Gains:
- Ability to explain actuarial concepts such as loss frequency, damage severity and premium adequacy with artificial intelligence and prepare an analysis draft
- Ability to test pricing model output, data quality and assumptions against independent accounting and actuarial principles
- Ability to understand that final tariff and technical provision decisions are the responsibility of competent actuaries and that artificial intelligence is only a support tool
The price of insurance is not a coincidence; It is calculated mathematically. Actuary (the field of expertise in insurance that measures risk with probability and statistics and establishes the balance of premium, provision and coverage) answers the question "what will this risk cost us on average and how can we cover it with a fair premium?" The actuary (the competent expert who makes this calculation and is legally responsible) brings together damage data, trends and assumptions and determines the tariff (price scale) and technical provisions (money set aside for future claims). Artificial intelligence (software that learns patterns from data and generates text, descriptions, and outlines) is a powerful helper in this heavy analytical work: explaining concepts, summarizing data, drafting analysis, suggesting code and formula. But the actuarial decision carries a signature and responsibility; Artificial intelligence cannot make this signature. In this unit, you will learn how to explain and analyze basic actuarial concepts with artificial intelligence and how to test the output against independent calculations and actuarial principles.
Basic concepts: Frequency, intensity and premium adequacy
Three concepts form the backbone of pricing:
- Claim frequency: The number of claims per policy (or unit) in a given period. For example, if there are 45 claims per year in 1,000 car insurance policies, the frequency is 0.045 (i.e. 4.5%).
- Damage severity: Average amount paid per claim. If a total of 3,150,000 TL was paid for 45 damages in the same portfolio, the severity is 70,000 TL.
- Expected damage cost (risk premium): Frequency × severity. In the example, 0.045 × 70,000 = 3,150 TL. This is the expected pure loss cost per policy; Gross premium is formed when expenses, commissions and profit margin are added.
- Premium adequacy: Whether the premium collected is sufficient to cover expected damages and expenses. Insufficient bonus puts the company in loss; Excessive premium creates uncompetitive and unfair pricing.
AI is very useful in explaining these concepts, extracting a frequency/intensity summary from a data set, and outlining analysis steps. However, you must always recalculate the numerical result.
Tip: Having the artificial intelligence explain an actuarial concept "with a numerical example, as if explaining it to a 10th grade student" is very useful both in team training and in explaining it to the customer. But check the number: even model multiplication can get it wrong.
Where does artificial intelligence stand in actuarial
The red line in the actuary is clear: the final tariff, technical provision and premium adequacy decision is the responsibility of the competent actuary. AI helps with the following tasks: concept clarification, data summary, exploratory analysis outline, formula/code suggestion, making the resulting text readable. It cannot have the final say in the following matters: approving the tariff, declaring provision adequacy, accepting assumptions as final. Because these decisions directly affect both the financial health of the company and the customer's right to a fair price and are subject to legislation (SEDDK - Insurance and Private Pension Regulation and Supervision Agency regulations).
Attention: A tariff or rate suggested by artificial intelligence will never be published without independent testing of data quality and assumptions. A wrong assumption (e.g. ignoring the impact of inflation) could put the entire portfolio in harm's way.
Step by step: Draft AI-supported actuarial analysis
- Anonymize and identify data. No personal data; aggregated numbers only. Specify period, portfolio and collateral type.
- Clarify the concept and method. Ask the AI to outline the steps of the frequency/intensity analysis.
- Question data quality. Are there any effects of missing data, outliers, period mismatches, IBNR (claims not yet reported)?
- Write the assumptions clearly. On what assumption are inflation, trend, development factors based?
- Calculate the result independently. Reproduce frequency, intensity and risk premium with your own spreadsheet.
- Test and confirm with actuarial principle. A competent actuary evaluates and signs in terms of premium adequacy, fairness and legislation.
three mini cases
Case 1 — Catching the wrong assumption. An assistant actuary received a draft analysis from artificial intelligence for the automobile insurance tariff. Artificial intelligence took the average damage severity of the past 3 years as 62,000 TL and calculated the risk premium. The actuary realized that the violence had not been adjusted for inflation to present value: with current vehicle repair costs, the violence was actually ~$89,000. This correction increased the risk premium by 43%. Without independent control, the tariff would be seriously inadequate.
