The short answer
Insurance and banking adopt AI at similar rates, and their transformations get stuck in different places: insurers in claims and underwriting work full of exceptions, documents and manual handoffs, banks in model risk governance, legacy core systems and controls around regulated decisions. Horizon helps both by interviewing the teams that run each process with AI to show where work actually breaks before AI is applied; at La Segunda, an insurer, every person on a team handling more than 700 active cases took part.
The UK's financial regulators found that 95% of insurers and 94% of international banks surveyed already use AI. Adoption is no longer the gap. What differs is the shape of the work AI has to change, and the rules around it.
Key takeaways
- Insurers and banks use AI at almost the same rate; their constraints differ.
- Insurance work is exception-heavy and document-heavy: claims, underwriting and policy servicing depend on judgment and handoffs.
- Banking AI meets formal model risk governance: validation, documentation and inventories apply to AI models too.
- The EU AI Act treats credit scoring and life and health insurance pricing as high-risk; back-office process optimization is outside that category.
- Both industries get the fastest value from low-risk operational processes, mapped from the people who run them.
- Horizon maps the process from the people who run it in both industries, with every finding traceable to its source, the kind of documentation regulated industries need.
How does AI transformation differ between insurance and banking?
| Insurance | Banking | |
|---|---|---|
| Core processes AI touches | Claims intake and settlement, underwriting, policy servicing, broker and partner operations | Customer onboarding and KYC, credit decisions, payments and reconciliation, fraud, back office |
| Where work gets stuck | Exceptions in claims, documents arriving incomplete, handoffs between adjusters, brokers and partners | Legacy core systems, controls layered over years, manual reconciliation between systems |
| Nature of the work | Case-based, judgment-heavy, many variants per product and region | Transaction-based, high volume, strongly standardized |
| Data situation | Unstructured documents, photos, emails and forms from outside parties | Structured transaction data in core systems, fragmented across legacy platforms |
| Regulatory focus on AI | Fair pricing and claims decisions; supervisory guidance on insurers' AI use | Model risk management; creditworthiness and fraud decisions |
| High-risk under the EU AI Act | Risk assessment and pricing for life and health insurance | Creditworthiness evaluation and credit scoring of individuals |
| Typical change challenge | Adjusters and underwriters fear losing judgment calls to automation | Many control owners, long approval chains for any model in production |
| Fast first target | Claims intake and exception handling | Reconciliation, onboarding back office and internal process optimization |
What do insurance and banking have in common?
Both industries are already deep into AI use. The Bank of England and the FCA surveyed 118 UK financial services firms in 2024: 75% were already using AI and another 10% planned to within three years. Insurance had the highest adoption at 95%, followed by international banks at 94%.
The use cases also overlap. Across firms, the most common current uses were optimizing internal processes (41%), cybersecurity (37%) and fraud detection (33%). A third of AI use cases were third-party implementations, and 46% of firms reported only a partial understanding of the AI technologies they use.
That last figure is the common risk. When firms deploy AI they do not fully understand, on processes they have not mapped, they automate whatever the process currently does, including its workarounds. Both industries share the first step: establish how the process actually runs before deciding where AI belongs. For a scoring method, see how to prioritize AI use cases by ROI.
Where does AI transformation get stuck in insurance?
Insurance work is organized around cases, and cases vary. A claim moves between intake, adjusters, experts, brokers, repair partners and payment, and each handoff can stop it. The documentation often arrives from outside the company, incomplete or in different formats.
Four patterns show up repeatedly:
- Exceptions dominate the effort. The standard path may cover most claims by count, and a minority of complex cases can consume most of the hours.
- Knowledge sits with experienced people. Adjusters and underwriters carry judgment that is rarely documented, which makes AI training data and process redesign harder.
- Partners and brokers add handoffs. Work waits on parties outside the insurer's systems, where no event log exists.
- Pricing and claims decisions are regulated. The EU AI Act classifies AI used for risk assessment and pricing of individuals in life and health insurance as high-risk. In the United States, 29 jurisdictions had adopted the NAIC Model Bulletin on insurers' use of AI systems as of November 2026, which sets governance expectations for how insurers use AI.
The practical consequence: insurers get faster value by applying AI to the operational layer first (intake, document handling, exception routing) and by mapping how experienced staff handle the hard cases before redesigning anything.
At La Segunda, an insurance company, Horizon ran AI interviews with a team handling more than 700 active cases. All nine people on the team took part, 100% of the scope, and the round produced 10 findings. Participants rated the experience 8.2 out of 10, and 90% said they would take part again, which matters in a function where people worry about automation taking over their judgment.
Where does AI transformation get stuck in banking?
Banking work is organized around transactions, and transactions run through systems built over decades. The obstacles are less about variation and more about the systems and controls around the work.
Four patterns show up repeatedly:
- Model risk governance applies to AI. US banking supervisors' guidance on model risk management (SR 11-7) defines a model as a quantitative method that turns input data into estimates, and expects independent validation, documentation detailed enough for outsiders to understand the model, ongoing monitoring and a model inventory. AI models used in decisions fall under that framework, which lengthens the path to production.
