Client Feedback and AI Calibration

3 min read

Every piece of client feedback teaches the AI what "a good candidate" means for that client. After enough signal, the AI proposes a concrete search refinement — you review and apply it, and the search gets sharper with every round.

How Clients Give Feedback

Clients rate candidates from two places, both one click:

In the portal — Excellent Fit / Good Fit / Not Right buttons on every candidate, with optional notes and a structured rejection reason (too junior, skills gap, salary too high, and so on)
In the submission email itself — the same feedback buttons are embedded, so a client who never opens the portal still feeds the loop

Every rating, note, and rejection reason is recorded against the role.

When the AI Proposes a Refinement

Once a role has 3 or more feedback items, the AI looks for the pattern behind the ratings and generates a calibration insight with three parts:

Part	Example
Pattern	"Strong matches all come from smaller companies where candidates owned projects end-to-end"
Suggestion	"Deprioritise large-enterprise backgrounds for this role"
Proposed change	Specific text to add to the search criteria

Structured rejection reasons are weighted heavily — three "too junior" rejections say more than three unexplained "Not Right" clicks.

Reviewing and Applying an Insight

The insight appears on the role for your review. If you apply it:

The proposed change is added to the search criteria
Sourcing filters are regenerated to match the refined brief
Screening questions are regenerated so the next candidates are vetted against what the client actually wants

You can also dismiss an insight — you know things about the client the feedback doesn't show.

Auto-Apply on Unambiguous Patterns

One exception to review-first: if the last 3+ rejections all share the same rejection category (for example, three "too junior" in a row), the pattern is treated as unambiguous and the insight is applied automatically — the search stops wasting everyone's time immediately. You're notified when this happens and can review or reverse the change.

The Bias Flag

Before any insight is saved, the AI checks whether the feedback pattern might be a proxy for a protected characteristic — for example, a pattern that tracks age, gender, or ethnicity rather than skills.

If the check trips:

The insight is not applied — not even suggested for one-click apply
Your agency admin is notified: "Calibration insight blocked — potential bias detected", with the AI's explanation of what looked wrong
The recruiter decides. A flagged pattern isn't an accusation — sometimes there's a legitimate, job-related explanation. But the call is made by a human, with eyes open, not silently baked into the search algorithm

This is part of PlacementFlow's auditable-AI posture: the calibration loop can't quietly learn a discriminatory preference from click patterns.

What the AI does here

Aggregates ratings, notes, and rejection reasons into a single pattern per role
Proposes one concrete, specific search change — not a vague "improve targeting"
Regenerates sourcing filters and screening questions when a change is applied
Always requires you for: applying insights (except unambiguous same-category streaks, which you're notified about) — and any bias-flagged pattern is always a human decision

Back to Client Portal

Client Feedback and AI Calibration

3 min read

How Clients Give Feedback

Clients rate candidates from two places, both one click:

In the portal — Excellent Fit / Good Fit / Not Right buttons on every candidate, with optional notes and a structured rejection reason (too junior, skills gap, salary too high, and so on)
In the submission email itself — the same feedback buttons are embedded, so a client who never opens the portal still feeds the loop

Every rating, note, and rejection reason is recorded against the role.

When the AI Proposes a Refinement

Once a role has 3 or more feedback items, the AI looks for the pattern behind the ratings and generates a calibration insight with three parts:

Part	Example
Pattern	"Strong matches all come from smaller companies where candidates owned projects end-to-end"
Suggestion	"Deprioritise large-enterprise backgrounds for this role"
Proposed change	Specific text to add to the search criteria

Structured rejection reasons are weighted heavily — three "too junior" rejections say more than three unexplained "Not Right" clicks.

Reviewing and Applying an Insight

The insight appears on the role for your review. If you apply it:

The proposed change is added to the search criteria
Sourcing filters are regenerated to match the refined brief
Screening questions are regenerated so the next candidates are vetted against what the client actually wants

You can also dismiss an insight — you know things about the client the feedback doesn't show.

Auto-Apply on Unambiguous Patterns

The Bias Flag

If the check trips:

The insight is not applied — not even suggested for one-click apply
Your agency admin is notified: "Calibration insight blocked — potential bias detected", with the AI's explanation of what looked wrong
The recruiter decides. A flagged pattern isn't an accusation — sometimes there's a legitimate, job-related explanation. But the call is made by a human, with eyes open, not silently baked into the search algorithm

This is part of PlacementFlow's auditable-AI posture: the calibration loop can't quietly learn a discriminatory preference from click patterns.

What the AI does here

Aggregates ratings, notes, and rejection reasons into a single pattern per role
Proposes one concrete, specific search change — not a vague "improve targeting"
Regenerates sourcing filters and screening questions when a change is applied
Always requires you for: applying insights (except unambiguous same-category streaks, which you're notified about) — and any bias-flagged pattern is always a human decision

Back to Client Portal