A Signal Works When It Costs the Wrong Side More

What he was looking at

Michael Spence was a graduate student at Harvard when an adviser told him to read Akerlof’s “Lemons.” He remembered it as “quite electrifying.” His 1972 thesis became this paper.

The market he picked was hiring. Think about it from the employer’s side. You don’t find out how productive an applicant is until long after you’ve hired them, so every hire is a bet.

“To hire someone, then, is frequently to purchase a lottery.” Spence (1973), p. 356

So what can the employer see? The applicant’s education, which is visible and costs something to get.

What he found

Education can sort applicants even if it teaches them nothing, as long as one condition holds.

“a signal will not effectively distinguish one applicant from another, unless the costs of signaling are negatively correlated with productive capability” p. 358

In plain terms, the signal has to be cheaper for the better applicants. If a degree takes less effort for the people who’ll be good at the job, they’re the ones who get it, and the degree becomes a reliable sign of who’s good. Does the degree have to make them better at the work? No. In his worked example it does nothing for the work itself.

“It is productive for the individual, but, in this example, it does not increase his real marginal product at all.” p. 364

He was candid about the cost. In the example, “everyone would prefer a situation in which there is no signaling” (p. 364). And more generally, “Systematic overinvestment in education is a distinct possibility” (p. 368).

Spence shared the 2001 Nobel prize with Akerlof and Stiglitz. In his Nobel lecture he recalls a journalist asking whether you could win the prize for simply noticing that some people in a market know things others don’t. “I thought it was pretty funny,” he said.

What it means

So what makes something a signal? It’s information created by a cost that’s different for different types of people. It doesn’t have to make anyone better. It just has to be easier for the better side to send, so that sending it does the sorting.

That also tells you when a signal fails. If it costs everyone about the same, the low-value side sends it too, and it doesn’t sort anybody.

Where it shows up in a revenue team

• A call instead of a web form, a booked appraisal, a data upload, a paid pilot. Each one only tells you something if it costs a poor-fit prospect more than a good-fit one.
• Friction that costs everyone the same doesn’t sort. It lowers your volume without changing the mix.
• You can design signals. Most of the data that predicts value isn’t in your systems yet. Picking the step or question that the right customers will take and the wrong ones won’t is the art to the science.