Annual reviews fail because they ask managers to remember a year. Monthly reviews only ask them to remember a month.
A monthly performance review is a twenty-minute conversation about the last four weeks, grounded in what was actually recorded rather than what anyone remembers. It replaces the annual review's central problem — reconstructing twelve months from the most recent six weeks — with twelve small, accurate records. By the time an annual summary or increment decision is needed, it writes itself from notes that already exist.
Three well-documented failure modes, all structural:
Monthly reviews dissolve all three, mostly by lowering the stakes of any single conversation.
Twenty minutes, once a month:
Question four is the one managers skip and employees value most. Silence after asking it is uncomfortable and productive — the useful answer usually arrives after the pause.
The difference between a useful review and a pleasant chat is whether it opens with evidence. If your team keeps a daily task log, you already have the month in front of you: what was worked on, for which client and project, how long it took, and what was blocked.
What that changes:
Reviews should start with the record and then discuss it — not start with an impression and then look for evidence supporting it.
After each review, record: what was delivered, what needs to change, and what was agreed for next month. Three or four lines. Two minutes.
The payoff is compounding. Twelve monthly notes make the annual summary a compilation exercise rather than an act of archaeology, they give the increment cycle a documented basis instead of an impression, and if a performance issue ever needs formal handling, the contemporaneous record already exists. Long-form review documents, by contrast, get written once, in December, and are useful to nobody.
If every monthly review might affect pay, employees will manage the conversation defensively and you will stop hearing anything real. Keep the monthly review about work, and hold compensation to its own cycle with its own meeting.
This does not mean the two are unconnected — twelve months of monthly notes should be the primary input to the increment decision. It means the employee knows which conversation they are in, which is what makes candour possible in the monthly one.
Drafting is a reasonable use: given a month of logged tasks, a model can produce a first summary of what someone worked on. That saves a manager ten minutes of assembly.
Two conditions make this acceptable. First, it is a draft that a human edits — nobody should receive an assessment their manager did not write and does not stand behind. Second, it is grounded in real logged work, not generating plausible-sounding performance language from nothing.
What AI must not do is decide. Ratings, increments, promotion and exit are judgements a person is accountable for. The full boundary is drawn here.
| Cadence | Monthly, 20 minutes per person |
|---|---|
| Format | What was delivered · what was blocked · next month's priorities · what you need from me |
| Starting point | The logged record of work, not recollection |
| Written output | 3–4 lines per review — delivered, change, agreed next |
| Annual review becomes | A compilation of twelve monthly notes |
| Keep separate | Performance conversation and pay conversation |
| AI's role | Draft from logged work for a human to edit — never to decide |
Monthly, in a short twenty-minute conversation, works far better for small teams than a single annual review. It removes recency bias, keeps feedback close to the work it refers to, and turns the annual summary into a compilation of notes that already exist rather than a reconstruction of the year.
Four: what did you deliver this month, what got blocked or took longer than expected and why, what are the two or three priorities for next month, and what do you need from me. Five minutes each. The last question is the one managers most often skip and employees most often value.
Base the review on a record rather than on memory. If the team logs daily work with client, project, time and blockers, the month is in front of both people and neither has to recall week two. Reviewing monthly rather than annually also shortens the window that memory has to cover.
Yes. If every monthly conversation might affect pay, employees manage it defensively and honest discussion stops. Keep monthly reviews about the work and hold compensation to its own cycle — while still using the accumulated monthly notes as the primary input to increment decisions.
Three or four lines: what was delivered, what needs to change, and what was agreed for next month. That is enough to make the annual summary a compilation, to give the increment cycle a documented basis, and to have a contemporaneous record if an issue ever needs formal handling.
AI can draft a summary from a month of logged work, saving a manager assembly time, provided a human edits and stands behind it. It should not produce assessments from nothing, and it should never make the decision — ratings, increments, promotions and exits are judgements a person must be accountable for.
Merik supports monthly performance reviews that sit next to the data they describe: the employee's task log and attendance for that month are in the same workspace, so a review starts from what was recorded rather than from what anyone recalls. Reviews are stored per month, which is what turns the annual summary into a compilation.
An AI draft is available and entirely optional. It is generated from the employee's own task log and attendance, it is off unless your workspace enables it, and it produces a draft a human always edits — the edit distance between the draft and what the manager actually submits is captured as a signal that the drafts are being reviewed rather than rubber-stamped. See the performance module, the feature list, or how it works.
Track outcomes, not screen time, and keep your team's trust.
How to structure revisions, effective dates and arrears cleanly.
Where a language model earns its place — and where it must not go.