Case 2 — Distortion of outlier. The total of 500 damages in one portfolio appeared to be 40,000,000 TL; but 22,000,000 TL of this was from a single major fire damage. Artificial intelligence summarized the average violence as 80,000 TL. The actuary evaluated this single major damage separately (because it does not recur every year); The average of "normal" damages was ~36,000 TL. Separating the outlier made the tariff realistic.
Case 3 — Justice check. In a health insurance pricing project, artificial intelligence offered very high increases to a certain age group. The actuary found that the recommendation came only from a small and noisy sample (small number of claims); It was not statistically reliable. Moreover, it was against the fair pricing principle required by the legislation. The actuary first made a reliability calculation with more extensive data and then made a decision.
Four copyable prompts
1) Concept explanation (for training/customer):
Act like an actuarial instructor. Explain the concepts of "loss frequency", "damage severity" and "risk premium" step by step with a single consecutive numerical example (e.g. 1,000 policies, 45 claims, 3,150,000 TL total payment). Show the formula and intermediate result at each step. Use simple and understandable language.
2) Analysis draft skeleton:
The following outlines the steps of a frequency-severity analysis for aggregated claims data (without personal data): [period, number of policies, number of claims, total payment]At each step: specify what is calculated, what assumptions are required, what risk of data quality exists (missing data, outliers, IBNR, inflation). I will recalculate the numerical result; you set up the method.
3) Data quality and assumption questioner:
Critique the following draft analysis from the perspective of a senior actuary:[paste draft]Ask: Has violence been adjusted for inflation? Were outliers removed? Is the sample reliable? Was IBNR taken into account? What is the trend assumption?List each weakness under "RISK" with a suggestion for correction.
4) Simplifying the result text:
Convert the following verified analysis result into a simple executive summary for a manager who is not an actuary: [paste result]Do not change numbers, add new numbers. Clearly state uncertainties and assumptions. Add note: "The final tariff decision is subject to the approval of the competent actuary."
Weak prompt / Strong prompt
Weak: "Calculate the car insurance premium with this data and write the tariff."
Problem: Direct tariff (red zone) requested; no assumptions, no data quality and no validation. The model fits unreliable rate from noisy data.
Strong: "Draft the steps of the frequency-intensity analysis; specify the assumption and data quality risk at each step. I will calculate the numerical result. Do not make the tariff decision."
Why it's good: AI establishes the method, the actuary keeps the calculation and decision; assumptions become visible.
comparison chart
step
AI role
Actuary role
Risk
Concept explanation
Explains, examples
Confirms accuracy
wrong example
Data summary
aggregates
Checks the source
Outlier distortion
Assumptions listing
Recommends
accept/reject
Missing/incorrect assumption
Risk premium calculation
draft formula
Independent accounts
Account error
Tariff/provision decision
—
Decides and signs
Insufficient premium
Common mistakes
- Getting the number of artificial intelligence directly. The model can even make multiplication/addition inaccurate; recalculate each result.
- Skipping inflation and trend correction. Establishing tariffs without bringing past violence into the present will damage the portfolio.
- Mixing the outlier with the mean. One major damage misleads the entire tariff; Evaluate separately.
- Relying on small sampling. The ratio from a small number of damages is unreliable; A creditworthiness calculation is required.
- Ignoring justice. It is against the law to offer an unfair raise to a group based on statistical noise.
In summary
In actuarial work, artificial intelligence explains concepts, summarizes data, outlines analysis and suggests formulas; but it does not decide on tariff, provision and premium adequacy. Use the concepts of frequency, intensity and risk premium correctly; question data quality (outlier, IBNR, inflation, sample reliability); calculate each number independently; Write the assumptions clearly. The final decision is the responsibility of the competent actuary and is subject to SEDDK regulations.
Application task
Generate frequency and severity summary with artificial intelligence from a damage data set you have (for testing purposes, aggregated). Then do the same calculation yourself in a spreadsheet. Note the difference between the two results; If there is a difference, find the reason (inflation, outlier, wrong multiplication). Critique your assumptions with prompt #3 and identify at least two data quality risks.
checklist
- [ ] I used only aggregated, non-personal data.
- [ ] I recalculated the frequency, intensity and risk premium myself.
- [ ] I applied inflation/trend correction.
- [ ] I evaluated outliers separately.
- [ ] I questioned the reliability of the sample.
- [ ] I wrote the assumptions clearly.
- [ ] I left the final tariff/provision decision to the competent actuary approval.