LensReading which lens this session carries.

Fairness readings computed on modeled data demonstrate the instrument, not a real audit result.

Govern · people-affecting decision · HR9A

Attrition risk flagging

Flags employees at elevated risk of leaving so that a retention conversation can happen.

All decision points

Who is accountable, and what the oversight actually is

Named, with the measure written out rather than implied by a role title.

Agent
Attrition Signal Agent
Autonomy: A2
Accountable human
HRBP for the population
HR business partner
Oversight measure
Flags are visible to the HR business partner only, never to the line manager, and expire after 60 days.
Review cadence
Quarterly
Reviewer: Priya Raman
51 days overdue
Monitoring status
Review overdue
Readings continue but the documented review cadence has been missed.

Regulatory position

How this decision is classified and what the affected person can see.

Risk class
High risk — EU AI Act Annex III(4)(b) and GDPR Article 22
Employment profiling. Predicting who will leave is profiling of workers, and the flag is capable of affecting how a person is treated.
Explanation surface
On request
An explanation is produced when the affected person asks for it.
How the outcome is contested
Priya Raman, Chief People Officer
Service level: 240 hours to a human response
Full risk record for this agent

Selection rate by cohort, by period

Impact ratio against the reference cohort.

CohortReferenceMar 2026Apr 2026May 2026Jun 2026Jul 2026Aug 2026
Women
Gender
Men
1.06
15.0% · n=1,690
1.08
15.4% · n=1,690
1.09
15.1% · n=1,690
1.11
15.6% · n=1,690
1.13
15.4% · n=1,690
1.15
15.8% · n=1,690
40 to 49
age-band
30 to 39
0.95
13.8% · n=883
0.94
12.7% · n=883
0.95
13.5% · n=883
0.94
13.5% · n=883
0.93
13.0% · n=883
0.94
13.3% · n=883
50 and over
age-band
30 to 39
0.88
11.9% · n=538
0.89
12.1% · n=538
0.87
12.4% · n=538
0.88
12.7% · n=538
0.86
12.0% · n=538
0.87
12.3% · n=538
Under 30
age-band
30 to 39
1.18
16.7% · n=1,037
1.19
16.7% · n=1,037
1.21
17.5% · n=1,037
1.22
17.5% · n=1,037
1.24
16.9% · n=1,037
1.26
17.1% · n=1,037
Over 10 years
tenure-band
2 to 5 years
0.79
11.3% · n=653
0.78
10.6% · n=653
0.77
10.8% · n=653
0.76
11.0% · n=653
0.75
10.3% · n=653
0.74
10.5% · n=653
Under 2 years
tenure-band
2 to 5 years
1.22
17.3% · n=1,306
1.24
17.7% · n=1,306
1.25
17.3% · n=1,306
1.27
17.8% · n=1,306
1.29
17.6% · n=1,306
1.31
18.0% · n=1,306
Within thresholdMoving but not yet outsideOutside the disparity thresholdSample below the reporting floor· means no reading exists for that period
Cohorts declared for monitoring that never produced a single reading
Gender · Non-binary and self-described — No reading has ever been produced for Non-binary and self-described on this decision. The cohort is declared in scope and the population exists, but it has never once been large enough to report against Men at the confidence this program requires.

Alarms on this decision point

Each with the action recorded against it and the person who owns it.

FAIR-0013Sustained driftMedium
170 days openAcknowledged
The ratio for Women has moved from 1.06 to 1.15 across 6 periods without ever breaching.
Action: Extrapolate the trend to the threshold, put a date on it, and bring that date to the next review rather than waiting for the breach.
Owner: Nadia Kovac · Data Protection Officer
FAIR-0014Never reportableLow
170 days openMonitoring
No reading has ever been produced for Non-binary and self-described on this decision.
Action: Either aggregate this cohort across decisions and periods so a reading becomes possible, or state on the record that this cohort is not monitored here and why. Leaving the cell empty is the one option that is not defensible.
Owner: Nadia Kovac · Data Protection Officer
FAIR-0015Threshold breachHigh
17 days openOpen
Under 30 is being selected at 126 percent of the rate for 30 to 39. For this decision selection is the adverse event, so anything above 125 percent breaches.
Action: Hold the decision at the current autonomy level, re-run the cohort with the ranking features ablated one at a time to find what is carrying the disparity, and record the finding against this decision point before the next cycle opens.
Owner: Priya Raman · Chief People Officer
FAIR-0016Threshold breachHigh
78 days openOpen
Under 2 years is being selected at 131 percent of the rate for 2 to 5 years.
Action: Hold the decision at the current autonomy level, re-run the cohort with the ranking features ablated one at a time to find what is carrying the disparity, and record the finding against this decision point before the next cycle opens.
Owner: Priya Raman · Chief People Officer
FAIR-0022Review overdueMedium
51 days openAcknowledged
The documented review cadence for this decision is quarterly.
Action: Hold the review, minute it, and record the departure so the cadence claim on this page stays true.
Owner: Priya Raman · Named reviewer for this decision

Appeals against this decision

Every contest recorded, with who reviewed it, how long they spent and what changed.

RefSubjectGroundsRouted toService levelHuman timeOutcomeEvidenceStatus
APL-0007
EMP-12008
employee · US
Employee learned of a retention flag during a conversation with their business partner and objects to the profiling.
Priya Raman
Chief People Officer
96 / 240 hr52 minOverturned
Flag withdrawn and the employee opted out of future flagging.
HR9A-1012#1Closed
How to read this page

The impact ratio is the cohort’s selection rate divided by the reference cohort’s. For a decision where being selected is the benefit, the threshold is four-fifths and a ratio below it is the alarm. For this decision, where being selected is the harm, the test is inverted and a ratio above 1.25 is the alarm. Both are stated on the reading rather than left to the reader.

A reading is only computed when the cohort and the reference cohort each carry at least thirty people and at least five events.

This decision runs in US, SG, AU. Local law shapes what may be measured: where a works council agreement or a local statute restricts cohort analysis, the exclusion is recorded in the monitoring status rather than silently applied.