Trust and Reputation: Making Strangers Willing to Transact | Marketplace Dynamics Tutorial

Why Trust Is the Core Product

A marketplace connects strangers. Strangers don't naturally trust each other enough to transact. The platform's job is to bridge that gap.

Think about what you're asking a buyer to do on Airbnb: send money, in advance, for a stay in a stranger's house in a city you've never been to. That is a huge trust ask. It works because the platform has built a trust stack: reviews, verified identity, payment escrow, host cancellation policies, insurance, a dispute-resolution team.

Take any of those away and the business collapses. Trust is not a feature; it is the product.

Every marketplace builds its own trust stack, calibrated to the risk of its transactions. Buying a used mug on Etsy is low-stakes; you can accept thinner verification. Hiring a nanny is high-stakes; you need background checks, ID verification, and possibly a phone screen.

The rule: trust investment scales with transaction risk.

The Trust Stack

A typical marketplace trust stack, in rough order of adoption:

Identity verification (email, phone, social login; upgrading to ID check, facial match, or full KYC as needed)
Payment infrastructure (escrow, credit card charge-backs, dispute flows)
Reviews and ratings (bilateral where possible, to prevent retaliation)
Fraud detection (behavioural signals, device fingerprinting, risk scoring)
Dispute resolution (human agents, SLAs, refund policies)
Insurance (platform-backed policies for high-value categories)

You layer on what the category needs. Not every marketplace needs all six. But the high-trust categories (lodging, labour, large-ticket goods) need something in every row.

Reviews and Ratings

The workhorse of marketplace trust. Users report on their experience, and the aggregated reputation constrains bad actors over time.

Design decisions that matter more than people expect:

Bilateral vs unilateral

Unilateral: only the buyer reviews the seller. Simple. Risks: buyers can punish sellers for things outside their control; sellers have no reputational lever
Bilateral: both sides review each other. Airbnb famously does this. Incentivises both sides to behave. Risk: retaliation (a bad guest might leave a bad host review to pre-empt being called out)

Airbnb's solution: reviews are hidden until both sides submit, or until the review window closes. Neither party can retaliate because neither knows what the other said.

Rating scales

5-star scales are the default. They also have a known problem: almost every rating is a 4 or a 5. "Rating inflation" on Uber, Airbnb, DoorDash converged to the point where anything below 4.5 is essentially a bad rating.

Alternatives:

Thumbs up / thumbs down (YouTube, Netflix). Clearer signal, but loses gradation
Multiple dimensions (cleanliness, accuracy, communication). More useful; harder to aggregate
Free text only with sentiment analysis. Richer but slow

No perfect answer. Many platforms run both: stars for the aggregate signal, text for the richer story.

Volume and time decay

A seller with 500 reviews averaging 4.8 is more trustworthy than one with 3 reviews averaging 5.0. Platforms often apply a Bayesian smoothing: new sellers start at the platform average and move toward their true rating as reviews accumulate.

Time decay matters too. A restaurant's 4.8 average from 2019 is not the same as 4.8 from last month. Recent reviews should weigh more.

Verification

What you verify depends on what can go wrong.

Email + phone: table stakes. Cheap, catches casual abuse
Government ID: catches identity fraud, enables legal recourse
Selfie match: confirms the ID belongs to the user
Background check: for high-trust roles (nannies, drivers in some jurisdictions)
Professional certification: for specialised categories (medical professionals, contractors)

Each step adds friction and cost. Don't over-verify. A seller on a £5 craft marketplace doesn't need a notarised passport. A nanny does.

Payments and Escrow

Payments are a trust mechanism disguised as payment infrastructure.

Escrow: the platform holds the buyer's money until the transaction completes. The seller knows they'll get paid; the buyer knows they can dispute if the product doesn't show. Uber, Airbnb, and most e-commerce platforms do this
Cancellation and refund policies: clear, codified rules that both sides agreed to on signup. Makes disputes deterministic
Chargebacks: most cards and PayPal let buyers reverse a charge. Understanding this is essential because it shifts risk to the platform or the seller
Payout timing: how fast sellers get paid. Faster is better for sellers (more liquidity), worse for risk (less time to catch fraud). Every platform picks a balance. Weekly is common

Payment providers (Stripe, Adyen, PayPal) carry most of the technical load. The platform's job is designing the dispute and payout policies on top.

