Problem Definition
Helix Assurance is a regional US personal auto insurer with about 1.3 million policies across eight states and about $1.6 billion of earned premium, an average of roughly $1,230 per policy. It handles about 150,000 claims a year through a traditional process: every claim goes to an adjuster, most repairable vehicles are inspected in person, and suspected fraud is flagged by a set of fixed rules written ten years ago.
Helix is losing money on underwriting. Its combined ratio (losses plus expenses, divided by premium) is about 104.7%, while the US private auto industry improved to a combined ratio of approximately 95 in 2024, and S&P Global projected approximately 93 for 2025. Claims are also slow. Helix's average repair cycle for repairable vehicles is 26 days, against approximately 19 days in J.D. Power's 2025 industry study, and customer complaints about claims have doubled in two years.
Costs per claim keep rising. Bodily-injury severity grew approximately 10% in 2025 according to CCC, and a record share of damaged vehicles is now declared a total loss. At the same time, industry estimates suggest that fraud occurs in about 10% of property and casualty insurance losses.
The new Chief Claims Officer has a $25 million budget for a data and analytics programme. She wants to know where it will pay off.
How should Helix use data to settle claims faster and reduce fraud, and how much can it improve its combined ratio?
Additional InformationAsk for dataInterviewer’s data
If asked, please share that:
- Helix has seven years of digitised claims history (about 1 million claims), including adjuster notes, photos, repair invoices and outcomes of fraud investigations
- Helix's Special Investigations Unit (SIU) has 40 investigators, each able to complete about 150 cases a year
- Customers can already upload photos through the mobile app, but adjusters rarely use them to write estimates
- About 35% of repairable-vehicle claimants use a rental car paid by Helix while their vehicle is repaired
- State insurance regulators expect insurers to be able to explain automated claim decisions and show that models are not unfairly discriminatory
Question 1Structuring
How would you structure the opportunity to use data in Helix's claims operation?
Hint · Structuring
Build 3–4 branches that are specific to this client and question, not a generic framework. Check they don't overlap and together cover the problem.
Try it first, then checkCheck my answerModel answer
1. Where the value is (claim journey)
- First notice of loss: collect structured data and photos at intake
- Triage: score each claim for complexity and fraud risk
- Assessment: photo-based estimates for simple damage, in-person inspection only when needed
- Settlement: automatic payment for low-risk, low-value claims
- Recovery: subrogation and salvage
2. Value levers
- Loss costs: less fraud and inflated claims, more accurate estimates
- Loss-adjustment expense (LAE): less adjuster time per claim
- Indirect costs: fewer rental-car days, lower storage fees
- Customer: faster settlement, better retention
3. Enablers and constraints
- Data: history, labels (confirmed fraud outcomes), photo quality
- Model: accuracy, explainability, bias testing
- Operations: SIU capacity, adjuster roles, repair-shop network
- Regulation: fair claims settlement rules, model governance
What the interviewer is looking forInterviewer’s viewInterviewer’s view
A good candidate follows the claim from first notice of loss to settlement and names where data can help. A strong candidate separates the two goals, speed and fraud, and notes that they pull in opposite directions: more checks catch more fraud but slow honest claims. An excellent candidate proposes one scoring engine that sends each claim down the right path (fast track, standard, investigation), and asks about constraints early: SIU capacity, regulation, data quality.
So What? cascade:
- Level 1: Helix is slower and less profitable than the industry
- Level 2: both problems come from treating every claim the same way. Low-risk claims wait in the same queue as suspicious ones
- Level 3: the value of data is in triage: sorting claims early so honest customers go fast and investigators only see the riskiest cases
Question 2Numeracy
Using Exhibit 1, how much fraud is in Helix's book, and how much net value would a new fraud model create vs. the current rules if SIU investigates the top 5% of claims?
Hint · Numeracy
Write the formula before you plug in numbers, keep units and zeros explicit, and sanity-check the order of magnitude at the end.
Exhibit 1Helix Claims Mix and Loss Costs (last 12 months)
| Coverage | Claims | Average severity | Incurred losses ($M) | Industry average severity (2024) |
|---|---|---|---|---|
| Collision | 60,000 | $5,500 | 330 | approx. $5,490 |
| Comprehensive | 30,000 | $2,300 | 69 | approx. $2,310 |
| Property-damage liability | 45,000 | $6,800 | 306 | approx. $6,770 |
| Bodily-injury liability | 15,000 | $28,000 | 420 | approx. $28,280 |
| Total | 150,000 | 1,125 |
Earned premium $1,600M; loss-adjustment expense (LAE) $190M; underwriting expenses $360M.
