The complete guide to following up on performance reviews

Why review follow-through usually fails

Most performance review processes end at compilation. A manager gathers feedback, writes a summary, delivers it, and the cycle closes. The next cycle starts from a blank page — new feedback, no reference back to what the last review actually asked the person to change.

That gap isn’t a discipline problem. It’s a data problem: last cycle’s feedback lives in a document (or a form submission) that nobody has a reason to reopen once the review is delivered. Even a conscientious manager who remembers the gist can’t point to what was actually said, so the next conversation restarts from impression rather than evidence.

What "follow-through" actually means

Follow-through is two things, not one: turning a cycle’s feedback into a small number of specific, checkable points, and then actually checking each one against the next cycle’s feedback before it’s forgotten.

The first half is where most processes stop. A review that ends with "focus on stakeholder communication" isn’t a point — it’s a theme. A point is checkable: it names the behavior, the evidence for it, and what changing it would look like. Themes make people nod; points make the next review’s question obvious.

Turning feedback into concrete points

Two inputs turn a page of prose into a short list of real points: the feedback text itself, and the rating trend per competency category across cycles. A category that’s been flat or falling for two cycles running is a different signal than one that dipped once and recovered — the first is worth a named point, the second usually isn’t.

A work profile (a template of competency categories, each scored per cycle) gives that trend a stable axis to measure against. Without it, "getting better" or "getting worse" is just a feeling reconstructed from memory each time.

• Name the behavior, not the trait — "missed two sprint estimates by more than 30%" not "poor planning"
• Cap it at three to four points per person per cycle — more than that and none of them get tracked
• Tie each point to the category it came from, so next cycle’s rating on that category is the evidence, not a fresh guess

Tracking whether it happened

A point only earns its place if something checks it later. That means: at the start of the next cycle, every point still open from a prior cycle gets evaluated against this cycle’s feedback and rating for that category, and lands in one of a few honest states — matched, partially matched, not matched, regressed, or not yet due if the next check-in hasn’t arrived.

That evaluation is what separates a follow-through process from a good-intentions process. "We talked about it" and "the category rating moved from a 2 to a 4 over two cycles" are different claims, and only one of them survives a skeptical manager asking for evidence.

Where automation actually helps — and where it doesn’t

Reading a cycle’s worth of feedback and drafting three specific points from it is exactly the kind of first-pass work an NLP summarizer is good at — it’s slow and easy to skip by hand, cheap to get a draft of by machine, and still needs a human to keep or discard each draft.

Deciding whether a point was actually met is a call that should stay reviewable: a "matched" status a manager can’t see the reasoning behind isn’t worth more than the paragraph it replaced. The automation that’s worth having drafts the check and shows its work, rather than replacing the check with a black-box status.

The next level up from checking once is not stopping there: drafting the actual follow-up conversation — a review cycle scoped specifically to the points still open — so re-checking isn’t a separate thing someone has to remember to schedule.

A simple template to start with

You don’t need software to start. One row per person, one column per open point, and four columns of state: point, evidence it was raised from, cycle it was raised in, and current status (open / partially met / met / regressed). Review that sheet at the start of every cycle, before writing anything new.

The moment that sheet gets past a handful of people, the manual version breaks down in the same two places every time: nobody rewrites vague feedback into specific points fast enough to keep up, and nobody remembers to reopen last cycle’s sheet before this cycle’s reviews are already half-written. That’s the specific gap tools built for this close — not the tracking spreadsheet itself, the two habits around it.