What Lemonade actually proved about the insurance industry
When Lemonade launched its renters insurance product in New York in 2016, the coverage was roughly conventional. What was new was the operating model: a peer-to-peer pool, a flat management fee that removed the insurer's incentive to deny claims, and a social-good mechanic where unclaimed funds were donated to causes customers chose. Behavioural economist Dan Ariely's involvement gave the design a clear intellectual lineage. The resulting product could pay simple claims in seconds.
A decade later, the durable lesson is not the specific commercial model. Lemonade has had years of mixed underwriting performance and has moved past the purity of its launch positioning. The durable lesson is engineering. The company showed what becomes possible when an insurance carrier is rebuilt as a software company: a single data model of the customer, an API-first core, a machine learning underwriting and claims engine, and a user experience engineered by product managers rather than approved by committee.
Why legacy insurance software became the main constraint
Most carriers are running on policy administration systems, claims platforms, and billing engines that were specified decades ago. Product changes require vendor tickets. Quote, bind, and issue workflows involve extract-transform-load between systems that do not know about each other. Claims data sits in one platform, customer data in another, billing in a third. The result is that a product manager cannot iterate, a data scientist cannot compose a reliable feature set, and a CIO cannot give a straight answer about how much a small product change will cost.
This is why the new entrants have an asymmetric advantage. They are not shipping better AI on the same infrastructure. They are shipping new infrastructure and putting AI on top. The incumbents that will stay relevant are the ones who are doing the unglamorous work of replacing or wrapping their cores, consolidating their data, and rebuilding their product and operational experience as software.
The architecture patterns that define modern insurance software
Across the InsurTech platforms Scrums.com has worked with, four patterns consistently show up in the estates that scale.
A modern policy administration system
The policy admin system is the system of record for contracts, coverage, endorsements, and renewals. Modern platforms such as Duck Creek, Guidewire Cloud, Socotra, and Insly expose their capabilities through APIs and treat product configuration as data rather than as custom code. For a new carrier or MGA, picking a modern policy admin is the single highest-leverage architectural decision. For an incumbent, migrating from a legacy policy admin is typically a multi-year programme with significant operational risk, which is why the strangler pattern (new products on the new platform, legacy products remaining until end of life) has become the default migration strategy.
A product configuration layer that non-engineers can use
The insurance industry ships product variants constantly: new coverages, new territories, new limits, new exclusions. Carriers that require engineering to edit a rater or reconfigure a product cycle times of weeks. Carriers that have pushed product configuration into a domain-specific tooling layer used by actuaries and product managers cycle in days. The engineering investment is substantial, but the competitive outcome compounds.
A claims platform engineered for straight-through processing
The Lemonade three-second claim was not magic. It was a decision engine with clear rules for the subset of claims that could be straight-through processed, combined with escalation paths for the rest. Every modern claims platform now has a version of this. The differentiator is the share of claims that can be safely auto-adjudicated, the explainability of the decisions, and the quality of the customer experience when escalation is required. Generative AI is compressing the non-auto-adjudicated claim cycle time as well, by drafting adjuster notes, summarising damage evidence, and accelerating reserve setting.
A unified data platform
Every high-value capability downstream, underwriting pricing, fraud detection, customer lifetime value analysis, catastrophe modelling, regulatory reporting, depends on a consolidated view of the policy, the claim, the customer, and the external signals around them. Carriers that still copy data between system silos struggle with every one of these capabilities. Carriers that have invested in a modern data platform, a feature store, and clean domain ownership can compose new capabilities quickly.
Where AI is creating measurable value in insurance right now
Insurance has some of the richest AI use cases of any regulated industry, because the industry has always been data-heavy.
Underwriting and pricing models have used machine learning for years. What is new is the use of richer, unstructured data (imagery, IoT telematics, connected home sensors, satellite data) inside models that can handle it. Generative AI assists the human underwriter for complex risks by summarising submissions, drafting referrals, and surfacing inconsistencies.
Claims handling benefits from both ML and generative AI. Image and video analysis speeds first-notice-of-loss triage for motor and property. Audio analysis flags call sentiment and potential fraud signals. Generative AI drafts adjuster narratives, customer communications, and reserves-setting memos.
Fraud detection remains a strong ML application. The shift has been from rules-based scorecards to network analytics that identify organised fraud rings and anomaly detection that catches opportunistic fraud at scale.
