Law firm structure creates a reporting relationship that nobody formally evaluates. A third-year litigation associate might take work from ten or twelve partners in a year. Each of those partners is measured on origination, realization, and the outcomes they deliver for clients. None of them is measured on how they staff, brief, correct, or develop the associates doing that work, unless the firm decides to ask.
That gap is what an upward review closes. Associates assess the partners they worked for, on the specific behaviors that make a supervisor effective or exhausting to work for.
What an upward review measures
The useful instruments are narrow. They ask about things an associate observed directly and a partner can act on, rather than about how much the associate likes the partner.
| Dimension | What the question actually asks |
|---|---|
| Feedback quality | Was the feedback specific enough to act on before the next draft |
| Feedback timing | Did it arrive while the work could still change |
| Work allocation | Was the assignment explained, or handed over with no context |
| Availability | Could the associate reach the partner when the matter needed a decision |
| Development | Did the associate get client contact and responsibility at the right pace |
Rated this way, a partner dimension becomes comparable across a practice group and across cycles. "Partners should give better feedback" is a sentiment. "Feedback timing in this group sits eleven points below the firm average" is something a practice group leader can open a conversation with.
The participation problem
Upward review programs fail quietly, through non-response, long before anyone concludes they are not working.
Across SRA-administered review cycles between 2023 and 2025, median associate participation was 87 percent. At the same firms, before SRA took over administration, in-house administered participation averaged 54 percent.
The difference is not enthusiasm for the survey. It is the answer to a question every associate asks before typing anything: who will see this, and can they work out it was me. When the firm holds the responses, the honest answer is that someone at the firm can, whatever the permissions screen says. A diplomatic non-answer is then the rational response, and so is not answering at all.
A program running at 54 percent carries a second problem underneath the first. Non-response is not evenly distributed: it concentrates among the people carrying the sharpest criticism, because they have the most to lose from being identified. What returns is weighted toward partners who were never the problem.
Small groups are where this gets difficult
In a small group, anonymity depends on how many people responded, not on what the invitation promised. A partner supervised by four associates cannot be given a breakdown of those four responses without being able to work out who said what, especially if one of them worked a distinctive matter.
Handling this properly means agreeing a reporting threshold before the instrument goes out, sized against each population rather than picked for tidiness. Cuts below it are reported in aggregate or withheld, never estimated.
Skip that step and the problem surfaces at readout, when a practice group leader asks for a breakdown that cannot be given safely. The associates hear about it, and the next cycle's response rate shows it.
What firms do with the results
Where scores move between cycles, three things are usually in place. Results go to the partner and to one person accountable for acting on them, normally the practice group leader. A firm-average benchmark sits beside each score, since a raw number tells a partner nothing about whether they are the problem. And the same dimensions are asked again soon enough to show movement.
Where nothing moves, the pattern is the reverse: results go to the partner alone, no benchmark is published, and the next cycle is twelve months out.
What upward reviews will not do
They will not tell you why a particular associate resigned. Exit data does that. They will not measure engagement across the firm, which needs a different instrument and a different population.
What they do is make one relationship visible. In our benchmark data, indexed engagement runs at 82 in year one, 68 in year two, and 47 across years three and four before recovering. The trough sits exactly where associates have enough experience to be valuable and enough exposure to several partners to compare them. That is the population an upward review is built to hear from, and the one most likely to leave without explaining why.