People analytics arrives at most firms as a dashboard. Headcount, attrition, time to fill, a diversity breakdown, all accurate and none of it answering a question anybody asked.
The useful version is narrower. A measure earns its place if somebody can name the decision it would change, and on that test most firms need three numbers rather than thirty.
One: attrition by cohort and by supervising partner
Firm-wide attrition is a board number. It tells leadership the rate and nothing about the cause, and it moves for reasons that have nothing to do with anything the firm did.
Cut the same data by class year and by the partners who supervised the leavers, and it starts naming things. A firm losing third and fourth-years at twice the rate of its second-years has a different problem from one losing people across the board. A practice group losing people who all worked substantially for the same two partners has a problem it can act on this quarter.
This is the measure firms are most reluctant to build, because it produces findings about named partners. That reluctance is the reason it is worth building: it is the only one of the three that identifies where to intervene.
Two: engagement by tenure, tracked across cycles
A single engagement score is a mood reading. The same score cut by tenure and compared with the previous cycle is a trajectory.
The shape is consistent enough to be useful as a baseline. Across SRA's benchmark, engagement runs at 82 in year one, 68 in year two, 47 across years three and four, and 55 across years five and six. A firm whose own curve dips earlier or deeper than that has something specific to look at, and a firm whose curve is flat has either an unusual culture or a participation problem, which the third measure below will tell it apart.
The value is in the comparison between cycles, which means the first cycle buys you very little and the third buys you most of it. This is the main argument for starting now rather than when the program is perfect.
Three: response integrity
The least glamorous of the three and the one that determines whether the other two mean anything.
Participation rate by population, the spread of the rating distribution, and the proportion of free-text fields left empty. These are measures of the instrument rather than of the firm, and they tell you whether to believe the rest.
A cycle at 54% participation with scores clustered in the top two bands and most comment boxes blank has not measured engagement. It has measured what people were willing to put in writing, and reporting it as engagement will send leadership after the wrong problem. For reference, independently administered upward reviews in SRA's programs run at 87% median participation, against 54% at the same firms when they administered the programs themselves.
What a firm already has, and what it has to collect
Most of the first measure is already in the building. Joiners and leavers sit in HR records, and who supervised whom can be derived from matter and time entries rather than reconstructed from memory.
The second and third have to be collected, and collected the same way each time. A question reworded between cycles produces a trend line that measures the rewording. This is the most common way a people analytics effort quietly destroys its own baseline in year two.
The constraint that shapes everything
Any cut fine enough to be actionable is also fine enough to identify people, and in a law firm the populations are small. A practice group of nine, split by tenure and gender, produces cells of one.
So the threshold is not a compliance afterthought, it is a design input. Decide before fielding which cuts will be reported and at what minimum, and accept that some questions the firm wants answered cannot be answered safely at its size. A firm that reports a cell of two has told everyone in that cell what their colleague said.
Who should own it
Not IT, and not whoever is best with a spreadsheet. The measures above only produce value if somebody has standing to put a finding about a named partner in front of leadership, which is a question of position rather than of analytical skill.
In practice this tends to work when professional development or the COO owns the questions and an external party administers anything confidential, because the person who has to present the finding should not also be the person the respondents think might read their answers.
SRA administers the collection and reporting side of this for US law firms, with suppression thresholds fixed before fielding and benchmark comparison across more than a hundred Am Law firms. The methodology sets out the threshold rules, the benchmark covers what the comparison data contains, and the figures quoted above are listed with their definitions and periods on figures and sources.