#601 1922 · Life Insurance Sales Research Bureau, Carnegie Institute of Technology (John M. Holcombe, Jr., formerly Manager, Sales Research Division, Phoenix Mutual Life Insurance Co.) · Life insurance / personnel selection
A life insurer aimed its own actuarial math at its hiring desk, scoring recruits the way it scored strangers' mortality
the problem
Sales managers hired agents on gut feel or a single crude proxy, and most washed out
background
In the early 1920s, life insurance companies hired sales agents largely on a manager's subjective impression of a candidate, sometimes narrowed to one crude proxy — Holcombe cites age alone as a common single-factor screen. Turnover among agents was severe and expensive: a company spent real money training and licensing a man before anyone learned whether he could actually sell on commission, and the industry had no systematic way to tell in advance who would succeed and who would quit within the year.
The obvious fix was better interviewing, a stricter age cutoff, or more experienced sales managers doing the choosing. None of it addressed the underlying problem: nobody had ever gone back and checked whether a manager's judgment, or any single proxy, actually correlated with which hires succeeded. One person's read on a stranger's future performance in a genuinely variable job — commission selling, subject to territory, timing and luck as much as skill — was simply trusted, never tested.
what everyone would do
Train sales managers to interview more carefully, or tighten a single screening rule like a minimum age. This fails because it never tests whether the manager's judgment, or the single proxy, actually predicts who succeeds — it substitutes one untested intuition for another, and turnover stayed high because nobody had gone back to check which biographical facts about past hires actually correlated with which of them made it.
what they saw
Holcombe saw that his own industry already had, sitting in its actuarial department, the exact tool his sales department needed and had never turned on itself: life insurance is nothing but combining many individually weak facts about a stranger into one number that predicts an outcome nobody can observe in advance. Hiring a salesman is the identical structure — predicting an unobservable future outcome from a set of weak biographical signals — and the company already knew how to do this kind of math. It had simply never occurred to anyone to point the actuary's method at the actuary's own hiring desk.
the move
Holcombe, running the newly formed Life Insurance Sales Research Bureau at Carnegie Institute of Technology, pointed his own industry's core method at itself. A life insurance company's entire business is combining many individually weak facts about a stranger — family history, occupation, habits — into one actuarially validated number that predicts an unobservable future outcome. Holcombe applied that exact logic to hiring: he pulled the personal-history blanks of 447 working agents, divided them into success and failure by an actual sales-outcome line, and empirically tested each biographical item — age at entry, marital status, dependents, insurance already owned, schooling, prior selling experience, home ownership, club memberships, investments — against real outcomes rather than assuming which ones mattered, discarding items that didn't predict and weighting the ones that did into a single composite score.
why it works
The method replaces one unverified intuition — a manager's read on a candidate — with a score built from an actual observed correlation between the same biographical facts and the same outcome, measured on hundreds of prior agents, so it inherits accuracy from real history rather than from any one person's memory or hunch. Discarding items that didn't predict (Holcombe found agents aged 24 to 32 scored no differently than the group as a whole) kept the score from being inflated by folk wisdom that felt plausible but didn't hold up, while weighting the items that did predict let many individually weak, unreliable signals compound into one usably strong one. Because the score was built on one sample of 447 agents and then tested on a separate sample of 148, the 80%-versus-43% gap reflects genuine predictive power rather than the method simply re-describing the data it was fitted to.
the payoff
On a held-out validation sample of 148 agents, men who scored above +1.5 on the composite succeeded 80% of the time; men scoring +1.5 or below succeeded only 43% of the time. Presented to the Taylor Society in March 1922 explicitly as a tool to cut the cost of agent turnover, Holcombe was careful to call the method tentative rather than proven at scale, and pointed to his own scatter chart showing at least one man who scored very high on the formula and still failed badly in the field.
where it breaks
Holcombe's own presentation is honest about the limits: the validation sample was still small by actuarial standards, the score hadn't been proven out on a truly independent future cohort at the time he spoke, and he stated plainly it should inform a sales manager's judgment, not replace it. His own scatter chart shows at least one agent who scored very high on the formula and still failed badly — a reminder that a composite score describes a population tendency, not a guarantee about any individual case. The approach also depends on having enough historical hires with recorded outcomes to find real correlations in the first place; a role or company too new, or too small, to have that history has nothing to validate a score against, and borrowing another organization's weights wholesale risks importing correlations that don't hold in a different context.
what came after
The Bureau continued operating after Holcombe's presentation — psychologist Marion Bills, working under Walter V. Bingham's Division of Applied Psychology at Carnegie Tech, conducted further sales-selection research there in 1924–25 — and Holcombe's 1922 paper is cited today in academic histories of industrial and organizational psychology as an early landmark in what the field would only much later start calling predictive hiring or people analytics.
references
- [1]A Case of Sales Research: Report on First Steps in a Study of the Selection of Life Insurance SalesmenBulletin of the Taylor Society, Vol. VII No. 3, 1922archive.org
- [2]The Early Years of Industrial and Organizational PsychologyCambridge University Press (Andrew J. Vinchur), 2018cambridge.org