A Dive into the Design and Features of Marketplaces

Marketplaces command a lot of focus from venture capitalists, which is unsurprising given some of the most well-known “unicorns” (e.g. Airbnb, Uber) are bonafide marketplaces. The attractiveness of these companies stems from the idea that marketplaces can “build moats” around their business through network effects. Other characteristics of marketplaces garner less attention.

It would be impossible to cover all the relevant marketplace design questions or important KPIs in a single blog post. But for entrepreneurs, investors, sellers, buyers, and partners, thinking critically about the nuances in marketplace design is vital. Slight tweaks can mean the difference between a successful marketplace and one that is D.O.A. We scoured the internet for as much public information* as we could find on ~75 marketplace startups. We analyzed the data for quantifiable insights, qualitative differences, and interesting takeaways related to the following marketplace features: match type, network effects, take rates, and multi-homing.

Who has the final say

Should a marketplace be designed such that the seller makes the final choice to enter into a transaction (“supplier picks”)? Should the buyer make the choice (“buyer picks”)? Should it require an opt-in from both buyer and seller (“double commit”)? The data reveals that marketplaces that are “supplier picks” or “double commit” skew towards services, rather than goods. Why? When a supplier is likely giving up time and skills, vs. a physical product in inventory, it is harder for that supplier to commit the time and effort necessary for the transaction upfront. To ensure high completion rates for matches, marketplaces are designed for the supplier/seller of services/time/skills to have the final say.

Interestingly, “supplier picks” marketplaces are rarer than we expected (10-15% of our sample), which is surprising since some of the major recent marketplace companies employ this model (e.g. Uber, Lyft, TaskRabbit, Fiverr). It’s worth noting that while Uber/Lyft technically function as supplier picks marketplaces, in reality, drivers have limited choice. There are penalties for declinations and they cannot select their riders.

Another reason that supplier choice is rare is marketplaces want to maximize and incentivize liquidity. By getting suppliers to upload inventory beforehand, marketplaces remove friction in matches. It’s harder for buyers to upload “requests” than it is for suppliers to upload an exact offering (e.g. headphones on Amazon). Buyers sometimes don’t know what they want, while suppliers do know what they have.

Why are Uber and Lyft “supplier picks” while GetAround and Turo are “buyer picks”? All involve a service centered around transportation in a car. We found that services involving time, sweat, etc. are more likely to require final confirmation from the supplier, and that a buyer purchasing commoditized services is less likely to care about small idiosyncrasies (e.g. if the Uber is a Honda Civic vs. Toyota Camry). GetAround could work like Uber, but, intuitively, it would be more work for a consumer to create new requests each time they use the service, and then have suppliers review and match those requests. Car-sharing marketplaces instead apply a more efficient matching system: one where suppliers set the availability of their cars and (largely) forget it.

When network effects aren’t network effects

The most important factor in a marketplace’s success is the power of its network effect(s). When founders, VCs, journalists, and analysts talk about network effects, the implication is largely that they are positive for a business, providing a path to quick growth, a (pseudo-) monopoly, and profits. But network effects can be positive or negative.

A “same-side” effect, or externality, refers to the impact one more buyer (seller) has on other buyers (sellers). Cross-side is the effect of an incremental buyer on sellers as a whole (and vice versa). Most, if not all, marketplaces have positive cross-side externalities. This is the classic “network effect” people refer to, where each additional player on one side of the market is good for the other side. However, many analyses ignore same-side externalities, which are often negative. Existing sellers usually don’t want one more seller on the platform, as that creates more competition for the buyers they are trying to attract.

Sometimes same-side externalities (and whether they are positive, neutral, or negative) help explain why some categories race to one “winner” (or standard), while others don’t. Take Microsoft Excel, for example.

First, is Excel really in a marketplace? Excel is a seller in the marketplace of spreadsheet solutions. Competitors with Microsoft Excel are crowded out by its dominance: if a buyer is using Excel, he or she is probably not looking for another spreadsheet solution.

Excel has a positive same-side externality: my Excel use does not hinder anyone else’s, and, in fact, produces a benefit. My usage allows you to share files with me, and vice versa. Also, the bugs I report end up as fixes for you. Nearly all software products scale in this way.

This consumer same-side positive externality is enough to make Excel historically dominant. Interestingly, the challenge to Excel is Google Sheets, which exhibits strong same-side consumer network effects by allowing users to work on a spreadsheet simultaneously. Compare Excel to Uber, where one’s ability to get a ride competes with yours (e.g. surge pricing), and drivers are competing to get riders. With both sides of this marketplace experiencing same-side negative externalities, we can see why both buyers and sellers look for alternatives (e.g. better driver commission on Lyft, the surge/dynamic pricing dance for riders).

