Review season usually runs on a document. A template gets emailed to every manager, comes back in fourteen slightly different versions, three of them saved over the previous year's file, and two never come back at all. The following January somebody needs to see what was agreed with a particular employee twelve months ago, and the answer depends on which folder a manager who has since left happened to use.
The questions on the form are the part everyone argues about. The records are the part that causes the actual damage, because a review that cannot be produced later is a review that did not happen as far as anyone outside the room is concerned. This guide covers both, starting with what the form is for, because that determines everything else about it.
The form has two jobs and they conflict
An employee review form is asked to do two things that pull against each other.
The first is to support a conversation. Here the form is a prompt sheet: it gets the manager and the employee to cover the same ground, stops the discussion becoming an argument about one recent incident, and leaves both with a written summary of what was agreed. For this job, open questions and plain language work best, and a rating is largely irrelevant.
The second is to produce a record. Ratings that can be compared across people and across years, evidence attached to conclusions, and a signed acknowledgement that the conversation happened. This job needs structure, consistency and storage that outlives the manager.
A form built only for the conversation leaves nothing usable a year later. A form built only for the record produces a scoring exercise that both parties resent and that gets filled in the night before. The workable answer is one form with the two parts visibly separated: a narrative section that belongs to the conversation, and a short structured section that belongs to the record.
Deciding which of the two matters more in a given organisation is a genuine choice, not a compromise to be avoided. A ten person company that will never compare ratings across teams should lean heavily towards the conversation. A company where pay decisions have to be defensible needs the record, and should say so on the form rather than pretending otherwise.
The question set
Six parts, in this order. The order is not decoration: reviewing the previous cycle before rating anything is what stops the whole exercise becoming a judgement of the last three weeks.
What was agreed last time, and what happened. List the goals from the previous cycle as they were written, and ask for the outcome of each. This is the single most valuable section and the one most often missing, because it requires that last year's form can be found.
Self assessment. The employee answers first, in their own words, before seeing any rating. What went well, what did not, what they needed and did not get.
Manager assessment against the same points. Written after reading the self assessment, and covering the same ground rather than a separate list of topics. Where the two disagree, the disagreement is the most useful thing in the document and should be recorded rather than smoothed over.
Ratings, if ratings are used. Kept short. Four to six items, each tied to something observable rather than to a personality trait. "Delivered agreed work to the dates committed" can be evidenced. "Shows initiative" cannot, and turns into a rating of how much the manager likes the person.
Goals for the next cycle. Written jointly, with dates, and few enough to be remembered. Three is usually the right number. Eight goals is a list nobody will read again until it is used as evidence.
Support needed. What the employee needs from the organisation to hit those goals: training, tools, a decision from someone else, time. This section is where a review stops being a one way assessment, and it is the first one cut when a form gets shortened.
Ask for evidence next to every rating
Put a short text field immediately beside each rated item, asking what it is based on. Two sentences is enough. This changes the character of the whole form.
A rating with no evidence beside it is an opinion recorded in a number, and it is the thing that falls apart when a pay decision or a dismissal is later questioned. A rating with one specific example beside it is a finding. The field also slows the manager down at exactly the right moment, which is the cheapest available defence against rating everyone at 4 because the deadline was yesterday.
The same field solves a smaller problem. Employees rarely dispute the number itself. They dispute the absence of a reason for it, and a form that requires a reason removes most of that argument before the meeting starts.
The question managers skip
Add one direct question about the manager: what would make the reporting relationship work better. It gets thin answers in year one and better answers later, once it becomes clear that answering it is safe. Without it, a review form asks an employee to be assessed by a process that never assesses itself.
Choosing a rating scale
| Approach | What it produces | Where it breaks |
|---|---|---|
| Three point | Fast, hard to game, clear | Almost everyone lands in the middle |
| Five point | Enough spread for pay decisions | Managers drift upward year on year |
| Behaviourally anchored, with written descriptions per point | The most consistent ratings across managers | Expensive to write and long to fill in |
| Narrative only, no rating | Honest, readable, good conversations | Cannot be compared or aggregated |
| Forced distribution across a team | Removes rating inflation | Damages trust quickly and is widely abandoned |
For most organisations a five point scale with a written description of each point, applied to four to six observable items, is the practical middle. The written descriptions are the part that gets dropped and they are the part that makes ratings mean the same thing in two different teams.
Whatever is chosen, two rules hold. Keep the scale identical between cycles, because a change in the middle makes the year on year comparison meaningless and the comparison is most of why ratings exist. And record the scale wording alongside each completed review, so a rating of 3 read three years later can be interpreted using the definitions that were in force at the time rather than the current ones.
