Are You Counting What Matters?

GroupLane · July 2026  ·  Realizing a return on benefits is a measurement problem, not a design problem.

You buy a large outcome based on a single number

Employer-paid benefits run about 30 percent of total compensation, and health insurance alone costs roughly $438,000 a year for a sixty-person company. That is the number everyone watches. On the other side of the equation sits turnover, now about $45,000 per head. With ten departures in a year — a 16% rate many would call a good year — the cost runs higher than the entire insurance spend. One side of that equation is measured to the dollar. The other is scattered across recruiting, overtime, training, and rework, and no report puts the two on the same page. A decision based on one visible number and one invisible number can only proceed in the direction indicated by the visible number.

You get what you measure

There is a rule here that has never once failed me. Tell me how you measure a man, and I will tell you how he behaves. Steven Kerr put it in the Academy of Management Journal in 1975, in a paper called On the Folly of Rewarding A, While Hoping for B: people work out what is rewarded and do that, often to the exclusion of everything else.

Hanoi, 1902. The French colonial government had a rat problem and put a bounty on them, one cent a tail. The city filled up with tailless rats — caught, docked, and turned loose to breed, on the sound logic that a dead rat is worth one tail and a live one is worth tails forever. Inspectors eventually found rat farms on the edge of town. Nobody in that story behaved irrationally. The government paid for tails and got tails, in volume.

The modern version is duller and better documented. Pay a crew for a low count of reported injuries and the reporting drops while the injuries keep happening, which is what the Government Accountability Office and OSHA both found with rate-based safety incentives. Now apply the rule to benefits. The one number the program is judged on is cost, so the program gets optimized to be cheap — and cheap is all you can be sure it will deliver.

The value lives in the middle of the chain, where nobody looks

A benefits program pays off through a chain of five links. The coverage exists. The employee knows what it does. He uses it when the moment comes. Using it changes what he does next, so he reports the injury in week one instead of week four, or he does not leave over fifty cents an hour. And that changed behavior finally shows up in turnover, in claims, in your comp modifier.

You measure the first link to the penny. Someone in another department measures the last one loosely and has never been asked to connect it back. Links two, three, and four are not measured by anyone — and those are the links that fail. Take link two. In one study 80 percent of workers said they understood their benefits, and 49 percent did when tested. Nobody measures whether people understand their coverage, so nobody manages it, so it does not happen, and a benefit a man cannot describe cannot change anything he does.

Measure the middle, and make someone answer for it

The fix is cheap and needs no consultant. Add a handful of measures, one for each link the cost number skips.

The data is readily available. The platform that already runs your enrollment records who enrolled and who used it, can host a short comprehension quiz, and timestamps every change — so the reporting lag and the enrolled-versus-not comparison are a query away. The numbers are not hard to get; the trouble is that nobody asks the system for them.

Make the numbers matter, because a measure nobody answers is a warning light with tape over it. Give each one a named owner. Decide in advance what happens when it moves. Put the whole set on the table at renewal, next to the premium — a number that gates a signature gets maintained, and a number in a quarterly deck does not. And reward the response, not the reading. Ask how many times a number moved the wrong way this year and what got done about each one. Acting has a date and a name on it, and it is the one thing that cannot be faked into looking good while producing nothing. You already know what the program costs. This is how you find out what it bought.

← All research