What Matching Is Not in Two-Sided Marketplaces

When people hear "matching", they often think about an algorithm that picks the best provider for a customer, or the best job for a worker. That is part of it, but it is not the whole story. In many marketplaces, the biggest failures come from treating matching as a simple ranking, recommendation, or automation problem.

Good matching is not just about finding a technically suitable option. It is about helping both sides of the market make a good decision, at the right time, with enough confidence to act.

To understand matching properly, it helps to first be clear about what matching is not.

All determination is negation.

Baruch Spinoza

Matching is not just search

Search helps users find options. Matching helps users move towards a good outcome.

In a marketplace, showing a list of available providers, jobs, services, or customers is only the beginning. A user may still need to compare options, assess fit, understand risk, check availability, negotiate, and make a decision.

If the marketplace simply improves search results but does not reduce decision effort, uncertainty, or friction, it has not really solved matching. It has only made discovery slightly better.

Matching is not just ranking

Ranking is often treated as the core of matching: put the "best" option at the top and the job is done.

But the "best" option is rarely universal. The best provider for one customer may be a poor fit for another. The best job for one worker may be unattractive to someone else. Fit depends on context, timing, preferences, constraints, trust, price, location, availability, and willingness to engage.

A ranking model can help, but matching requires more than ordering a list. It requires understanding whether both sides are likely to see value in the interaction.

Matching is not optimisation for one side only

A common mistake is to optimise for the customer side and forget the supply side, or the other way around.

For example, a marketplace may recommend the highest-quality providers to every customer. That may look good for customers in the short term, but those providers may become overloaded, ignore requests, or become selective. Other capable providers may receive no opportunities. Over time, the marketplace becomes less healthy.

In a two-sided marketplace, matching must consider both sides. A good match is not just "good for demand" or "good for supply". It is good enough for both sides to engage.

Matching is not giving users more options

More choice can feel like a better marketplace, but too many options often create confusion.

If a customer receives 50 possible providers, they may not know who to contact. If a provider receives many low-fit opportunities, they may ignore them. In both cases, the marketplace creates noise instead of confidence.

Good matching often means showing fewer, better, more actionable options. The goal is not to maximise the number of options. The goal is to increase the chance of a successful interaction.

Matching is not forcing the perfect match

In real marketplaces, perfect matches are rare.

Availability changes. Preferences are incomplete. Users do not always know what they want. Supply may be limited. Demand may be urgent. Price, timing, distance, and trust all create trade-offs.

A matching system that waits for the perfect match may create delays and missed opportunities. A better system helps users understand the best available trade-offs and make a confident decision.

Matching is often about finding a good-enough match that can actually happen.

Matching is not just personalisation

Personalisation is about tailoring the experience to a user. Matching is about creating a successful interaction between two parties.

A marketplace can personalise recommendations for one user and still create poor outcomes overall. For example, it may keep recommending the same highly popular providers, even if those providers are unlikely to respond. Or it may recommend jobs that look relevant but are not attractive enough for workers to accept.

Matching needs to consider mutual fit, not just individual preference.

Matching is not the same as allocation

Allocation means the platform decides who gets what. Matching usually means the platform helps two sides find and choose each other.

This distinction matters. In some marketplaces, the platform can directly assign supply to demand. In many others, users still need to express interest, accept, negotiate, or build trust.

If the marketplace behaves as if it can allocate, but users still behave as if they have choice, the system will fail. It may recommend or push interactions that one side does not actually want.

Good matching respects the level of agency users expect in the marketplace.

Matching is not only an AI problem

AI can improve matching, but it does not replace marketplace design.

A model may predict fit, rank options, summarise profiles, generate recommendations, or automate parts of the interaction. But if the underlying product flow is unclear, incentives are misaligned, supply is thin, or users do not trust the recommendation, AI will not fix the problem.

Matching is a product, data, design, operations, and market design problem. AI is one tool inside that system.

Matching is not a solution to every marketplace problem

Sometimes what looks like a matching problem is actually a liquidity problem, a quality problem, a pricing problem, or a trust problem.

