Why Input Structure Matters
A customer-level price decision is only as sound as the policy record supplied to it. If one field describes a six-month term while another describes a full year, or if the recorded premium is last term’s amount rather than the renewal offer, later calculations can be internally consistent and still answer the wrong question. The first task is therefore to define a common policy-period basis and assemble comparable, traceable inputs.
For each policy being considered, the essential inputs include its current renewal premium, its prospective expected loss cost, relevant variable expenses, a renewal probability associated with a stated price, and a decision-eligibility status. These are not interchangeable quantities. The expected loss cost describes expected covered claims; the premium is the amount charged; expenses depend on their contractual or operational basis; and the renewal probability describes customer behavior under a specified offer.
A renewal model may supply a probability \(r_i(p)\) for policy \(i\) under a specified offered price \(p\). That value is meaningful only with its price basis and renewal period attached. A probability predicted for the current offer must not silently be treated as the probability at another price. The model’s intended use and limitations, including whether it supports the relevant policy and price context, should be respected.
Define the Basis Before the Fields
Start by specifying the unit of decision: typically one policy at one renewal date for one coverage term. Record the term length, effective dates, coverages, limits, and deductibles that define the risk being priced. If the policy is changing materially, its expected loss cost and behavioral prediction may no longer describe the same risk or offer.
The current premium should ordinarily mean the premium for the renewal offer being evaluated, before any hypothetical price change. Do not substitute the expiring premium merely because it is readily available. Also specify whether the amount includes taxes, installment charges, policy fees, or other charges. A renewal model trained on a particular definition of price should receive a consistent definition in the decision record.
Expected loss cost is prospective: it is the expected covered loss amount for the policy’s defined period and coverage terms. It is not necessarily the observed claims paid in the prior term. It should reflect the selected actuarial assumptions and any applicable adjustments, such as the expected exposure period or coverage changes. Where claim frequency and severity are used, their time basis and definitions must agree.
Variable expenses should be represented according to how they actually vary. A commission or premium tax may be a percentage of premium; a servicing charge may be a fixed amount per policy term; another expense may vary with claim count. Keep the basis explicit rather than combining unlike items into an unlabeled “expense” field. Fixed expenses that do not vary with the individual price decision should not be misclassified as premium-sensitive costs.
Expected Loss Cost: A Short Justification
Suppose aggregate loss \(L\) is the sum of the severities \(Z_j\) for the \(N\) claims in the policy period, with \(L=0\) when \(N=0\). If claim severities have a common finite mean and are independent of claim count, then conditioning on \(N\) gives:
Taking expectations of both sides and applying iterated expectation yields:
Thus the familiar frequency-times-severity calculation follows from the stated assumptions. It gives an expected aggregate loss amount, not a premium. The formula requires a common severity mean and the stated independence (or an appropriate conditional severity model); if severity depends on claim count or the risk mix, the simple product may not represent the expected loss.
Worked Example: Aligning a Six-Month Policy Record
A hypothetical insurer is preparing a renewal decision for a six-month auto policy. The current renewal offer is $840 for the six-month term. The prospective expected loss cost is $510 for that same term. Commission is 12% of premium, a premium tax is 2% of premium, and a servicing expense is $18 per policy term. The retention model gives a 0.740 probability of renewal at the $840 offer. Assemble and check the relevant inputs.
Method and assumptions. Use one six-month policy term as the unit for all dollar amounts. Assume the $840 is the renewal premium before the stated commission, tax, and servicing expense; the $510 loss cost is already estimated for the same coverages and six-month period. No interest or mortality assumption applies to this short-term property-casualty pricing record.
Premium and loss fields. Record current renewal premium as $840 per six-month term and expected loss cost as $510 per six-month term. Do not multiply or divide either by two: both already use the same term.
Expense fields. The commission rate is 0.12 and the premium-tax rate is 0.02, so the combined premium-based expense rate is \(0.12+0.02=0.14\), or 14%. At the current offer, commission is \(0.12(840)=\$100.80\), and premium tax is \(0.02(840)=\$16.80\). The servicing expense remains $18 per term. The current-price expense check is therefore \(\$100.80+\$16.80+\$18=\$135.60\).