A market-clearing take rate has many inputs

In the freelancing or “gig economy” category, take rates (the platform’s commission) can vary dramatically. UpWork’s commission is 10%. TaskRabbit’s is 15%, Fiverr’s is 20%. Handy’s is 10-15%. UpCounsel’s is 18%, while Catalant’s is 25%. What explains these variations? By looking at the data*, the variations are largely explained by transaction size, quality, the work the marketplace is doing, and availability of alternatives.

Larger transaction sizes make sellers more sensitive to the net commission. However, these transactions usually come with more “work” for the marketplace (e.g. vetting, managing payment, contracting, service staff, etc.), which can mitigate sensitivity to the rake. For example, Catalant takes up to 25% “commission” on the total value of the working relationship, which can be a sizable GMV/transaction value, but the company shoulders a lot of matchmaking work.

Avoiding take rate competition and circumvention

Marketplaces will sometimes work to obfuscate their true take rates to avoid downward pressure on pricing or participants escaping. They may also remove the ability to escape (disintermediate) the marketplace altogether by locking up the inventory/payment on one side of the market.

For example, Thumbtack made the (likely correct) calculation that lots of participants eventually circumvent the marketplace, communicating and transacting off the platform which would be hard to monitor and fight. So Thumbtack wants to use a system where sellers pay for leads, allowing them to extract value from the work it facilitates on an ongoing basis. If Thumbtack can accurately price the lead upfront, it can capture substantial economic value from the relationship without worrying about circumvention.

In marketplaces where the interactions are one-off and/or real-time, circumvention is less of an issue. Are you really going to call that same Uber driver who gave you his card every time you need a ride?

Beyond handling basic matching and payment processing, great marketplaces strive to create a better user experience than alternatives on both sides. GrubHub beat out the “menu stapled to the delivery bag” by saving payment information, order information, using previous orders to make recommendations, dynamically adjusting delivery ranges, and reducing errors compared to phone orders.

Multi-homing and how it affects economic rent

Consider the economic rent Microsoft Excel has extracted over the last few decades vs. the (lack of) profits Uber has raked. In marketplaces where your use of a product impacts my use, I’m more likely to “multi-home” (use multiple similar services). Lyft has built a massive business targeting exactly this weakness in Uber: if I’m a driver that wants to enter the market, it may be more attractive for me to jump on Lyft since there may be less driver competition, better expected pay, etc. Unfortunately for Lyft, Uber can target the same multi-homing weaknesses in Lyft.

The magnitude of these same-side negative externalities impacts success of the marketplace and mitigation required. If a dog-walker starts walking my neighbor’s dog before mine, the slight delay in my dog’s walk has a negligible impact on the value I receive. But a rider jumping in front of me to take an Uber I need to the airport right now (think “wait time” or “surge pricing”) is a much bigger negative externality, so I check the Lyft app. Incidentally, Wag has a 40% take rate for on-demand dog-walking, compared to ~20% for Uber/Lyft commission. The average take rate in our dataset was between 15-20%.

In software marketplaces, there is usually a same-side negative externality on the supply-side, but not on the demand-side. A developer’s use of an API from a marketplace does not negatively impact another developer’s use of the API (other than potential minor load issues), and probably improves the quality of the API over time. This is (part of) why VCs love B2B SaaS companies – they could become one-sided marketplaces where scaling does not come with the headache of same-side negative externalities for buyers, but rather lots of same-side buyer benefits.

In other markets, where negative same-side externalities do affect buyers, companies attempt to mitigate multi-homing by employing strategies like price competition (Amazon, enabled largely by cost advantages), subscription models (Instacart), pay-per-lead (Thumbtack), differentiation (product/service, delivery, timing, quality), bundling (OrderHub by GrubHub), branding (Airbnb), supply exclusivity (Uber, Lyft driver bonuses), membership perks (Amazon), and many others.

Origin obsesses over marketplaces, but we want to invest in those with the best chance of success. And we judge that, in part, by evaluating all the features (we discussed a few here) of a particular marketplace. How does it induce single-homing? Is there an unfair advantage in user experience? Is it maximizing take rate in a category? Can it uniquely nurture positive externalities? Founders should design their marketplaces to amplify same-side positive externalities, curb negative externalities, produce the best experience possible for all participants, stave off disintermediation with creative solutions, and justify margin-producing pricing by providing lots of value.

If you have a startup we should talk to, feedback on our dataset or our conclusions, or simply want to chat marketplaces, drop me a line at prashant@originventures.com.

*The information on marketplaces was gathered from press releases, websites, forums, FAQs, social media, and other publicly available information that any savvy internet user could find. We did not use proprietary or confidential data from any Origin portfolio company or company we have evaluated for an investment.

**A note of thanks to Chris Eng, Nikhita Giridhar, and Erin Martinez for help in collecting data/inputs for this piece, Blaze O’Byrne for editing, and Devon Leichtman for editing and publishing.