- Credit decisions are high-risk. The EU AI Act classifies AI used to evaluate the creditworthiness of individuals or set their credit score as high-risk, with fraud detection excluded from that point.
- Legacy cores fragment the data. Customer, product and transaction data sit in different systems, and staff bridge them manually.
- Controls accumulate. Each incident adds a check; few checks are ever removed. AI applied on top of an over-controlled process inherits its delays.
The practical consequence: banks get faster value where AI changes internal operations without making regulated decisions about customers, such as reconciliation, onboarding back office and internal process optimization, while the regulated use cases go through model governance in parallel.
Reconciliation is a good example of where the friction hides. At PedidosYa, a payments-heavy delivery platform operating in 15 markets, manual reconciliation between two systems in one market generated more than 1,000 discrepancies a week and took about two hours of weekly analysis, against about 40 minutes in other markets, because of a visualization error that forced line-by-line comparison. The same pattern, two systems that disagree and a person bridging them, runs through bank back offices. For more on how this kind of loss builds up between teams, see end-to-end operations visibility.
In regulated financial operations, AFAP SURA, a pension fund administrator in Uruguay with 600 employees, assessed its international transfer process, where information was spread across five systems and an error could produce a financial loss that is hard to recover. In under seven days, Horizon delivered 63 prioritized insights covering bottlenecks, repetitive tasks and operational risk points, with a pilot ROI of 186%.
Where should each industry start?
| Step | Insurance | Banking |
|---|---|---|
| 1. Pick a first process | Claims intake or one product's exception handling | Reconciliation, onboarding back office or an internal operations process |
| 2. Map it from the people who run it | Adjusters, intake teams, experts and partner coordinators | Operations analysts, control owners and the teams that bridge systems |
| 3. Separate regulated from operational use | Keep life and health pricing and claims decisions on a governed track | Keep credit and customer decisions under model risk governance |
| 4. Score the opportunities | Value, evidence, data readiness, adoption | Same criteria, plus validation effort for any model |
| 5. Pilot in one area | One product line or region | One market, entity or back-office team |
In both cases, a structured business process assessment gives the first process a baseline, and the pilot shows whether the change holds before it scales. Programs that skip the mapping step tend to fail for the reasons covered in why digital transformations fail.
Where Horizon fits
Horizon is an AI-powered continuous discovery platform. In both industries, it covers the step that comes before AI is applied: establishing how the process actually runs and where it breaks.
Discovery Cycles run AI-led interviews across every role in the process, from adjusters and underwriters to operations analysts and control owners. The Insights Dashboard ranks findings by impact and effort with traceability to the input behind each one, which supports the documentation regulated industries need. The Process Library documents the process as it runs, and the Initiatives Dashboard turns priorities into business cases with owners. Comparing teams, products or markets shows which differences are required by local rules and which are workarounds.
What sets Horizon apart in financial services is coverage with an audit trail. Traditional discovery samples 10 to 20 interviews; Horizon interviews every role in scope at the same time, so the map includes the exceptions in claims and the workarounds around controls, and each finding points back to the input behind it. At La Segunda, every person on a team handling more than 700 active cases took part.
For how the interviews work and how to design them so employees trust them, see AI employee interviews at scale.
Map one insurance or banking process from the people who run it, in one week. Run Horizon's one-week challenge: AI interviews across the process and a ranked list of what to fix first. Start the one-week challenge
FAQ
Is AI adoption higher in insurance or in banking?
They are close. In the Bank of England and FCA 2024 survey of UK financial services firms, 95% of insurers and 94% of international banks were already using AI, the two highest rates among the sectors surveyed.
Which AI use cases are high-risk under the EU AI Act in financial services?
Annex III of the EU AI Act lists AI used to evaluate the creditworthiness of individuals or set their credit score, with fraud detection excluded, and AI used for risk assessment and pricing of individuals in life and health insurance. Internal process optimization falls outside those points.
How does model risk management affect AI in banks?
US supervisory guidance on model risk management (SR 11-7) expects banks to validate models independently, document them in enough detail for outsiders to understand them, monitor them and keep a model inventory. AI models used in decisions fall under that framework, which adds time and review before production.
Where should an insurer start with AI transformation?
With an operational process full of exceptions, such as claims intake or document handling, mapped from the adjusters and intake teams who run it. Pricing and claims decisions go on a separate, governed track.
Can the same discovery approach work in insurance and banking?
Yes. Both industries need to know how a process actually runs before applying AI, and both have the knowledge with the people who do the work. What changes is the process chosen first and the governance around regulated decisions.
The bottleneck decides the starting point
Insurers and banks are adopting AI at almost the same pace. Their transformations differ in where the work gets stuck: exceptions and handoffs in insurance, governance and legacy systems in banking. In both, the first step is the same: see how the process actually runs, then decide where AI belongs.
Horizon makes that first step fast in both industries: every role in the process heard in days, and a ranked list of where AI belongs.
See it. Fix it. Lead it.