Discovery & Search Costs#
Why looking for a partner is a search problem
Four terms, before anything else#
- search cost
- cost per candidate
- signal quality
- false-positive rate
Get them straight now — the rest of the semester reuses them without redefinition.
Search cost#
Time spent evaluating one candidate is time not spent finding a better one.
The moment you stop looking is a decision — even when it doesn’t feel like one.
What sets the search cost#
A search technology — a friend group, a workplace, an app — doesn’t just supply candidates. It sets the price of finding them.
| Term | What it means |
|---|---|
| Cost per candidate | time + effort to evaluate one option |
| Signal quality | how much a candidate’s information tells you the truth |
| False-positive rate | how often “promising” turns out to be wrong |
Search problem vs. matching problem#
- Search: one side deciding when to stop looking
- Matching: both sides’ preferences reconciled at once
This course spends three more weeks on the former (screening: weeks 4–6) before it earns the latter (Week 8).
More options is not obviously an improvement.
Why more isn’t obviously better#
A thicker market — more candidates — sounds like a pure win. Roth’s work on market design names the catch: thickness brings congestion, and congestion is its own failure mode, independent of how many good candidates are actually in the pool.
More candidates only helps if the cost per candidate stays low enough to actually evaluate them.
Now, back to the point#
Same four terms. One scenario.
Two channels, one persona#
Persona: Priya, new to the city, trying both at once.
| Metric | Channel A — mutual friends | Channel B — a swiping app |
|---|---|---|
| Candidates seen per week | ~1 | ~40 |
| Cost per candidate | high (a whole dinner) | low (a few seconds) |
| Signal quality | high (a friend vouched) | low (a curated photo set) |
| False-positive rate | low | high |
What changes about the strategy#
| Question | Channel A | Channel B |
|---|---|---|
| Optimal move | say yes to a decent first date | screen hard before agreeing to meet |
| Why | each candidate is already pre-filtered and expensive to skip | most candidates are noise; the cost of looking is what’s cheap, not the cost of being wrong |
| Stopping rule | stop early | keep sampling, filter later |
Mapping it back#
- Search cost — an order of magnitude apart between channels, not a rounding difference
- Cost per candidate — is what makes “meet everyone” rational on A and irrational on B
- Signal quality — is why A can skip screening that B can’t
- False-positive rate — is why B’s 40 candidates aren’t 40 real options
A cheaper channel doesn’t hand you a better outcome. It hands you a different optimization problem.
Tutorial 2#
Model a fictional persona’s search technology as a cost function — for two different channels, side by side.
What today doesn’t settle#
Today names the trade-off; it doesn’t solve it. Given a real cost function, when exactly should you stop looking?
Week 6 does the math properly.
See you next week#
Week 3: Cheap Talk vs. Costly Signals