Source: Helix Assurance case file
Additional InformationAsk for dataInterviewer’s data
Share Exhibit 1 and the following:
- Assume about 10% of Helix's incurred losses are fraudulent or inflated (in line with the industry rule of thumb)
- About 6% of claims (9,000) contain some fraud, with an average fraudulent amount of $12,500
- When SIU confirms a fraud, Helix avoids about 80% of the fraudulent amount, or about $10,000 per case
- Each SIU investigation costs about $1,500 in external costs (records, surveillance, experts)
- Current rules flag 6,000 claims a year, and 12% of those turn out to be fraud
- At the top-5% threshold the model flags 7,500 claims, and 30% of those turn out to be fraud
Try it first, then checkCheck my answerModel answer
Step 1: Combined ratio check
- Loss ratio: $1,125M / $1,600M = 70.3%
- LAE ratio: $190M / $1,600M = 11.9%
- Expense ratio: $360M / $1,600M = 22.5%
- Combined ratio: ($1,125M + $190M + $360M) / $1,600M = $1,675M / $1,600M = 104.7%
Step 2: Size of the fraud problem
- 10% x $1,125M = $112.5M
- Check: 9,000 claims x $12,500 = $112.5M
Step 3: Current rules
- Frauds caught: 6,000 x 12% = 720 (720 / 9,000 = 8% of all fraud)
- Savings: 720 x $10,000 = $7.2M
- Cost: 6,000 x $1,500 = $9.0M
- Net: -$1.8M a year
Step 4: Model at the top-5% threshold
- Frauds caught: 7,500 x 30% = 2,250 (2,250 / 9,000 = 25% of all fraud)
- Savings: 2,250 x $10,000 = $22.5M
- Cost: 7,500 x $1,500 = $11.25M
- Net: +$11.25M a year
Step 5: Improvement vs. today
- $11.25M - (-$1.8M) = about $13M a year, or about 0.8 points of combined ratio ($13M / $1,600M)
What the interviewer is looking forInterviewer’s viewInterviewer’s view
The candidate should first check the combined ratio from the exhibit, then size the fraud and compare the two approaches. A strong candidate notices that the current rules lose money: they cost more to investigate than they save. An excellent candidate points out that Helix's average severities match the industry, so the problem is not that Helix pays too much per claim on average. It is that it cannot tell the good claims from the bad ones.
So What? cascade:
- Level 1: fraud costs Helix about $112.5M a year
- Level 2: the current rules catch only 8% of fraud cases and lose $1.8M a year. The model catches 25% and makes $11.25M
- Level 3: precision (the share of flagged claims that are truly fraud) is what makes an investigation programme profitable. Better targeting beats more investigations
Question 3Judgement & Insights
Exhibit 2 shows the model's results at three thresholds. Which threshold should Helix choose, and why?
Hint · Judgement & Insights
Read the exhibit title, axes and units first. Lead with the ‘so what’, then back it with one or two numbers.
Exhibit 2Fraud Model Performance by Threshold
| Threshold (share of claims flagged) | Claims flagged | Precision | Frauds caught | Share of all fraud caught | Honest customers flagged | Savings ($M) | Investigation cost ($M) | Net value ($M) |
|---|---|---|---|---|---|---|---|---|
| Top 2% | 3,000 | 45% | 1,350 | 15% | 1,650 | 13.5 | 4.5 | 9.0 |
| Top 5% | 7,500 | 30% | 2,250 | 25% | 5,250 | 22.5 | 11.25 | 11.25 |
| Top 10% | 15,000 | 18% | 2,700 | 30% | 12,300 | 27.0 | 22.5 | 4.5 |
| Current rules | 6,000 | 12% | 720 | 8% | 5,280 | 7.2 | 9.0 | -1.8 |
Source: Helix Assurance case file
Additional InformationAsk for dataInterviewer’s data
Share Exhibit 2 and remind the candidate:
- SIU capacity is 40 investigators x 150 cases = 6,000 cases a year
- Hiring an investigator costs about $110,000 a year, fully loaded
- An SIU referral adds about 10 days to the claim for the customer, whether or not they turned out to be honest
Try it first, then checkCheck my answerModel answer
Marginal analysis: moving from top 2% to top 5%
- Extra frauds caught: 2,250 - 1,350 = 900
- Extra net value: $11.25M - $9.0M = $2.25M
- Extra honest customers delayed 10 days: 5,250 - 1,650 = 3,600
- Extra capacity needed: 7,500 - 6,000 = 1,500 cases, or 1,500 / 150 = 10 investigators, about $1.1M a year
- Net gain after hiring: $2.25M - $1.1M = about $1.15M, or roughly $320 per extra honest customer delayed
Top 10% is clearly worse: net value falls to $4.5M, and more than 12,000 honest customers are delayed.
