Definition
TACI measures the frequency and severity of contradictory recommendations across AI systems acting on the same employee.
The Measurement Gap
Measurement Gap and Decision Impact
When an attrition model flags an employee as a flight risk while the performance system recommends them for a stretch assignment, or when the learning platform recommends upskilling in an area the skills platform already classifies as proficient: TACI detects and severity-weights these contradictions to produce an overall coherence score.
The Decision It Supports:
Are the organization's multiple AI talent systems working in concert or at cross-purposes? Most organizations have deployed AI across recruiting, skills management, learning, performance, compensation, and retention independently. TACI asks whether these independently deployed systems produce coherent recommendations when their outputs converge on the same employee.
Why It Requires a New Metric:
Legacy talent management operated through human coordination, a manager or HR business partner synthesized inputs from multiple sources and resolved contradictions through judgment. AI systems operate independently, each optimizing its own objective without awareness of what other systems are recommending for the same employee. No legacy metric measures cross-system coherence because the coherence function was embedded in human judgment, not measured as a system property.
The measurement gap TACI addresses is the absence of any mechanism to detect when multiple AI talent systems produce contradictory recommendations for the same employee, a failure mode that did not exist before organizations deployed AI across multiple HR functions independently.
Consider a concrete scenario. The attrition prediction model flags an employee as a high flight risk based on tenure, compensation trajectory, and peer departure signals. Simultaneously, the performance AI recommends the same employee for a high-visibility stretch assignment based on performance ratings and skill match. The learning platform recommends foundational training in a competency the skills platform already classifies as proficient. The compensation engine's equity analysis suggests the employee is above-market while the attrition model's compensation sensitivity analysis identifies below-market pay as a flight risk driver.
None of these systems know what the others are recommending. Each is individually defensible based on its own data and optimization objective. But the aggregate employee experience is incoherent: the organization simultaneously signals that it values the employee (stretch assignment), doubts their commitment (flight risk flag), questions their competence (foundational training recommendation), and considers them overpaid (equity analysis) while also worrying they will leave because they are underpaid (attrition driver).
This incoherence erodes employee trust in AI-driven processes and creates operational confusion for managers who receive contradictory signals from authoritative-looking AI outputs. More practically, it wastes organizational resources by triggering retention interventions for employees simultaneously flagged for development investment, or recommending training in skills already validated as proficient.
TACI makes cross-system incoherence visible by auditing the set of active AI recommendations for each employee with two or more active AI touchpoints. It identifies contradictions, classifies them by severity (minor, material, directly opposing), and computes a weighted coherence score. The result tells the organization whether its AI ecosystem is producing a coherent talent strategy or a collection of conflicting signals.
Without TACI, cross-system contradictions accumulate invisibly until they surface as employee confusion, manager frustration, or trust erosion in AI-driven processes.
Formula
Formula:
TACI = 1 − (Σ severity_j × contradiction_j) / (N_employees × Max_possible_contradictions)
Variable Definitions:
| Variable | Definition | Unit | Source |
|---|---|---|---|
| contradiction_j | Binary indicator for each contradiction type j identified in the cross-system audit | Binary (0 or 1) | Cross-system recommendation audit |
| severity_j | Impact weight for contradiction type j: 1 = minor inconsistency, 3 = material conflict, 5 = directly opposing recommendations | Integer (1, 3, or 5) | Defined in contradiction taxonomy |
| N_employees | Number of employees with two or more active AI recommendations being audited | Integer | Cross-system integration layer |
| Max_possible_contradictions | Theoretical maximum contradictions per employee multiplied by N_employees | Integer | Computed from system count |
Units: 0 to 1 ratio. 1.0 = perfect coherence (no contradictions). 0 = maximum incoherence.
All thresholds are proposed starting points for organizational calibration, not industry standards.
Edge Cases and Exclusions:
- Not all contradictions are errors: some represent naturally tensioned recommendations (develop a skill vs. build on an existing strength). Define a contradiction taxonomy that distinguishes truly incoherent recommendations from legitimate tensions.
- The cross-system audit mechanism most organizations do not have is the primary implementation barrier
- Begin with high-stakes employees (top performers, succession candidates, flight risks) where contradictions have the most impact
Data Requirements:
| Data Element | Source System | Classification |
|---|---|---|
| Attrition model predictions per employee | People analytics platform | NATIVE |
| Skills-based recommendations per employee | Skills/talent marketplace | NATIVE |
| Compensation recommendations per employee | Compensation platform | NATIVE |
| Learning recommendations per employee | LMS/LXP | NATIVE |
| Performance actions per employee | Performance system | NATIVE |
| Cross-system recommendation audit results | Custom integration layer | NOT CAPTURED |
Proposed formula, pending validation pilots.