Behavior and eligibility fields. Record renewal probability as 0.740, explicitly labeled as the probability at the $840 six-month offer. If the policy satisfies the insurer’s documented eligibility rules, mark it eligible; the problem supplies no reason to exclude it. The assembled record has matching six-month bases for premium, loss cost, and fixed servicing expense. The expense arithmetic checks to $135.60. This is an input validation, not a comparison of alternative prices.
Common mistake avoided: treating the $18 servicing amount as an 18% expense rate would produce a nonsensical percentage input. It is a dollar amount per policy term and must stay separate from the two percentage-based expenses.
Worked Example: Converting Annual Frequency and Severity to Term Loss Cost
A hypothetical six-month property policy has an annual claim count \(N_{12}\sim\operatorname{Poisson}(0.08)\). Assume claims occur at a constant rate through the year, so the six-month count \(N_6\sim\operatorname{Poisson}(0.04)\). Claim severities are independent of claim count and each other, with \(Z\sim\operatorname{Gamma}(2,6250)\), using shape 2 and scale $6,250. The current six-month renewal premium is $760, premium-based commission is 9%, servicing expense is $15 per term, and the modeled renewal probability at $760 is 0.680. Determine the expected loss-cost input and lay out the remaining fields.
Method and assumptions. Use the six-month policy term. The Gamma parameterization is shape-scale, so \(E[Z]=2(6250)=\$12,500\). The Poisson rate is prorated by time under the stated constant-rate assumption. No interest or mortality assumption applies.
Step 1: Calculate expected claim count for six months. The annual mean is 0.08, hence the six-month mean is \(0.08(6/12)=0.04\). This agrees with the supplied \(N_6\) distribution.
Step 2: Calculate expected aggregate loss. By the frequency-severity result and the stated independence, \(E[L_6]=E[N_6]E[Z]=0.04(\$12,500)=\$500\). Thus the expected loss-cost field is $500 per six-month term.
Step 3: Record price and expenses on their proper bases. Enter $760 per six-month term as the current renewal premium, 9% as the premium-based commission rate, and $15 per six-month term as servicing expense. As a check at the current offer, commission is \(0.09(\$760)=\$68.40\); the $15 servicing amount is not multiplied by the premium.
Step 4: Attach the behavior input. Record 0.680 as the renewal probability at the $760 offer, not as a price-independent probability. The inputs are aligned because all term-dollar values describe six months. The frequency-severity calculation checks to $500.
Common mistake avoided: using \(0.08(\$12,500)=\$1,000\) as the six-month loss cost applies the annual claim frequency to a six-month policy. The correction is to use the six-month mean, \(0.04\), giving $500.
Decision Eligibility Is an Input, Not an Afterthought
Not every record should automatically enter an individual price decision. Eligibility rules should be established and applied consistently before a policy’s inputs are treated as actionable. The rules may reflect the scope of the pricing program, data quality, coverage changes, legal or regulatory requirements, or whether the prediction process is appropriate for that record. The exact rules depend on the insurer and applicable requirements; they must be stated rather than inferred from a missing field.
Store both an eligibility flag and a reason code. A single “not eligible” value without an explanation is difficult to audit and may conceal different issues that require different remedies. A missing premium, for example, is not the same condition as a policy whose coverage changed and therefore needs a refreshed loss estimate. Eligibility should not be used to erase records from reporting: excluded policies still need to be counted, classified by reason, and handled through the applicable alternative process.
Specify policy, renewal date, coverage term, and the coverages and limits represented by the cost estimate.
Confirm that current premium, expected loss cost, and per-term expenses use the same term and clearly stated fee basis.
Label each renewal probability with the price and renewal context to which it applies.
Set the flag from explicit rules, record a reason for exclusions, and route excluded policies to an appropriate alternative process.
Worked Example: Applying Eligibility and Recording Exclusions
A hypothetical insurer reviews three policies for a renewal pricing program. Its written rule requires a valid current renewal premium, an expected loss cost for the renewal’s actual coverage, and no unresolved material coverage or exposure change. A policy failing any requirement is excluded from the individual decision run and receives a reason code. All amounts below are for one annual renewal term.
