Issued April 15, 2024 — six months post-close, first full data profiling run completed. The assumed review band was 8% of member population; the actual profiling shows 26%. This is not an unexpected variance; it is the structurally guaranteed consequence of the Class 5 estimate, which was legally imposed. Before close, member-level data was not available for examination.
Identity Resolution Review Band Discovery and Workload Capacity Analysis
Contents
1. Executive Summary
The initial identity resolution profiling has revealed a review band of 26% of the combined member population, significantly higher than the 8% assumed in the pre-close Deal Model (Class 5 estimate). This translates to a clerical review queue of 109,200 records instead of the assumed 33,600. At current steward capacity and throughput assumptions, the workload requires 273 working days to complete, far exceeding the 84 days initially assumed and exceeding the TSA migration window.
This is not an estimating failure. The Class 5 estimate was legally constrained: member-level data was not available for examination before the close. The actual profiling reveals what the structure of the data truly is, not a failure to predict it under enforced blindness. However, the result is a material constraint on the schedule and requires escalation and change control.
This workload does not fit inside the TSA window under current staffing. Three potential levers are unavailable: offshore capacity is barred by the regulatory condition (R-07, DOI data processing location restriction), throughput cannot be raised by pressure without increasing false-positive rate, and hiring and onboarding run weeks. The only viable lever is time, which requires a TSA extension.
2. Profiling Scope & Methodology
The profiling run examined all 420,000 Cumberland Valley members against the 1,800,000 ACME member base, using the same matching engine configuration that was used in internal profiling within each source before integration. The profiling was completed across three weeks (April 1–15, 2024), with data extracted from both organizations' source systems and staged in a non-production EMPI sandbox environment.
Three matching categories were evaluated:
Deterministic matching (SSN + DOB exact, or member ID exact match): Records that match on unique identifiers with no human review required. These become part of the consolidated record immediately on acceptance of the match.
Probabilistic auto-matching (Jaro-Winkler, Soundex/Double Metaphone, CASS/NCOA, above configured thresholds): Matches that exceed the automatic acceptance threshold and require no clerical review. Typical confidence: 92%+. These are accepted programmatically without worklist queuing.
Clerical review band (probabilistic matches below auto threshold, or cases requiring override): Records requiring human expert judgment. Matched by data stewards using the survivor-per-attribute method, with a documented decision on each pair reviewed.
Internal duplicates within target system (Cumberland Valley to Cumberland Valley): Duplicates identified within the source organization before integration. These are resolved in the source-system cleansing phase, not in the integration layer.
3. Assumed vs. Actual Match Rates
| Outcome | Assumed (Class 5) | Actual (Profiling) | Variance |
|---|---|---|---|
| Deterministic exact match | 65% | 41% | −24 pp |
| Probabilistic auto-match | 27% | 33% | +6 pp |
| Clerical review band | 8% | 26% | +18 pp |
| Internal duplicates (target) | 3% | 9% | +6 pp |
Deterministic shortfall (−24 pp): The assumed model expected 65% of records to match on SSN + DOB or member ID. Actual profiling shows only 41%. Root cause analysis of the 24% gap shows three drivers: (1) SSN completeness in the Cumberland Valley file is 67% (vs. the assumed 92%), contributing −15 pp; (2) duplicate SSN assignment within the target file (members issued multiple SSNs over time) contributes −5 pp; (3) DOB recorded differently or missing accounts for −4 pp. These are data quality issues inherent to the source system, not matching-engine failures.
Probabilistic recovery (+6 pp): Matches that fell below deterministic thresholds were captured by the probabilistic engine with confidence above the auto-acceptance threshold, adding 6 pp to the resolved set. This is within expectations and reflects the tuning of the matching algorithm.
Review band expansion (+18 pp): The clerical review band grew from the assumed 8% to 26%—a 3.25x expansion. This is the consequence of the deterministic shortfall: records that cannot match on unique identifiers require probabilistic evaluation below the auto-acceptance threshold, and such matches fall into the review band. These are pairs where the similarity score is high enough to merit consideration but low enough to require human judgment on whether they represent the same person.
The review band is not error: A wide review band is the correct response to poor deterministic coverage. Forcing probabilistic matches above human judgment (to avoid review) increases false positives, which are harder to detect and correct than false negatives. False positives merge two people in the system; nothing in normal operations surfaces that they were ever separate. False negatives produce a duplicate, which is annoying but visible and fixable.
4. Review Band Workload Impact
The review band queue is calculated as the number of records requiring clerical review, multiplied by the average handling time per record:
Assumed workload: 420,000 members × 8% = 33,600 records
Actual workload: 420,000 members × 26% = 109,200 records
The increase from 33,600 to 109,200 represents an additional 75,600 records requiring steward review. At 100 records per steward per working day, this is not a matter of efficiency or effort; it is a matter of duration and sequencing.
The actual review band is larger than the assumed one, but the distribution of difficulty within the band is unchanged. This is important: speeding up the stewards does not resolve the crisis, because a steward working faster at identity matching is a steward spending less time on the uncertain cases—and those are exactly the cases where false positives originate. A false positive merger is invisible until a clinician reads the wrong chart months later, or a claimant discovers they have coverage history for conditions they never had. Throughput pressure produces the wrong kind of speed.