Insurance

For high-value or high-stakes transactions, platform-backed insurance moves risk from users to the platform.

Airbnb's Host Guarantee / AirCover: up to $1M in damage coverage
Uber's commercial auto insurance: fills gaps in driver personal policies
eBay's Money Back Guarantee: buyer protection against fraud or misrepresentation

Insurance reassures both sides. It is expensive; it pays for itself through higher conversion and lower churn. Platforms that launch high-risk categories without insurance learn the hard way.

Dispute Resolution

No trust stack is complete without humans who handle the bad cases.

Good dispute resolution:

Clear SLAs ("we respond within 24 hours")
Published rules buyers and sellers agreed to
A tiered path: bot first, then automated refunds, then a human agent
Evidence-driven outcomes, not just sympathy-driven

Bad dispute resolution:

Defaulting to "buyer is always right", driving honest sellers away
Defaulting to "seller is always right", driving buyers away
Slow response times that let both sides stew
Opaque processes that neither side feels heard by

The dispute team is a cost centre. It is also the team most responsible for the perception of fairness. Under-investing here breaks the marketplace quietly.

Handling Bad Actors

Some fraction of your users will be bad. Scammers, fake sellers, abusive buyers, platform manipulators.

Tools:

Account limits and holds: new sellers have lower volume caps until trust is built
Risk scoring: machine-learned flag for suspicious behaviour
Manual review queues: flagged transactions go to a human before being paid
Strict bans: repeat offenders are removed. With high-friction categories, bans need to persist across new accounts (ID, payment, device)

The hardest part is the grey zone. A seller with 10 good transactions and one bad one: coincidence or early warning? Platforms that err too hard one way lose honest supply; the other way, the marketplace becomes a dumping ground.

The fraud arms race

Bad actors evolve. Fraud that worked last year gets blocked, so it mutates. Fake accounts become harder to detect. Fake reviews generate more naturally. Phishing targets your support team.

Every mature marketplace has a fraud team. It is never finished. Budget for it early; it is cheaper to design fraud-resistant systems upfront than to retrofit them.

Trust Recovery

Sometimes the marketplace fails a user. A booking falls through, a product is fraudulent, a driver is rude. The platform's response defines the relationship going forward.

The pattern that works:

Acknowledge fast
Refund or re-match quickly, even before blame is assigned
Investigate in the background
Follow up with a resolution and (usually) a small goodwill gesture

The pattern that fails:

Fight the user on whether they're right
Make them jump through procedural hoops
Win the argument; lose the user

A marketplace's reputation is more fragile than it appears. One public incident handled badly can undo years of brand work.

Trust in Cold Start

Early marketplaces have no reviews, no reputations, no data. Users have to take more risk per transaction. This compounds the cold-start problem.

Ways to bridge:

Show your work. Explain verification, insurance, escrow prominently
Take the risk yourself. Early Airbnb refunded guests directly if hosts flaked; early Uber covered ride problems out of pocket. Expensive, but buys you the behaviour of a mature platform before you earn it
Use proxy trust. Link to external credentials (LinkedIn, Google reviews, industry certifications) to borrow trust from other platforms
Limit the blast radius. Small transactions first. Build reputation. Open higher-ticket as both sides grow trust

Common Pitfalls

"We'll trust users until they prove otherwise." You'll be disappointed. Trust-by-default is a reasonable stance for a friend, not a platform. Build the scaffolding to detect bad behaviour before it costs you

"Our ratings look great." Rating inflation makes a 4.7 average mean "average" on some platforms. Compare against the distribution, not the absolute

"Insurance is too expensive to launch with." Insurance is often priced as a percentage of GMV. If you can't afford it, the risk of going without it is worse. Build it in from the start

"We don't have bad actors yet." You do; you just haven't noticed. Build the detection infrastructure before you need it

"Reviews will keep sellers honest." Only if buyers leave honest reviews. Fear of retaliation, reciprocity pressure, and rating inflation all conspire against honest reviews. Design against these from day one

Next Steps

Continue to 07-network-effects.md for the compounding dynamics that make marketplaces defensible.