Provisioning Entitlement Utilization (PEU)

GMI-IT-030IT OperationsPilot-Ready

Definition

Provisioning Entitlement Utilization (PEU) is a proposed measure of what percentage of AI-provisioned access rights new hires actually used within the first 90 days: quantifying the gap between what peer-based provisioning delivers and what the role actually requires. The calculation is straightforward: count the entitlements the AI provisioned for a new hire, then count how many of them the hire actually accessed within 90 days. Divide. The resulting percentage tells you how much of the initial access grant was necessary versus how much was deadweight from the start. The formula, in three layers:

The Measurement Gap

The new hire showed up on her first day and was provisioned with forty-seven entitlements. The IAM system had done exactly what it was designed to do: analyzed her peer group (eight other people in the Senior Data Engineer role who had been in that role between two and six years) and provisioned access based on what they had. The logic was sound. The result was not.

Her peers had accumulated entitlements across multiple projects, some of which were years past completion. One peer had production database access from a migration project eighteen months ago that nobody ever revoked. Another had write access to a deprecated data pipeline that still existed in the infrastructure. A third had read access to three Salesforce reporting environments from before the company's CRM consolidation. Each of those entitlements had survived annual certification reviews: individually, they looked low-risk; the reviewers recognized the system names and approved without question. Collectively, they represented a blast radius far larger than any Senior Data Engineer actually needed on Day 1.

The new hire's provisioning reflected the aggregate of those residual entitlements. She started with access to systems she wouldn't learn about for months, wouldn't need for her current projects, and that nobody explained to her. From a security posture standpoint, her account was over-provisioned from the moment it was created. If her credentials were compromised in her first week on an otherwise unremarkable phishing attempt, the attacker had the full set, not just what was necessary for the job she was hired to do.

The legacy metrics the IAM team tracked didn't see any of this. Provisioning time was the primary operational metric: how quickly did she get her access? Four hours: well within SLA. Access certification completion rate tracked whether quarterly reviews were finished on time: they were. Orphaned account count tracked whether terminated employees' accounts were cleaned up: irrelevant here, she had just started. MFA adoption tracked whether she was enrolled in multi-factor authentication: she enrolled on Day 1.

Not one of those metrics measured what the peer-provisioning model had actually given her versus what she needed. The structural problem is that legacy IAM metrics measure process execution (did provisioning happen fast enough, did reviews complete, did cleanup happen) and were designed before AI started making provisioning recommendations. When a human IAM admin provisioned a new hire, they made judgment calls. They knew the role, asked the hiring manager, provisioned a baseline set, and let the hire request additional access as needed. When AI provisions based on peer analysis, those judgment calls are replaced with statistical inference from the peer group's current entitlement set. If the peers have accumulated years of residual access, the new hire gets it from Day 1.

There is no metric that tells you whether what the AI provisioned matched what the hire actually used. That gap (between entitlements granted and entitlements needed) is what Provisioning Entitlement Utilization is designed to quantify.

Formula

Provisioning Entitlement Utilization (PEU) is a proposed measure of what percentage of AI-provisioned access rights new hires actually used within the first 90 days: quantifying the gap between what peer-based provisioning delivers and what the role actually requires. The calculation is straightforward: count the entitlements the AI provisioned for a new hire, then count how many of them the hire actually accessed within 90 days. Divide. The resulting percentage tells you how much of the initial access grant was necessary versus how much was deadweight from the start.

The formula, in three layers:

Plain language: For each user provisioned with AI-recommended entitlements during the measurement period, calculate what percentage of their provisioned entitlements they actually accessed at least once in the first 90 days. Aggregate across all provisioned users in the period.

Notation:PEU = Σ E_used,i / Σ E_provisioned,i

Where:

  • E_provisioned,i = total entitlements AI-provisioned for user i during onboarding or role change
  • E_used,i = entitlements user i actually accessed (at least one confirmed access event) within 90 days of provisioning date
  • Summation runs over all AI-provisioned users in the measurement periodWorked Example The IAM team provisioned 20 new hires in Q1, each receiving an average of 35 AI-recommended entitlements: 700 total provisioned entitlements across the cohort. In the 90 days following provisioning, those 20 users accessed an average of 18 of their entitlements each: 360 confirmed access events out of 700 provisioned. PEU = 360 / 700 = 51.4%. More than half of provisioned access went unused in the first 90 days. That's the peer model's entitlement inflation, made visible.

Sign convention:

Higher PEU is better. A PEU of 100% would mean every access grant was needed and used: unlikely in practice, since some legitimate entitlements are held for contingency (emergency access, audit tooling, break-glass accounts). A realistic well-performing system might target 70–80%. Low PEU means the peer model is inflating access from Day 1.

Edge Cases

Users who leave the organization within 90 days should be excluded from the cohort: their access utilization will be artificially low regardless of provisioning quality, and including them distorts the metric. Seasonal entitlements are the main edge case requiring judgment. A financial analyst's access to the year-end reporting environment won't be used if she starts in February. Track known seasonal entitlements separately with an extended window (full fiscal year) rather than letting them contaminate the 90-day PEU calculation. Maintain a documented list of seasonal entitlements by role that gets reviewed annually. Mandatory compliance entitlements (access every employee must have regardless of role (payroll portal, benefits system, corporate VPN, security awareness training)) should be excluded from the PEU calculation entirely. You're measuring the quality of role-specific peer-based provisioning decisions, not the baseline access that comes with employment. For small cohorts (fewer than 5 provisioned users in the measurement period) PEU will be sensitive to individual outliers. Use rolling 90-day provisioning cohorts to maintain sample sizes above 10 before drawing conclusions.

Cross-references:

PEU is a foundational IAM metric: it measures the starting condition from which all subsequent access accumulation problems emerge. There's no upstream KPI in this framework that PEU depends on; the peer model's output quality is the direct measurement target. Downstream, PEU feeds directly into Certification Auto-Approval Accuracy (CAAA, GMI-IT-028). The entitlements that show up as unused in the PEU calculation are the same entitlements that will appear in the next certification campaign. If the AI certification tool auto-approves them again (because they look low-risk by usage profile) the access accumulates into the next generation of peer group data. Low PEU now predicts certification quality problems in six months. Further downstream, low PEU is the measurable origin of entitlement creep: the legacy metric that compliance teams have tracked for years without being able to attribute it to a specific cause. The mechanism is now traceable: peer-provisioned access that goes unused doesn't get proactively removed, survives certification, accumulates through role changes, and becomes the peer baseline for the next new hire. PEU quantifies how much of that accumulation originates at Day 1.PEU is classified as Pilot-Ready. The IGA platform logs what was provisioned; the SIEM or application access logs confirm what was used. The join between them is real data engineering work, but both sources exist in virtually every mature IAM deployment. You don't need new infrastructure or new tooling to compute PEU.

Proposed formula, pending validation pilots.

At a Glance

Category
AI-Era IT Operations
← Procurement Forecast Variance (PFV)PROXY DISCRIMINATION DETECTION GAP (PDDG) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment