Verisk Analytics: Who should own the next layer of insurance AI?
Download case PDF (6 pages)Request teacher’s pack
Decision setting: 26 February 2025. This case uses information publicly available by 11:59 p.m. Eastern Time on that date.
A new product, an old strategic question
Verisk sells data, analytics, and software to insurers. Its products support decisions in underwriting, claims, catastrophe risk, and related workflows. Verisk says these products combine proprietary and contributory data, insurance expertise, statistical models, analytics, and software, often embedded in customers’ workflows. In 2024, it served all of the 100 largest U.S. property and casualty (P&C) insurers for the lines in which it competes.¹
Generative AI raises both an opportunity and a threat for a business built around information and decisions. It may help insurers find facts in large records, handle routine work, and make better use of expert staff. It may also reproduce or substitute for some information products, while introducing errors, privacy and security exposure, and questions about who is accountable for an output. Verisk’s 2024 Form 10-K identifies both the possibility that generative AI could reduce the relevance of existing products and risks from Verisk’s own use of AI.¹
Verisk had already made a product choice. In February 2024 it announced an automated medical-record summary feature in Discovery Navigator, its platform for P&C claims professionals. The feature used Amazon Bedrock, which provides API access to foundation models. Verisk said the existing Discovery Navigator platform had been “up to 90% faster” than manual record review with “up to 95% accuracy.” Those were company claims about platform performance; the release did not measure the new feature’s incremental effect or disclose adoption, error rates, customer savings, or product revenue.²
On 26 February 2025, Verisk reported that 2024 revenue grew 7.5% to $2.882 billion and adjusted EBITDA grew 9.9% to $1.576 billion. Claims revenue rose 8.7% to $857 million. Verisk attributed claims growth to property-estimating and anti-fraud solutions—not to generative AI. Management said it would continue investing in innovation and technology while returning cash to shareholders. The company did not report generative-AI revenue, development costs, usage, or customer outcomes.³
A decision about the next tranche
Imagine you are advising Verisk’s executive leadership on the next stage of AI product investment. This is a hypothetical assignment; public information does not establish that a particular funding decision was pending. Verisk is actively adding generative-AI features to its services, but the 10-K does not name the projects or their budgets.¹ Your task is to recommend how Verisk should scale AI across insurance workflows and what evidence should govern further investment.
Three approaches are plausible:
- Build more of the stack in-house. Develop more model and application capability under Verisk’s control. This could give the company tighter control over performance, data handling, and product direction. It would require talent and investment in a rapidly changing technical field.
- Partner first. Use external models and cloud infrastructure broadly, concentrating Verisk’s effort on insurance data, use cases, and distribution. This could speed experimentation and limit duplicative development. It would increase dependence on providers and make contract rights, portability, cost, and reliability important.
- Use a hybrid. Keep control of the insurance-specific application, permitted data, evaluation, governance, and customer workflow, while partnering for general-purpose model or cloud capabilities where that is advantageous. This could balance control and speed, but would add integration and supplier-management work and may not prevent partners from moving closer to the customer.
The options are analytical alternatives, not descriptions of an announced Verisk plan. Any proposed budget, pilot size, customer cohort, or target in your recommendation should be stated as an assumption.
