Most performance problems don’t arrive suddenly. They build for months before anyone names them. By the time the hard conversation happens, the signals were already there; they just weren’t being tracked. A feedback system is the structure that makes those signals visible while there is still time to act.
The reframe from the previous article applies here: the annual review format surfaces problems too late. The lightweight feedback cadence from Why Small Business Performance Reviews Don’t Happen — and What to Do Instead creates the regular conversations. This article builds the system that captures what those conversations reveal.
A conversation that produces a note is different from a conversation that produces only a memory. The note is the thing that makes patterns visible. Without it, each conversation starts from scratch, and the accumulating picture of someone’s performance is never available at the moment it’s needed.
The Early-Warning Feedback System gives you the structure for capturing that picture consistently.
Why Terminations Surprise Both Parties
The obvious failure: the owner has known for months that something wasn’t working. They delayed the hard conversation because there was no comfortable moment for it. When the conversation finally happens, it feels like it came from nowhere to the team member who never received the feedback.
The less visible cost is what this does to the owner’s credibility. A team member who is terminated without a documented performance history will almost always believe the decision was unfair or sudden. When the record doesn’t exist, that belief is hard to contradict.
The deepest problem is that the absence of feedback removes the possibility of course correction. Someone who is underperforming and receiving clear, consistent feedback can often correct. Someone who is underperforming and receiving no feedback has no basis for understanding what needs to change. The termination becomes the first clear signal they’ve received, and by then it’s too late.
A feedback system doesn’t prevent terminations. It prevents the terminations that should have been avoided.
The Early-Warning Feedback System
The system has three elements. Each one turns a conversation into a data point that accumulates over time.
Element 1 captures the check-in notes
After every monthly check-in, write two to four sentences: what was discussed, how the person is tracking against expectations, and any signal worth noting.
This note doesn’t need to be a formal assessment. It needs to be specific enough to reconstruct the conversation later. “Told me the X project is going well; expressed concern about the Y workload; performance on output metric on track” is sufficient. “Good conversation” is not.
The check-in note is the earliest indicator in the system. Individual notes are often unremarkable. Patterns across six months are often revealing.
Element 2 tags the patterns
Every three to four check-ins, read the notes from the last quarter together. What has been consistent across the conversations? What has changed? What signal, positive or negative, has appeared more than once?
A team member who mentioned workload pressure in three of the last four check-ins is showing a signal. A team member whose notes have been uniformly strong for six months is showing a different signal. Neither is obvious from any individual conversation. Both are obvious from the accumulated record.
Tagging the pattern is a one-sentence observation added to the quarterly review note. “Consistent workload stress across the last four check-ins” is the kind of one-sentence observation that makes the record useful.
Element 3 creates the pre-conversation summary
Before any difficult performance conversation, read the full check-in record for the person. The pre-conversation summary covers three points: the pattern, the signals that support it, and where the record shows a gap from expectations.
This summary is not the script for the conversation. It is the preparation that ensures the conversation is grounded in the actual record rather than in the memory of a recent incident. A conversation grounded in six months of observation is different from one grounded in last week’s incident.
Before: The difficult conversation is based on a recent incident and the other party disputes the characterization. After: The difficult conversation is based on a documented pattern and the record is available to support it.
How the Living Library Tracks Your Feedback Cadence
Without a system, the feedback history for any team member exists in scattered notes and in memory. Neither holds up over time.
The Living Library is the part of the Kiluma platform that holds the business’s accumulated people-records. The Conductor, Kiluma’s context-aware AI, reads from the Performance Collection where the check-in notes and quarterly observations are stored.
When a concern starts to develop, the owner opens the platform and finds the feedback record already current. The check-in notes from the last six months are there. The quarterly pattern observations are there. The record shows not just that something seems off, but when the signal first appeared and how it has developed over time.
That reading changes when the hard conversation happens. Instead of starting from recent memory, the owner starts from a documented pattern that has been building for months. The conversation is harder to dispute and more likely to produce change.
Start Capturing After the Next Check-in
After the next monthly check-in, open a document and write three sentences about what was discussed and what it showed. Don’t overthink the format. Date it and save it to the person’s file in the Performance Collection.
Do that after every check-in, for every direct report. The system starts from the first note, not from the first crisis.
When the Record Exists Before the Problem Peaks
The owner who has six months of documented check-ins addresses performance from a sustained record. The owner who doesn’t is reacting to a recent incident. The feedback system is what creates the record. Try Kiluma free for 14 days at kiluma.ai.