Customer operations is the fastest-growing use case. RAG-based assistants handle a meaningful share of tier-one customer queries for modern carriers, with human handoff for anything outside their grounded knowledge base.
Regulatory context that shapes what you can ship
Insurance regulation has moved quickly on AI. The EU AI Act classifies insurance pricing systems as high-risk, requiring documented risk management, data governance, technical documentation, and post-market monitoring. The NAIC in the United States has issued model bulletins on AI use in insurance that multiple state regulators have adopted, focused on governance, testing, and unfair discrimination. UK and Canadian supervisors have issued comparable expectations.
Operational resilience obligations also apply. Insurers subject to DORA in the EU (since January 2025) carry full ICT risk management, incident reporting, and third-party oversight requirements, with EIOPA coordinating implementation across the insurance sector. Equivalent expectations are rising in UK, Australian, and APAC supervisory rule-books. These are not compliance afterthoughts. They are architectural constraints on where AI and critical workloads can run.
Build, buy, or partner: the insurance-specific calculus
Insurance carriers face the same three choices as banks, with a specific twist. The policy admin, claims, and billing cores are dominated by a short list of vendors, and for most carriers the right answer is to buy the core and build around it. Differentiation lives in the layers above: product configuration, customer experience, data platform, underwriting intelligence, and claims automation.
Partnering with a software development firm that has shipped regulated InsurTech is often the fastest way to close the gap. The selection criteria matter: look for partners with production experience in insurance, fluency in the vendor ecosystem (Duck Creek, Guidewire, Socotra, and similar), a track record of passing regulatory audits, and a clear stance on IP ownership and data residency.
What Lemonade still gets right, and what any modern insurer needs
The specific Lemonade commercial model has evolved, and the company's financial results have varied. The operating posture remains instructive. The insurers that will compound value in the next decade share a short list of traits.
- They treat software as the product, not as a cost centre. Engineering leadership sits on the executive team and carries P&L accountability.
- They instrument everything. Product decisions are made from data, not from quarterly working-group consensus.
- They separate product configuration from platform change. Actuaries and product managers can ship variants without an engineering sprint.
- They engineer for claims experience, not just for pricing discipline. Customers judge insurers on the moment of claim, and that moment is a software-delivered experience.
- They treat AI as a platform capability, not as a feature. The evaluation harness, the governance layer, and the observability stack are shared infrastructure, not project-by-project afterthoughts.
How Scrums.com partners with insurers and InsurTechs
Scrums.com builds production-grade software for insurance carriers, brokers, MGAs, and InsurTech startups as part of our insurance software development services. Our engineering teams bring experience with Guidewire, Duck Creek, and Socotra ecosystems, modern policy admin configuration, claims automation, underwriting AI, and customer-facing builds engineered for the security, resilience, and regulatory demands of the sector.
If you are modernising your insurance estate or launching a new InsurTech proposition, start a project with us.
FAQ
What did Lemonade actually change about insurance?
Lemonade's commercial model (a flat management fee with unclaimed funds donated to charity) was the headline at launch. The durable change was engineering: a single customer data model, an API-first core, and a claims automation platform that proved what becomes possible when an insurer is rebuilt as a software company. The lesson has outlasted the specifics of the launch business model.
Should an insurer replace or wrap its legacy policy admin system?
Both are valid. Full replacement offers the cleanest long-term outcome but is a multi-year, high-risk programme. The strangler pattern (new products on a modern platform while legacy products remain until end of life) is the more common migration strategy for incumbent carriers. The right choice depends on execution appetite and competitive pressure.
Where does AI deliver the most value in insurance?
Underwriting and pricing, claims handling (triage, auto-adjudication, drafting), fraud detection, and customer operations. The winning carriers treat AI as shared platform infrastructure with evaluation, governance, and observability baked in, rather than as a feature-by-feature project.
How do the EU AI Act and DORA affect insurers?
The EU AI Act classifies insurance pricing as high-risk, requiring documented risk management, data governance, technical documentation, and post-market monitoring. DORA, in force since January 2025, adds operational resilience obligations that extend to AI vendors, model hosting, and other critical ICT third parties.
Should we build InsurTech software in-house or partner with an external firm?
Most carriers buy the core and build differentiation in the layers above. External software partners with production insurance experience can accelerate that work while in-house teams retain strategic ownership. The selection criteria that matter are domain experience, vendor ecosystem fluency, and regulatory track record.