Recommendation: a two-tier design
- Top 2%: full SIU investigation (3,000 cases, well within capacity). Net value about $9.0M
- Next 3% (the 2%-5% band): a one-day desk review by trained adjusters using the model's reasons (for example, a mismatch between photos and invoice, or a recent policy change). Escalate only the clear cases to SIU
- Use the freed SIU capacity (about 3,000 cases) for organised-fraud rings and bodily-injury claims, where fraudulent amounts are largest
- Retrain the model every quarter using investigation outcomes, and test that flag rates do not differ unfairly across customer groups
This captures most of the value of the top-5% option without hiring and without making thousands of honest customers wait ten days.
What the interviewer is looking forInterviewer’s viewInterviewer’s view
The top-5% option has the highest net value, so a candidate reading only the last column will choose it. But it needs 7,500 investigations against a capacity of 6,000, and it delays more than three times as many honest customers as the top-2% option. A strong candidate works out the marginal return of moving from 2% to 5%. An excellent candidate breaks the choice open: send the top 2% to SIU and give the next band a quick desk review.
So What? cascade:
- Level 1: top 5% has the highest net value ($11.25M)
- Level 2: going from 2% to 5% adds only $2.25M, needs about 10 more investigators (about $1.1M), and delays 3,600 more honest customers
- Level 3: a hard threshold is the wrong design. Match the depth of each check to the risk level, so honest customers in the grey zone are cleared in a day, not ten
Question 4Numeracy
If the same model sends simple, low-risk damage claims to a no-adjuster "fast track", what is the annual value, and what happens to Helix's average repair cycle time?
Hint · Numeracy
Write the formula before you plug in numbers, keep units and zeros explicit, and sanity-check the order of magnitude at the end.
On the fast track, the estimate is written from the customer's photos and payment is made automatically, with no adjuster.
Additional InformationAsk for dataInterviewer’s data
Share the following:
- Physical-damage claims (collision plus comprehensive) are 90,000 a year
- The model classifies 40% of them as low-risk and simple enough for the fast track
- Adjuster handling cost: about $900 per claim today, about $250 on the fast track
- Cycle time for fast-tracked claims falls from 26 days to 10 days. Other claims stay at 26 days
- 35% of these claimants use a rental car, at about $45 a day
- Without an adjuster, Helix expects to overpay by about 3% on fast-tracked claims, whose average severity is $3,000
Try it first, then checkCheck my answerModel answer
Step 1: Claims fast-tracked
- 90,000 x 40% = 36,000
Step 2: Handling savings (LAE)
- 36,000 x ($900 - $250) = 36,000 x $650 = $23.4M
Step 3: Rental-car savings
- Days saved per claim: 26 - 10 = 16
- 36,000 x 35% = 12,600 claimants with a rental car
- 12,600 x 16 days x $45 = $9.07M
Step 4: Extra leakage
- 36,000 x $3,000 x 3% = -$3.24M
Step 5: Net value
- $23.4M + $9.07M - $3.24M = about $29.2M a year, about 1.8 points of combined ratio ($29.2M / $1,600M)
Step 6: New average cycle time for repairable claims
- 40% x 10 days + 60% x 26 days = 4.0 + 15.6 = 19.6 days, close to the industry figure of approximately 19.3 days
What the interviewer is looking forInterviewer’s viewInterviewer’s view
A good candidate adds the handling and rental savings. A strong candidate subtracts the extra leakage from paying claims without an adjuster. An excellent candidate works out the new blended cycle time, compares it with the industry figure, and notes that the fast track should exclude any claim with a high fraud score, so the two uses of the model support each other.