Assets and constraints
Verisk’s position includes deep relationships and workflow access. Its 10-K describes regular customer advisory panels and direct sales teams, and says that alliances can integrate other providers’ offerings so clients obtain information from a single source and fit it more easily into existing workflows.¹ Hosted subscriptions made up 81% of 2024 Insurance segment revenue; these generally run one to five years and renew automatically. That recurring base could help a new feature reach existing customers, but the filing does not report AI-feature adoption, renewal effects, or willingness to pay.¹
Data is not a single asset with uniform rights. Verisk says it relies on data from customers, government services, public records, and other external sources. It generally does not own the information in its repositories. Some supplier agreements are short term; sources may raise prices, stop contributing, or grant competitors exclusive access. The filing does not say which datasets may be used to train or expose models.¹
Infrastructure creates a related choice. Verisk had used Amazon Bedrock for Discovery Navigator, and its filing says that migration to cloud-based infrastructure has concentrated systems among a smaller number of providers. Verisk says it cannot easily switch cloud providers and identifies outage, pricing, contract lock-in, regulatory, and other risks. The filing does not identify Bedrock as the specific source of this company-wide concentration.¹ ²
For a knowledge business, quality and trust are strategic concerns as well as technical requirements. Verisk says its solutions support insurer decisions and that errors or delays could hurt market acceptance, increase liability, and damage its reputation. The 10-K also identifies possible privacy, security, bias, compliance, intellectual-property, and cybersecurity issues in using generative AI.¹
Exhibits
Exhibit 1. Verisk’s disclosed business at a glance
| Measure | FY2024 / company description |
|---|---|
| Revenue | $2,881.7m; 7.5% reported growth; 7.1% organic constant-currency growth |
| Adjusted EBITDA | $1,576m; 9.9% growth |
| Operating cash flow | $1,144m |
| Free cash flow | $920m |
| Claims revenue | $857m; 8.7% reported growth, 7.7% organic constant-currency growth |
| Underwriting revenue | $2,025m; 7.0% reported growth, 6.8% organic constant-currency growth |
| 2024 Insurance segment revenue from hosted subscriptions | 81% |
| 2024 revenue from solutions to U.S. P&C primary insurers | Approximately 70% |
Source: Verisk 2024 Form 10-K and 26 February 2025 results release. “Organic constant currency” and adjusted EBITDA are company-defined non-GAAP measures. Results are for the company or segment, not for AI products. Claims growth was attributed to property estimating and anti-fraud solutions.¹ ³
Exhibit 2. One announced GenAI product
| Feature | What was disclosed |
|---|---|
| Workflow | Medical-record review for P&C bodily-injury claims |
| Product | Discovery Navigator |
| Announced capability | Automated summary of treatment data and medical information |
| Foundation-model access | Amazon Bedrock API |
| Potential benefit described by Verisk | Faster review and more efficient claim handling |
| Reported performance language | “Up to 90% faster” and “up to 95% accuracy” for the platform relative to manual review |
| Incremental feature performance, usage, cost, and realized outcome | Not disclosed |
Source: Verisk’s 8 February 2024 release. Performance figures are company claims about Discovery Navigator; they are not an independent evaluation or a measure of the new GenAI feature’s incremental impact.²
Exhibit 3. Risk and control questions in the 10-K
| Issue | Company disclosure |
|---|---|
| Data access | Verisk depends on customer, government, public-record, and other data; it generally does not own repository information. Supplier agreements can be short term. |
| Cloud dependence | Systems have migrated toward fewer large cloud infrastructure suppliers; Verisk says switching providers is difficult. |
| Product substitution | Generative AI could duplicate or reduce the relevance of some existing products. |
| New-product risks | Verisk identifies ethical, compliance, privacy, bias, intellectual-property, security, and reputational risks from its own AI use. |
| Product quality | Errors or delays can harm acceptance, reputation, revenue, and increase liability or service costs. |
Source: Verisk 2024 Form 10-K, risk factors. Company-identified risks are not evidence that any event has occurred.¹
Questions for discussion
- Which approach—build, partner, or hybrid—would you recommend for Verisk’s next stage of insurance-workflow AI? What assumptions drive your choice?
- Which assets should Verisk control directly, and which capabilities could it obtain through contracts or alliances? How do data-use rights affect your answer?
- What should Verisk measure in an initial deployment? Design a test that distinguishes feature impact from changes in claim mix, staffing, or other workflow conditions.
- What contract and technical conditions would preserve Verisk’s options if a model or cloud provider changed its terms, quality, or reliability?
- What result would cause you to expand, redesign, or stop the effort?
Selected sources
- Verisk Analytics, Inc., 2024 Form 10-K, filed 26 February 2025, accession 0001437749-25-005160, SEC filing, Item 1 (business and strategy, SEC HTML lines 212–263; product development and alliances, lines 494–501); risk factors (data dependence, lines 701–711; cloud and AI, lines 847–870); revenue model (lines 1267–1273).
- Verisk, “Verisk Launches Generative AI Tool to Support Faster Insurance Claim Processing,” 8 February 2024, official release, pp. 1–2.
- Verisk, “Verisk Reports Fourth-Quarter 2024 and Full-Year 2024 Financial Results,” 26 February 2025, results release, lines 65–78, 82–114, and 143–165.