Keep the record straight
This is where document based reviews fail, and the failures are mundane.
One version, not fourteen. A form that exists as a file emailed around will diverge, because managers add rows, delete sections and rename it. A form that exists as one online form cannot diverge, and that alone removes most of the reconciliation work at the end of a cycle.
The previous cycle has to be retrievable. Reviews for the same person across years should join up into one history rather than sitting as unrelated files. Without that, the first section of the form, the review of what was agreed, cannot be completed honestly.
Sign off has to be recorded as an event. What the employee acknowledged, when, and what the document said at that moment. Acknowledgement of a document that has since been edited is not evidence of anything, so the version acknowledged needs to be the version stored.
Access has to be decided rather than assumed. A review contains parts the employee wrote, parts the manager wrote, and often notes that are not shared with the employee at all. Those need to be separate fields with different visibility, not one text box with a convention about what goes in it. Conventions leak.
What is kept, and for how long, is a legal question. Retention periods, what has to be produced on request, and what an employee is entitled to see vary by jurisdiction and by the kind of employment. The list above is about the mechanics of not losing things. Whoever handles employment matters should confirm the retention rules before the first cycle runs.
The practical consequence of all five points is that a review process needs a form with structured storage behind it rather than a template plus a shared drive. Each submission needs an owner, a status, a history of changes, and fields that only some people can see. Looking at what the tools after the form actually do is the relevant comparison, because producing a pretty form is not the difficult part of running a review cycle.
One more record that gets forgotten: what was agreed in the meeting itself, as distinct from what the form said beforehand. Reviews change in the room, and a document that captures only the pre meeting positions will not match what either party remembers. Add a short field completed after the conversation, with the date it was held, and treat that as the authoritative version.
Running the cycle without chasing
The administrative load of a review cycle is almost entirely chasing, and it is proportional to how invisible the state of each review is.
Give every review a status: not started, self assessment done, manager section done, meeting held, acknowledged. With that visible in one list, the chasing becomes a filter rather than a series of individual questions, and the person coordinating the cycle can see at a glance which nine of sixty are stuck and at which step.
Set the sequence so the self assessment has its own deadline, several days before the manager's. Sent as one deadline, the two collapse into one evening and the self assessment stops influencing anything, which was its purpose.
Send reminders from the same place the reviews live, so what was sent to whom is on the record. Chasing from a personal inbox means nobody else can tell whether a late review has been chased twice or not at all, which is how a cycle ends with four reviews quietly missing.
Then export the structured part at the end, once, as a single file. If the ratings live in sixty separate documents, the summary for whoever makes pay decisions gets assembled by hand under time pressure, and that is where errors enter the numbers that matter most. The questions people ask cover how internal fields appear as columns in an export, which is the mechanism that makes this a single step.
What to change first
Move the form itself off a document and onto one online form, because that removes version drift, missing submissions and most of the chasing in a single change. Then add the section that reviews what was agreed last cycle, which is the part that makes the whole exercise cumulative rather than annual theatre. When the cycle is being coordinated by hand across more than about twenty people, see what one list with a status per review looks like in Halict.
Q1. What should an employee review form include?
Six parts: the outcome of goals agreed last cycle, a self assessment written first, a manager assessment covering the same points, a short set of ratings tied to observable work, goals for the next cycle with dates, and what support the employee needs. A question about what would improve the reporting relationship is worth adding.
Q2. Should the employee fill in a self assessment before the manager writes anything?
Yes, with its own earlier deadline. Written afterwards, or at the same time, it becomes a response to the manager's view rather than an independent account, and the disagreements between the two are the most useful content in the document. A separate deadline is what keeps the sequence intact.
Q3. What rating scale works best for performance reviews?
A five point scale with a written description of each point, applied to four to six observable items, is the practical default. The written descriptions are what make a rating mean the same thing in two different teams. Keep the scale identical between cycles and store the wording with each completed review so old ratings stay interpretable.
Q4. How long should completed review forms be kept?
Retention periods and what an employee is entitled to see vary by jurisdiction and by the kind of employment, so confirm the rules with whoever handles employment matters before the first cycle. What is within your control is that reviews for the same person join up across years, that the version acknowledged is the version stored, and that nothing depends on a departed manager's folder.
Q5. Can one form be used for every role in the company?
The structure can be shared while the rated items differ, which is the arrangement that keeps exports comparable. Duplicating one form and editing the items for each role family keeps the sections, the scale and the column order identical, so the results still assemble into one file. Separate forms written from scratch will not.