers. One decides who sets the number; the other decides how much the platform keeps of it. Conflating them leads to fixing the wrong problem.
• Seller-set pricing protects supply trust and cold-start liquidity but creates price dispersion that erodes buyer confidence as the catalog matures.
• Platform-set pricing protects demand trust and conversion predictability but requires the platform to replace price as a supply-side incentive with something else.
• Algorithmic/dynamic pricing solves real-time liquidity better than any fixed model, at the cost of a fairness perception the platform must actively explain, not just deploy.
• The right model follows the good, not convenience: differentiated, subjective goods lean seller-set; fungible, time-sensitive goods lean platform-set or dynamic.
• Migration from seller-set to managed pricing works best staged — suggestions, then bands, then full control — rather than as a single abrupt switch.
• Pricing decisions are feedback loops, not one-time settings, and mapping the loop before shipping the change catches second-order supply or demand effects a spreadsheet comparison misses.