Audit your review form

Creates a report on how useful last cycle's review answers were, plus a rewritten form that can be used in the next cycle.

This skill audits your performance or 360 review forms using the current form and a CSV of last cycle's answers. It measures how useful the answers were, question by question, and gives you a rewritten form in a table: past question on the left, new question on the right. The report is written in plain language for an HR or People lead.

When to use it

  • People say your reviews are useless and you can't point to why.
  • Peer feedback comes back as "great colleague, nothing to improve".
  • Managers give the same score for every competency.
  • The form has grown past 15 questions and answers get shorter toward the end.
  • A new cycle starts soon and you want better questions, built from last cycle and not from scratch.

What you get

A shareable document with:

  1. Summary: three or four lines on what's wrong and the three changes.
  2. What last cycle's data shows: numbered findings, each with its number and what it means.
  3. Peer and upward: new questions, one table per form.
  4. Self and manager: new questions, one table per form.
  5. Next steps: three or four actions, including how to check next cycle.

In each table, the past question is kept verbatim, with its description and scale labels, so you see the contrast. Every past question is accounted for: rewritten, merged, or removed with a reason. New forms have at most 15 questions, and peer and upward forms aim for 8 to 10.

How it works

  1. It collects the inputs. The current forms with full question text, descriptions and scale labels; last cycle's answers as a CSV; and context such as headcount, review types and answer language. It checks your connected review tool and drives first.
  2. It lists every question set. One per distinct form, and it flags template problems such as an "explain your rating" box repeated under every competency.
  3. It analyses the answers. A script computes, per review type: how the rating scale is used, the share of forms with one score for every competency, the share with the top score on everything, and for each open question the median length, the share of non-answers, and the share with a concrete example. Then it reads samples itself, because numbers say how long answers are, not whether they help.
  4. It shows you the statistics before rewriting anything.
  5. It diagnoses the wording. It checks each question against known patterns: one rating covering four to six behaviours, peers asked to rate on performance labels, no "Can't assess" option, overlapping self-review questions, no anchor to the period's goals.
  6. It rewrites. One observable behaviour per rating. A frequency scale with "Can't assess" for peer and upward forms. Every open question asks for a real situation. Competency names are kept where possible so results compare year over year, and it says when a change breaks the trend.

How it reads the numbers. It compares your results with reference ranges from five companies' anonymised review exports, and says they are a guide, since roles, languages and company size differ. It tests an explanation before writing it: for example, whether blanket top scores actually rise when the reviewer group is small.

What it never does. It doesn't name an employee or reviewer, or quote a named person's answer. It recomputes every number it quotes. It suggests platform features only as "if your tool supports it". Without a CSV it says it can only review the wording, and skips the data section.

How to run it

Install the Effy AI plugin for Claude or ChatGPT. One install adds every Effy AI skill, this one included. Then connect the tools you want the skill to read from, such as email, calendar, your drive or your meeting notetaker, and ask for what you need in plain words: the assistant picks the right skill.

Before you start: export last cycle's answers as a CSV, with all review types in one file or one file per type. Anonymised reviewer names are fine. Remove names if your AI policy requires it.

  1. Attach the current form and the CSV, and ask, for example: "Effy, use your skill to build our review form."
  2. Read the statistics and react.
  3. Review the report and decide on the open choices it lists.

Next in the cycle: "Draft a performance review" helps each participant write with evidence, and "Analyze a performance review cycle" prepares calibration.

Click. Try. Use.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.