HomeIntelligenceEarly Warning Engine

Early Warning Engine

Detect Assignment Failure
6–12 Weeks Before the Phone Call Comes

Ten validated risk signal patterns. A prioritised intervention queue. Specific action scripts for every flag type. Not a risk score to interpret — a named pattern the mobility manager immediately recognises, with a specific action to take.

See a Live Demo → See Population Dashboard

The Core Design Principle

Named Patterns. Not Risk Scores.

A risk score is a number that the manager must interpret. A named pattern is a situation they immediately recognise — and each one comes with a specific recommended action. The Early Warning Engine doesn't produce dashboards to monitor. It produces an intervention queue to act on.

PatternWhat It DetectsTypical TimingSignal CombinationDefault Intervention
EW-001
Post-Honeymoon Withdrawal
Engagement drops sharply after high early activity — culture shock onsetWeeks 6–12Eng ↓ + trend: decliningActivate social mission chain
EW-002
Partner Isolation Risk
Partner not enrolled after week 6 AND family domain below thresholdWeek 6+partner_enrolled=false + family < 0.45Send partner enrollment invitation
EW-003
Culture Shock Trough
Sentiment and engagement declining simultaneously below thresholdsWeeks 8–14sentiment < -0.3 + eng < 0.5Direct mobility manager outreach
EW-004
Social Stagnation
Social domain flat below 0.35 despite ongoing engagement — destination-specificWeek 8+social < 0.35 + velocity < 0.01Activate destination social missions
EW-005
Pre-Repatriation Flight Risk
Declining signals in the 90 days before return date90 days before returneng ↓ + sentiment ↓ + days_until_return < 90Early repatriation module activation
EW-006
Family Adjustment Lag
Family flat below 0.4 while Workplace above 0.65 — hidden household failureWeek 10+family < 0.4 + workplace > 0.65Partner enrollment + family missions
EW-007
Mission Disengagement
Zero missions completed in 3+ weeks when 3 were expectedAny phasemissions_completed_4w = 0Personalised re-engagement outreach
EW-008
Partner Distress Signal
Partner enrolled but engagement very low AND sentiment decliningAny phasepartner_eng < 0.25 + partner_sentiment ↓Direct outreach to employee
EW-009
Assignment Completion Risk
Week 16+ with overall progress below 35% AND negative sentimentWeek 16+week > 16 + progress < 0.35 + sentiment < 0HR escalation + intensive support
EW-010
Repatriation Retention Risk
Both engagement and sentiment declining simultaneously during repatriation phaseRepatriation phasephase=repatriation + eng ↓ + sentiment ↓Retention coaching track activation

The Intervention Queue

Not Alerts to Acknowledge. Actions to Take.

Each risk flag generates one card in the mobility manager's intervention queue. The card shows the assignee, the pattern name, a 2–3 sentence plain-language explanation — written without numbers or data terminology — and a primary action button. The manager decides and acts. They do not interpret data.

  • "See what to say" expands a specific action script — email copy, talking points, or outreach message generated for this specific pattern and assignee profile
  • High-severity flags fire email alerts immediately — defined as patterns with >35% historical correlation with early returns
  • "Snooze 7 days" and "Dismiss with reason" prevent alert fatigue without silencing genuine risks
  • Resolved tab shows past flags and their 30-day outcomes — did the signals improve after the intervention?
  • Pattern Library lets managers see all 10 patterns and their current historical effectiveness rates
Intervention Queue — 2 Items
Marcus R. · Tokyo · Week 9
EW-001 · Post-Honeymoon Withdrawal
High
Marcus was consistently active through week 8, then engagement dropped sharply and hasn't recovered. This pattern is consistent with the post-honeymoon culture shock phase and typically precedes an early return request if untreated.
Sarah K. · Amsterdam · Week 8
EW-002 · Partner Isolation Risk
Medium
Partner not enrolled after 8 weeks. Family domain stagnating at 31% — below the 45% threshold expected at this stage of the assignment.

How the Patterns Get Better Over Time

Every Intervention Outcome Teaches the System.

When a mobility manager acts on an intervention, the platform tracks what happened 30 and 90 days later. Did the signals improve? Did the assignment complete? The Platform Intelligence Engine uses this outcome data to rank interventions by effectiveness, refine signal thresholds, and discover new patterns not yet in the seeded set of ten.

30 days
Outcome assessment runs and updates intervention effectiveness ranking for each pattern type
90 days
Secondary assessment: is the assignment on track, completed, or ended early?
50+
New outcome labels required to trigger a full model retraining run improving prediction accuracy
EW-011+
New patterns promoted from the AI pattern discovery layer when validated at >0.4 confidence

See the Early Warning Engine in Action.

We'll run a live early warning scenario for your destination mix — showing exactly which patterns would fire and what the intervention queue would recommend.

Book a Team Demo → See Pricing

FAQ

Common Questions

How does the system decide high vs. medium severity?
Severity is based on historical correlation between the pattern and early returns in the platform outcome dataset. Patterns with >35% correlation fire as high severity. Thresholds update as new outcome data accumulates — patterns that prove less predictive over time are down-ranked.
What happens if I dismiss a flag?
Dismissed flags are suppressed for 14 days. They re-fire if signals remain at threshold after the suppression window. Dismissal reasons are logged so the system can distinguish false positives from genuine dismissals in the effectiveness data.
Can I add custom risk patterns?
Custom patterns are available at the Team Intelligence tier. You define the signal conditions using the same field set as the seeded patterns; the platform evaluates them alongside the seeded set every 6 hours.
Does the system email me when a high-severity flag fires?
Yes. High-severity flags trigger an immediate email and in-app notification. Medium severity is batched into a daily digest. Alert preferences are fully configurable — including filtering alerts by destination, phase, or assignment type.
What is the "action script" that expands?
A generated message specific to the intervention type. For partner enrollment: a warm email invitation to the partner explaining what Bearings offers them. For direct outreach: a suggested message the mobility manager can send to the employee, framed as routine programme check-in. For manager prompts: 5 specific talking points for the next one-on-one.
Are there false positives?
Yes — the Early Warning Engine errs toward sensitivity rather than specificity. A dismissed flag is better than a missed early return. The effectiveness data tracks false positives separately, and pattern thresholds adjust accordingly over time.