Most US law firms run something they call the associate evaluation: a form, a cycle, a meeting, a rating that feeds compensation. Ask two partners at the same firm what it is for and the answers usually differ.

An associate evaluation is not a single instrument. It is up to four, each answering a different question, and they cannot be collapsed into one form without losing what each is good for.

The four instruments inside one evaluation

The downward evaluation is the partner's assessment of the associate. It is attributed, and it should be: a judgment used in compensation and progression decisions needs an author who can be asked about it. Where a rating affects somebody's pay, the reviewee has to know whose judgment it is in order to contest it.

The self-assessment is the associate's account of their own year. Its value is not the content but the comparison. Where an associate rates themselves well above or below their reviewers, you have the most useful conversation in the cycle, and you have it in a form that does not depend on anyone being persuasive.

The upward review runs the other way: the associate on the partners who supervise them. This one has to be confidential and reported only above a threshold, because an associate assessing the partner who staffs their matters, approves their bonus and votes on their progression answers honestly only when identification is not available to the firm. Most firms either skip it or administer it themselves.

The 360 asks the same competency questions of everyone who works with the lawyer, split by rater group so peer, upward and downward views can be compared rather than averaged into a number that describes nobody's experience.

Running the first two and calling it an evaluation is common. It produces a record of what partners think, which is worth having and is not the same as knowing how the associate is actually supervised.

Who gets asked, which decides more than the form does

In most firms the reviewer list comes from an email. Partners are asked which associates they worked with over the past year and reply from memory, often weeks after the question and months after the work.

Those lists are wrong in predictable directions. Recent matters are over-represented, short assignments go missing, and the associate pulled onto a difficult matter for six weeks is sometimes left off.

The firm already holds a dated record of who worked with whom, in the time entries. Deriving the rater pool from matter and billing data rather than recollection does three things: it shows before fielding which reviewees have enough raters to report at all, it makes an omission a visible fact rather than an oversight, and it replaces case-by-case judgment about who is close enough to assess somebody with a consistent floor in hours.

This is covered at length in a separate piece on rater selection.

Criteria that predict something

Evaluation criteria at law firms tend to drift toward what is easy to count. Billable hours are the obvious case: they are recorded, comparable and almost useless as an evaluation criterion, because they measure how much work an associate was given at least as much as how well it was done.

The criteria that carry information are behavioral and specific to the work:

  • Work product against the standard the matter required, not against the associate's seniority in the abstract.
  • Legal judgment, meaning the quality of the call made when the answer was not in a precedent.
  • Responsiveness and reliability, which partners rate consistently and associates are rarely told about until it is a problem.
  • Client handling under deadline, where it is visible.
  • Supervision of juniors, which starts mattering years before anyone is formally responsible for it.
  • Development of the practice, scoped to what is realistic at that level rather than imported from the partnership criteria.

A criterion earns its place on the form if a reviewer can give an example of it from the past year. Anything that cannot be exemplified collects a middle rating from every reviewer, and a column where everyone scores in the middle cannot be acted on.

The cycle

A program that works runs about eight weeks from design to readout, and the sequence matters more than the duration.

Instrument design and population definition come first, including the reporting threshold, which has to be fixed before anyone is invited. Once responses exist the threshold cannot be loosened to produce a cut somebody wanted, and a firm that learns this in week six has lost the chance to design around it.

Self-assessments are collected and held until reviewers have filed, so a reviewer's rating is not anchored to what the associate said about themselves. Fielding and reminders follow. Then aggregation and threshold suppression, free-text review for identifiability, and only then reports: individual, practice group, firm.

Calibration belongs after the individual reports and before anything is communicated. A consistently generous or harsh reviewer can only be shown it against their own group. Without that comparison, a 4 from a generous reviewer and a 4 from a severe one enter the record as the same number.

Five ways evaluations fail

The results are uniformly positive. A distribution with no spread has usually measured the reviewers' reluctance rather than the associates' performance. This is the most common failure and the hardest to raise, because nobody wants to argue that the ratings are too kind.

The upward half is administered by the firm. Associates read "confidential within our platform" correctly, as different from "not available to the firm". SRA's median participation on independently administered upward reviews is 87%, against 54% at the same firms when they ran the programs in-house. Those are the same firms asking much the same questions, with the responses held somewhere else.

Thresholds are decided after the fact. In a practice group of three, a partner cannot be given an individual upward report without exposing the respondents, and the arithmetic does not improve because leadership wants the cut.

The cycle ends at the report. Development actions agreed in a readout and recorded in a document get revisited when somebody remembers. Most of the value in an evaluation sits in the part after the report, which is where an annual process tends to stop.

Nothing goes back to the participants. Not the raw results, but what the firm heard and what it intends to do. The first cycle usually works without it. Participation in the second drops, and a program that stops after one cycle never produces a second data point to compare against.

What the data says about timing

Evaluations are usually scheduled for the firm's convenience. The engagement data suggests when they matter most. 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.

The third and fourth year trough is where firms lose people they have spent four years training, and it is the cohort whose evaluations most need to produce a conversation rather than a rating. A firm that finds the dip a year later, in an exit survey, is reading it about people who have already gone.

Where to start

If the firm runs downward evaluations and nothing else, add the upward half and administer it outside the firm. If it runs both and the results are flat, the problem is custody or threshold design rather than the questions. If the results have spread and nothing happens afterwards, the problem is in the follow-up, which is a question about who owns the actions rather than about the instrument.

SRA designs and administers upward reviews, downward evaluations and 360-degree feedback for US law firms, and will tell you what your group sizes support before an instrument is agreed. The methodology sets out how thresholds are fixed and what a report can carry, and every figure above is listed with its definition and period on figures and sources.