Monitoring Retention, Quote Conversion, and Selection

Why Monitor Retention, Conversion, and Selection?

A rate launch can change more than premium. A policyholder may choose not to renew after receiving a higher renewal offer; a prospective customer may decline a new-business quote; and the risks that remain or bind may differ from those that leave. Monitoring these outcomes helps determine whether observed production changes are consistent with the intended response to the new rates or could instead reflect changes in distribution, underwriting, or the measured population.

The three measures answer different questions. Retention describes what happens to eligible existing business at renewal. Quote-to-bind conversion describes what happens to eligible new-business quotes. Selection monitoring examines whether the mix of risks retained or bound changes. None of these measures, on its own, proves that a price change caused the observed result. Comparisons require consistent definitions, cohorts, timing, and exposure basis.

Define the Population Before Calculating a Rate

For renewal retention, state which policies had a genuine renewal opportunity and the date or period by which renewal is determined. A policy that terminated before a renewal offer was prepared may not belong in the eligible denominator, depending on the monitoring definition. Report such terminations separately rather than silently treating them as either renewals or nonrenewals. Also decide how to handle rewrites, transfers between products, cancellations after renewal, and policies with multiple locations or vehicles. A policy-count rate gives each policy equal weight; an exposure-weighted measure gives more weight to policies with more exposure. These are different measures and should not be interchanged.

For quote conversion, define a quote as a unique, eligible quoting opportunity. Specify whether revised quotes for the same applicant are deduplicated, how withdrawn or incomplete quotes are treated, and how long a quote can take to bind. A quote made near the end of the observation period needs enough time to mature; otherwise its outcome may be recorded as a nonbind simply because the decision is not yet known. State the binding event used, such as an issued policy or an accepted bind request, and apply it consistently.

Use comparable pre- and post-launch cohorts. Align the renewal effective dates or quote dates, observation windows, and development of outcomes. Reconcile counts to policy and quote records, and check that premiums, exposures, and eligibility rules have not changed unnoticed. Earlier material on monitoring baselines and observation windows and reconciling written premium, earned premium, and exposure provides useful context for these checks.

Conversion Rates and Channel Mix

An overall quote conversion rate is a weighted average of the conversion rates for the channels or segments included in the cohort. If \(n_{jt}\) is the number of eligible quotes in channel \(j\) during period \(t\), and \(r_{jt}\) is that channel's bind rate, then the overall rate is \(\sum_j w_{jt}r_{jt}\), where \(w_{jt}=n_{jt}/\sum_j n_{jt}\). The weights change when the channel composition changes. Consequently, an overall rate can move even if no channel's own conversion rate changes, or it can mask the direction or size of within-channel changes.

A fixed-weight standardized rate answers a narrower comparison question: what would each period's conversion rate have been if both periods had the same channel mix? Choose and disclose the reference weights, such as the baseline distribution. This removes the arithmetic effect of channel-share changes from that comparison, but it does not establish that price caused the remaining difference. Underwriting rules, lead quality, agent practices, quote timing, and other conditions may also have changed.

The justification follows directly from the weighted-average identity. The observed aggregate is \(\sum_j w_{jt}r_{jt}\), so it depends on both the segment rates and their shares. Holding the weights at the same \(w_j\) in each period makes the comparison \(\sum_j w_j(r_{j1}-r_{j0})\). Thus, differences in channel shares cannot by themselves change this standardized comparison. It still reflects any within-channel changes, including changes in applicants, underwriting, or distribution practices.

Monitor Selection in the Business That Remains or Binds

A retention or conversion rate reports how often a policy or quote produces a desired outcome; it does not describe the risk composition of those outcomes. For that, compare characteristics such as territory, coverage, limit, deductible, prior loss history, or a modeled expected loss measure across eligible, retained, and bound business. Use consistent definitions and exposure units. A shift in mix can change portfolio-level expected cost even when each segment's expected cost is unchanged.

For a simple two-segment illustration, let \(w_s\) be segment \(s\)'s share of policies and \(c_s\) its expected cost per policy under a stated model. The portfolio average is \(\sum_s w_sc_s\). If the shares change, the average changes even if the segment costs do not. This is a monitoring comparison, not a prediction that every individual policy will have that cost. Where policy sizes differ materially, compare exposure-weighted measures or a common unit of exposure instead of treating every policy as equivalent.

A useful monitoring table follows the same cohort through the relevant stages: eligible renewal offers or eligible quotes; renewed or bound outcomes; channel and underwriting characteristics; and exposure or expected-risk measures. Compare like with like, including underwriting rules and eligibility. If a territory becomes ineligible, for example, a change in the retained mix may result from the rule change rather than customer response to price. If the source of a difference is uncertain, describe it as an observed association and investigate before attributing cause.

Interpretation, Common Traps, and Practical Checks

A before-and-after change is not automatically a price elasticity or causal estimate. To interpret a rate launch, examine whether the periods differ in channel participation, underwriting eligibility, quote quality, exposure size, product design, or observation maturity. Compare segment-level outcomes where the data support it, and keep separate measures for retention, conversion, and risk mix. A favorable average risk shift can accompany a retention decline; neither result alone tells whether the resulting portfolio is adequate or desirable.

A common denominator error is to divide renewed policies by all policies originally scheduled to reach renewal, even when some never received a renewal opportunity under the stated definition. In the first example, the incorrect calculation \(768/1{,}000=76.80\%\) would understate eligible-policy retention, which is \(768/960=80.00\%\). Conversely, excluding policies that did receive a valid offer but did not renew would overstate retention. The correction is to define eligibility in advance, apply it consistently, reconcile the outcomes, and report excluded or separately classified policies.

• Keep outcome windows consistent. A late quote should not be labeled a nonbind before its observation window has elapsed.
• Deduplicate opportunities consistently. Multiple revisions for one applicant should not inflate the quote denominator unless the metric explicitly counts quote versions.
• Report channel and risk mix. An aggregate rate can move because the composition changed, even if segment rates did not.
• Do not infer causation from standardization. Fixed weights address a specified mix difference, not unmeasured changes within segments.
• Reconcile with exposure and production records. Policy counts, exposure, written premium, and earned premium measure different things; use the measure that answers the stated question.