The Renewal Book That Won't Balance
Meadowbrook Mutual's personal lines unit renewed 4,200 auto policies last quarter. Underwriting had promised the CFO a 96% retention figure, but the finance reconciliation shows net written premium of $3.86M against a prior-term book of $4.20M. The retention report says 95.8% of policies renewed. Digging in, you find: 180 policies were cancelled flat mid-term (no premium earned), 240 were rewritten at renewal under a new policy number because the insured changed vehicles, and 90 were non-renewed for claims history. The CFO wants the gap explained before the board meeting. The retention report is calculated as renewed policies divided by policies eligible for renewal, counting rewrites as new business and excluding flat cancellations from the denominator entirely.
1. Is the 95.8% retention figure a defensible measure of what the CFO actually cares about? Explain what it is measuring versus what the premium drop suggests.
Read the worked answer
The 95.8% figure is a policy-count retention metric, and it is internally consistent with its own definition - but it is not measuring what the CFO cares about, which is premium retention (the revenue base holding).
Two distortions matter:
Rewrites are counted as new business. The 240 rewritten policies are economically renewals - the same customer, the same risk, reissued under a new number. Excluding them from the numerator understates retention. If they were counted as renewals, retention rises.
Flat cancellations are excluded from the denominator. The 180 flat-cancelled policies were genuinely at risk and largely represent lost business. Dropping them from the denominator inflates the ratio by shrinking the base.
The premium evidence contradicts the count: NWP fell from $4.20M to $3.86M, an 8.1% decline. Even allowing for rate and exposure changes, a book retaining 95.8% of policies would not normally shed 8% of premium unless average premium per policy fell sharply - which the facts don't support. The honest read is that count-retention and premium-retention have diverged, driven by rewrites (which may carry different premiums), flat cancellations, and the 90 non-renewals.
The defensible move is to present both a corrected count metric (rewrites as renewals, flat cancellations in the denominator) and a premium retention metric, and explain the reconciliation between them. Presenting 95.8% alone to the board would be misleading.
2. Recompute a more honest policy-count retention figure using the facts given, showing your arithmetic and stating your assumptions.
Read the worked answer
Start from the report's logic and correct it.
The report's denominator excludes the 180 flat cancellations. Its numerator excludes the 240 rewrites. We don't have the raw eligible count, so reconstruct it.
If 95.8% = renewed / eligible (per the report's definition), and renewed = eligible − 90 non-renewals (assuming non-renewals are the only non-renewed category in their base), then:
- 0.958 = (E − 90) / E
- 0.958E = E − 90
- 0.042E = 90
- E ≈ 2,143
That is implausibly low against a 4,200-policy book, which signals the report's denominator is not what I assumed - a good reason to flag the metric as unreliable rather than trust the reconstruction. The arithmetic itself is telling us the stated definition can't produce 95.8% on this book, so the report is either using a different base or contains an error.
A cleaner approach: build retention on the full eligible population. Eligible ≈ 4,200 (policies in force at start of term, treating the book as the exposure base). Renewed (including rewrites as renewals) = 4,200 − 180 flat cancellations − 90 non-renewals = 3,930. Retention ≈ 3,930 / 4,200 ≈ 93.6%.
Caveat: if flat cancellations occurred after renewal (mid-term), they shouldn't reduce retention at all - they're post-renewal attrition. The facts say 'cancelled flat mid-term,' which suggests they were in force and then cancelled, meaning they may have renewed first. That distinction changes the numerator and is the key assumption to verify before publishing any number. The model answer should state this explicitly rather than assert 93.6% as fact.