Question 5Synthesis
Summarise your recommendation for the Chief Claims Officer, including the expected effect on the combined ratio.
Hint · Synthesis
Answer first: the recommendation, two or three reasons with numbers, then risks and next steps.
Try it first, then checkCheck my answerModel answer
Recommendation: build one claims-triage engine that fast-tracks simple, low-risk claims and sends only the riskiest claims to investigators.
| Initiative | Annual value | Combined-ratio impact |
|---|---|---|
| Fast track for 40% of physical-damage claims | about $29.2M | about 1.8 pts |
| Fraud triage (top 2% to SIU, next 3% desk review), vs. current rules' -$1.8M | about $10.8M | about 0.7 pts |
| Total | about $40M | about 2.5 pts |
(Fraud triage: $9.0M net from the top-2% band minus the current -$1.8M = $10.8M, before any extra gain from desk reviews.)
Why
- It pays back the $25M budget in well under a year
- It cuts the repair cycle from 26 to about 20 days, in line with the industry
- It turns the fraud programme from a net cost into a profit, while delaying far fewer honest customers
What it does not do: the combined ratio moves from about 104.7% to about 102%, still well above the industry's approximately 93 to 95. Helix must also review pricing adequacy, especially for bodily injury, where industry severity rose approximately 10% in 2025.
Plan
- Months 0-6: launch the fast track in two states. Add photo-quality checks in the app. Set up model governance (explainability, bias testing)
- Months 6-12: roll out fraud triage, train desk reviewers, move SIU toward fraud rings and bodily-injury claims
- Months 12-18: extend to all states, retrain models quarterly, and combine with a pricing review
Risks: regulator challenge to automated decisions (keep a human review for any denial), photo estimates missing hidden damage on newer vehicles (exclude cars with complex safety sensors from the fast track), and adjuster resistance (redeploy adjusters to complex and injury claims).
What the interviewer is looking forInterviewer’s viewInterviewer’s view
Strong answers lead with the combined-ratio impact and are honest that data alone does not close the gap to the industry. Excellent candidates sequence the programme: fast track first (quick, visible customer benefit), fraud triage second (needs investigator training and model governance), and they state what else Helix must do, especially rate adequacy.
Data Sources
Market facts in this case come from public sources. Figures are rounded for interview math. Helix Assurance and all company figures are fictional.
| Fact used in the case | Publisher | Title | Year | URL |
|---|---|---|---|---|
| 2024 average claim severity: bodily injury approximately $28,278, property damage approximately $6,770, collision approximately $5,489, comprehensive approximately $2,306; average auto expenditure approximately $1,282 (2023) | Insurance Information Institute (Triple-I) | Facts + Statistics: Auto insurance | 2026 | https://www.iii.org/fact-statistic/facts-statistics-auto-insurance |
| US private passenger auto combined ratio approximately 95.3 in 2024 | Carrier Management | 2024 P/C Insurance Combined Ratio: Best in More Than a Decade | 2025 | https://www.carriermanagement.com/news/2025/05/13/275145.htm |
| Personal auto combined ratio projected at approximately 92.7 for 2025 (S&P Global Market Intelligence) | Carrier Management | Good Times for U.S. P/C Insurers May Not Last; Auto Challenges Ahead | 2026 | https://www.carriermanagement.com/news/2026/01/06/283094.htm |
| Average repair cycle time for repairable vehicles approximately 19.3 days in 2025, down from 22.3 days | J.D. Power | 2025 U.S. Auto Claims Satisfaction Study | 2025 | https://www.jdpower.com/business/press-releases/2025-us-auto-claims-satisfaction-study/ |
| Bodily-injury severity up approximately 10.3% year on year; total-loss frequency a record approximately 23.1% of claims | CCC Intelligent Solutions | CCC Crash Course 2026 Report Finds Higher Severity and Record Total Loss Frequency | 2026 | https://ir.cccis.com/news-releases/news-release-details/ccc-crash-course-2026-report-finds-higher-severity-and-record |
| Fraud occurs in about 10% of P&C insurance losses (industry rule of thumb); US insurance fraud approximately $308.6B a year across all lines | Coalition Against Insurance Fraud | Insurance Fraud Statistics (figures from The Impact of Insurance Fraud on the U.S. Economy, 2022) | 2022 | https://insurancefraud.org/fraud-stats/ |
Free in the DrillCase case library · Editorial policy