5. Throughput & Duration Analysis
| Parameter | Assumed | Actual |
|---|---|---|
| Steward headcount | 4 | 4 (planned), 8 (post-CR) |
| Throughput per steward per day | 100 records | 100 records |
| Total daily capacity (4 stewards) | 400 records | 400 records |
| Review queue | 33,600 records | 109,200 records |
| Duration at 4 stewards | 84 working days | 273 working days |
| TSA migration window available | ~90 working days post-close | |
At the planned steward capacity of 4 FTEs, the 109,200-record queue requires 273 working days to complete. The TSA provides a ~90 working-day window between close and the point where operations must be fully independent (all interim systems retired). The queue requires 3x the available time.
Doubling the steward team to 8 FTEs (via CR-002, which adds a second cohort for the review band) reduces the duration to 137 working days, which still exceeds the window. However, 137 days falls within the negotiated TSA extension period (up to 18 months from close), making it manageable with a formal TSA term extension.
6. Lever Analysis & Constraints
Four potential levers exist to reduce the time-to-completion:
1. Increase steward headcount — Viable for incremental capacity. CR-002 adds 4 stewards (cohort 2) in the planned pipeline. However, recruiting and onboarding additional stewards beyond CR-002 requires 3–4 weeks per hire, and effectiveness ramps over 2–4 weeks. Adding staff immediately provides relief, but not unlimited relief; there is a maximum effective team size before coordination costs outweigh throughput gains. CR-002 is in the planned staffing model. Beyond that, curve flattens.
2. Offshore capacity — Not available. The state insurance regulator's Form A approval imposed a condition (R-07): member-level data cannot be processed offshore. This is a regulatory constraint, not a preference, and it was anticipated and accepted into the Risk Register as R-07 (marked "Accepted," because it is not mitigable—it is a law). CR-003 proposed adding 12 offshore data stewards to work down the queue; the proposal was rejected because it violates the regulatory condition. This lever is permanently closed.
On accepting R-07: A state data-processing restriction is not something a program can mitigate; it can only be complied with. The Risk Register records it as "Accepted" with a named accepter (R. Cadwallader, Regulatory Affairs) and the constraint reflected in platform policy and resourcing. That is more honest than inventing a mitigation plan that amounts to "we will follow the law." Registers that contain no accepted risks are usually registers where acceptance is happening without a name against it.
3. Raise throughput per steward — Counterproductive. The assumed throughput of 100 records per steward per day is already at the boundary where accuracy drops. The review band contains the uncertain cases: pairs where the matching score is high enough to be plausible but low enough to require judgment. A steward working at 150 records per day instead of 100 does so by spending 33% less time on each case—time that would normally be spent evaluating the uncertain cases. This increases false positives, the error type that does not surface in normal operations and is hardest to detect and correct post-close. Throughput cannot be a lever.
4. Reduce the review band size — Not available. The review band is determined by the matching-engine threshold configuration, which was tuned to minimize false positives. Raising the threshold to force more records into the auto-match category reduces review band size but increases the false-positive rate. The program has already accepted conservative thresholds (wider review band, fewer auto-matches) as the correct choice. The band size is therefore a parameter of the data, not a dial to be turned.
The only viable lever is time. Time is the rank 1 constraint in the Risk Register (R-02, marked Critical and owned by the Program Manager). Extending the TSA term to provide the 273 days (or, pragmatically, ~200 days after accounting for ramp and coordination) is the only change that resolves the crisis without introducing worse risks.
7. Recommendation & Escalation
The profiling results require immediate escalation to the Steering Committee and activation of CR-004 (TSA extension, three months, $3.45M). The Steering Committee decision is binary: (1) approve the TSA extension and complete identity resolution in parallel with operational cutover risk acceptance, or (2) constrain the review band by raising the false-positive threshold, accepting the risk that some merged identities will not be detected until well post-close.
There is no third option where the review band completes on the original timeline. The options differ in which constraint gives—whether time extends, or data quality lowers. The program's charter (Artifact 1, §3) establishes data integrity as a non-negotiable for Day 1 cutover. Therefore, time is the constraint that must give.
Steering Committee action required: Approve CR-004 (TSA extension to 18 months, enabling ~200+ working days for completion of identity resolution at the current steward throughput and accuracy). This consumes the negotiated margin but does not breach the contractual maximum. Rejection of CR-004 forces a decision to accept higher false-positive rate in identity resolution in order to meet the original cutover date.
Supporting materials for the Steering decision are attached to this report: detailed match-quality diagnostics (Appendix A), steward SOP and quality-acceptance criteria (referenced from Artifact 33), and the Change Request CR-004 with financial and schedule impact (Artifact 44, Change Control Log).
Owner: Dr. A. Ravindran (Data & EMPI)
Escalation path: C. Tyrrell (Program Manager) → D. Ashmore (Executive Sponsor) → Steering Committee
Decision required by: Target April 22, 2024 (7 calendar days; CR-004 approval window)
Related artifacts: 32 — EMPI Match Rules · 33 — Steward Standard Operating Procedures · 44 — Change Control Log (CR-004) · 28 — Risk Register (R-02, R-07)