Collections Analytics — Leadership Brief

7 months of data (Jan–Jul 2026) · 30,000 accounts · one row per payment_id · revised Sep 2026 (see CORRECTIONS.md)

+11% is one month, not a trend Recovery flat: ₹18.72 Cr in Jan and in Jul
March was +11.0% on February, but February was −9.1% on January. Month-on-month changes from February to July average +0.3%. Real duplicate payments are small (325 successful rows, ₹2.45 Cr, ~1.9%) and evenly spread, so they do not explain March. Contact, promise-kept rate, productivity and portfolio mix are all flat.
Avg MoM change, Feb–Jul
+0.3%
Reported: +11% (March only)
Total Recovery (Jan–Jul)
₹126.8 Cr
After removing ₹2.45 Cr of duplicates
Accounts paying / portfolio
7.8%
July · range 7.2–8.1% Flat
Right-party contact
76–79%
of dispositioned calls Flat
Channel lift vs baseline
≈ 0
All 95% intervals include zero

Monthly Recovery — deduplicated by payment_id vs raw (₹ Crores)

Last-touch credit, 30-day window (₹ Cr) · plus ₹80.8 Cr (64%) with no contact

Metric Trend (Jan–Jul 2026)

Month Paying rate PTP Kept% ₹/Agent-Hr

Data Forensics — Issues Found

₹10 Cr Recommendation

→ Test before committing: no channel shows measurable lift
Step 1 (proposed ≤ ₹1.5 Cr)8-week randomised holdout, 20% per channel
Detectable lift≈ 1.3 pp (80% power, 5% level)
Step 2 (proposed ≤ ₹1.5 Cr)Fix data at source: payment_id, one PTP code, UTC, agent master
Step 3 (≥ ₹7 Cr)Deploy to the channel the test proves, by lift per rupee
Why not WhatsApp now?Most credit only because it sends the most messages; lift +0.07 pp
April strategy shiftDiD +1.3 pp, 95% CI −0.2 to +2.9: no detectable effect
Why not more agents?₹/agent-hour flat (₹16.1k–17.0k); contact shows no lift