Policy A has a current renewal premium of $1,200, expected loss cost of $690 for unchanged coverage, a 10% commission rate, a $24 per-term servicing expense, and a modeled renewal probability of 0.810 at the $1,200 offer. Its data meet all three requirements, so it is eligible. Check the expense basis: commission at the stated offer is \(0.10(\$1,200)=\$120\), while servicing remains $24 per term.
Policy B has a current renewal premium of $1,050 and a predicted renewal probability of 0.770 at that offer. However, a material garaging change is unresolved, so the recorded loss cost still describes the old exposure. It fails the coverage/exposure consistency requirement and is excluded with the reason code “unresolved exposure change.” The available probability does not cure the stale cost input.
Policy C has an expected loss cost of $540 and a renewal probability of 0.700 at the displayed offer, but the current renewal premium is missing. It fails the valid-premium requirement and is excluded with the reason code “missing renewal premium.” It should not be assigned a zero premium or treated as eligible merely because other fields are present.
The eligibility tally is one eligible policy and two excluded policies: one for unresolved exposure change and one for missing premium. The tally checks against the three reviewed records. Each exclusion has a distinct reason that identifies what must be resolved before the record could be reconsidered.
Common mistake avoided: admitting Policy B because its renewal probability is available confuses behavioral-model output with complete pricing inputs. Eligibility requires the cost and behavior fields to describe the decision being made, as well as satisfying the stated rules.
Common Input Errors and Practical Checks
Several errors recur when actuarial estimates and renewal predictions are joined at policy level. They are preventable with a small set of explicit checks before any candidate prices are evaluated.
- Mixing periods. An annual expected loss cost cannot be compared directly with a six-month premium. Convert using an appropriate exposure basis and assumptions; do not assume simple time proration when risk varies seasonally or coverage changes.
- Using the wrong premium. Expiring premium and current renewal premium are different fields. Identify which amount is the baseline offer and use a consistent definition of premium throughout.
- Combining unlike expense bases. A percentage of premium and a fixed dollar cost per term require separate inputs. Confirm whether fees are included in the modeled price or handled separately.
- Detaching probability from price. Store the renewal probability with its price point, term, and model context. A probability at one price does not by itself specify behavior at another.
- Ignoring missing or stale data. A numeric value is not necessarily a valid value for the current risk. Coverage changes and missing fields need explicit treatment, not silent substitutions.
A practical record should retain enough provenance to reproduce its fields: source and valuation date for the loss estimate, source and definition for the premium, expense basis, model version or prediction context, and eligibility reason. This does not replace actuarial judgment; it makes the judgment and the resulting inputs reviewable.
Check Your Understanding
For each problem, keep the policy-period basis and the meaning of every field explicit.
- A hypothetical three-month policy has an expected loss cost of $240 for the quarter and a current renewal premium of $390 for the same quarter. Commission is 8% of premium and an inspection charge is $12 per quarter. What are the commission and total of these two expense amounts at the current offer? (Answer: commission $31.20; total expenses $43.20.)
- A hypothetical policy has annual Poisson claim frequency with mean 0.12. Assume a constant claim rate through the year and independent claim severities with \(Z\sim\operatorname{Gamma}(3,4000)\), shape-scale. What is the expected loss cost for a six-month term? (Answer: \(0.06(3\times\$4,000)=\$720\).)
- A renewal prediction is 0.760 at a $900 annual offer. The expected loss cost is $580 for six months, and the premium record is annual. What must be resolved before these fields can be treated as one annual decision record? (Answer: align the loss-cost period to the annual term using an appropriate exposure assumption; the 0.760 prediction must also be labeled as applying to the $900 annual offer.)
- A policy has a valid renewal premium and a probability prediction, but the policy’s deductible changed after the loss-cost estimate was produced. Under an eligibility rule requiring costs to match current coverage, how should it be handled? (Answer: exclude it pending a loss-cost estimate consistent with the changed deductible, and record the coverage-change reason.)
- A hypothetical annual policy has an $1,100 current renewal premium, a 7% commission, $20 per-term servicing expense, and expected loss cost of $640. The predicted renewal probability is 0.690 at $1,100. State the correctly based input values and the commission arithmetic. (Answer: premium $1,100 per year; loss cost $640 per year; commission rate 7%; servicing $20 per year; renewal probability 0.690 at $1,100; commission \(0.07(\$1,100)=\$77\).)