Skip to content

Latest commit

 

History

History
75 lines (60 loc) · 4.62 KB

File metadata and controls

75 lines (60 loc) · 4.62 KB

Decision Memo: Add Product-Category Benchmarking to Servicer Complaint Review

To: VP, Consumer Response Operations From: Business Analyst, Consumer Response Analytics Date: 2026-07-09 Decision requested: Approve building a within-product-category servicer scorecard (vs. the current practice of comparing servicers against a single portfolio-wide relief-rate target)

Portfolio artifact — self-directed analysis on public CFPB data; the addressee role is illustrative, not a real engagement (see repo README).


Recommendation

Stop comparing complaint-handling outcomes across companies using a single portfolio-wide relief rate. It produces false positives. Instead, benchmark each company only against peers in its own product category, and flag deviations from that baseline. This CA CFPB dataset shows exactly how the current approach fails, and exactly which company it would have wrongly flagged.

The problem, quantified

Relief rate (share of complaints closed with monetary or non-monetary relief) varies 14x by product alone — 2.6% for student loans vs. 36.8% for credit cards — before any company-level effect is considered (KPI-02, sql/10_relief_outcomes.sql). A portfolio-wide comparison naively flagged MOHELA (939 CA student-loan complaints, 2.8% relief rate) as an outlier against an "all other companies" pool averaging 23.4% relief. That comparison is wrong: student loan is structurally the lowest-relief product in the entire dataset, and MOHELA's rate is in line with its own peers (Nelnet 0.0%, Maximus 0.0%, Navient 0.6%). Flagging MOHELA on this basis would waste review effort on a company that isn't actually deviating from its category.

Evidence

  1. Relief rate by product ranges from 2.6% (student loan) to 36.8% (credit card) — a structural difference in what "relief" means per product, not a quality signal (sql/10_relief_outcomes.sql).
  2. Within student loan specifically, 4 of 8 servicers with sufficient volume show 0.0% relief (Maximus, EdFinancial, Nelnet, and the Federal Student Aid contractor) — MOHELA's 2.8% is roughly mid-pack, not a low outlier (sql/10_relief_outcomes.sql).
  3. The real company-level outlier is EdFinancial Services: 50.7% timely-response rate, roughly half of every other reliably-sized peer servicer (88–100%; excludes the Federal Student Aid contractor entry, n=57, flagged separately below as too thin to interpret), while still granting 0% relief (sql/10_relief_outcomes.sql, student-loan servicer breakdown).

Options considered

Option Cost/effort Expected impact Why / why not
Do nothing (keep portfolio-wide comparison) None Continues generating false-positive flags on structurally low-relief products (student loans, mortgages, money transfer) while missing real outliers like EdFinancial Cheapest, but actively misdirects review effort
Recommended: within-category benchmarking (KPI-04) Low — same underlying data, different grouping logic in the KPI query Correctly identifies EdFinancial-style outliers; stops flagging structurally-low-relief products as if they were badly-run ones Requires updating the KPI definition and any existing dashboard/report built on the flat portfolio-wide rate
Build a full statistical control model (regression adjusting for product, complaint type, region) High — new modeling work, ongoing maintenance Marginal accuracy gain over simple within-category peer comparison at this data volume Not justified yet; revisit if false-positive/negative rates on the simple version prove too high in practice

Risks & open questions

  • n=57 for the Federal Student Aid contractor entry is thin — its 0% relief and 0% timely rate is a striking combination but shouldn't be acted on without a larger sample or a longer date window.
  • Federal student loan servicers operate under different, more constrained relief authority than banks or private lenders by regulation — some of the category-wide low relief rate may be a genuine structural ceiling, not a solvable service-quality problem. Within-category benchmarking controls for this at the category level, but can't fully separate "structurally constrained" from "poorly run" at the individual-servicer level without a longer time series or a second data source.
  • This analysis is CA-only; whether EdFinancial's timeliness gap holds nationally is unverified.

Next step

Pilot the within-category scorecard on the student loan product category (8 servicers, already scoped) for one quarter, using KPI-04 as defined in the KPI dictionary, before extending to the other 9 product categories.