For example, if there are not enough providers in a region, better matching will not magically create supply. If customers do not trust providers, ranking them differently may not help. If jobs are unattractive, recommending them more aggressively will not make providers accept them.

Before improving matching, it is important to ask: is the marketplace failing because users cannot find the right options, or because the right options do not exist?

Matching is not measured by clicks alone

Clicks, views, and applications can be useful signals, but they do not always represent successful matching.

A user may click because they are confused. A provider may apply without being a good fit. A customer may receive applications but still not choose anyone. A match may happen but lead to poor retention.

Good matching should be measured through outcomes, not just activity. Useful metrics may include response rate, acceptance rate, time to match, successful transaction rate, repeat engagement, satisfaction, retention, and marketplace balance.

The best metric depends on the marketplace, but the principle is the same: matching should create value, not just movement.

Matching is not static

A good match today may not be a good match tomorrow.

Availability changes. Users learn. Preferences evolve. Supply and demand shift. Marketplace conditions change by season, location, category, price, and urgency.

This means matching cannot be treated as a one-off model or fixed rule. It needs feedback loops. It needs monitoring. It needs to learn from what users actually do, not only what they say they want.

Matching is not just about efficiency

Efficiency matters, but matching also shapes trust, fairness, and marketplace health.

A system that always optimises for short-term conversion may create long-term damage. It may over-expose some providers, under-expose others, reduce diversity, or create winner-takes-all dynamics. It may also push users into matches that convert quickly but do not last.

Good matching balances short-term outcomes with long-term marketplace health.

So, what is matching?

Matching is the process of helping the right demand and the right supply find, assess, and choose each other under real-world constraints.

It reduces uncertainty. It lowers decision effort. It improves confidence. It helps both sides understand why an interaction is worth pursuing. It makes the marketplace feel less like a directory and more like a trusted system.

The best matching systems do not just answer "who is available?" or "who ranks highest?"

They answer better questions:

• Who is likely to be a good fit?
• Who is likely to respond?
• Who is likely to accept?
• Who is likely to deliver a good outcome?
• What trade-offs should the user understand?
• What information is missing before a decision can be made?

And most importantly: will this interaction create value for both sides?

That is the real work of matching in a two-sided marketplace.

A slightly geeky way to say it

If we wanted to describe matching in a simple mathematical way, the wrong version would look like this:

\[ \text{Matching} \neq \arg\max_{s \in S} \text{score}(d, s) \]

Where \(d\) is the demand side, for example a customer or client, and \(s\) is the supply side, for example a provider or worker.

In plain English: matching is not just picking the provider with the highest score for a customer.

It is also not:

\[ \text{Matching} \neq \text{Search}(d) \]

\[ \text{Matching} \neq \text{Ranking}(S) \]

\[ \text{Matching} \neq \text{Personalisation}(d) \]

\[ \text{Matching} \neq \max |S| \]

More options do not automatically mean a better match. In fact, too many options can reduce confidence and make decisions harder.

A better way to think about matching is:

\[ \text{Match}(d, s) = f(\text{fit}, \text{availability}, \text{trust}, \text{constraints}, \text{incentives}, \text{willingness}) \]

Or, even more simply:

\[ \text{Good Match} = \text{Mutual Fit} \times \text{Availability} \times \text{Trust} \times \text{Willingness to Act} \]

The multiplication is intentional. If any critical part is close to zero, the whole match becomes weak.

\[ \text{High Customer Fit} \times 0 \text{ Provider Interest} = 0 \text{ Real Match} \]

This is why a recommendation that one side loves but the other side ignores is not really a match. It is just a recommendation with optimism.

A successful marketplace match needs both sides to have enough reason to move forward:

\[ \text{Successful Match}(d, s) = \begin{cases} 1, & \text{if both sides are willing and able to proceed} \\ 0, & \text{otherwise} \end{cases} \]

So the simplest formula may be:

\[ \text{Matching} = \text{fit that can actually happen} \]

The best matching systems are not the ones that maximise clicks, lists, or recommendations. They are the ones that increase the probability of a good interaction between both sides of the marketplace.

\[ \text{Great Matching} \Rightarrow P(\text{good outcome for both sides}) \uparrow \]