Pet insurer ManyPets reworks claims and pricing with AI, cutting processing time in half
ManyPets has embedded AI across claims processing, pricing, customer service and marketing, with concrete results: 55% of claims now pay on the same day, up from roughly half that level a year earlier. The pet insurer processes hundreds of thousands of claims annually, and the shift reflects a broader strategy to reduce friction at moments customers notice most.
Pierre du Toit, chief data officer at ManyPets, describes the work not as a generic digital overhaul but as using data and AI to improve core business operations. The company started two years ago with enough accumulated data to move beyond basic analytics into more sophisticated applications.
Claims automation through Millie
The clearest example is claims processing. ManyPets built an AI system called Millie that screens 100% of incoming claims. The models, trained on historical claims data, determine whether to approve claims as submitted, flag pre-existing conditions, calculate deductions, or route cases for manual review.
Speed matters in pet insurance for reasons beyond productivity. Claims arrive when owners are stressed-managing a sick animal and a veterinary bill simultaneously. Faster settlement improves not just efficiency but reassurance and trust. Customer feedback on Net Promoter Score and Trustpilot has reflected the change.
Around 80% of claims now settle within five days, a significant jump from earlier performance.
Pricing and profitability
ManyPets has also invested heavily in pricing models. The company built a proprietary dataset covering customers, pets, health and wellness information, giving it detailed risk assessment capability. This supports more accurate pricing and fairer customer outcomes.
The commercial impact is material. Du Toit said the pricing work has supported significant year-on-year growth in new business and helped turn the company profitable. AI here functions as an underwriting tool, not just a service enhancement.
Broader deployment across the business
AI extends beyond claims and pricing. In marketing, the company uses it to standardize tone of voice, test propositions with synthetic audience segments and support campaign development. Engineers and data scientists use AI agents to accelerate coding. Customer service teams use it to route inbound messages. Reporting systems flag anomalies and give business users faster access to insights.
The company also created a "TailMates" campaign using AI to generate video content for pets and owners, showing a more creative application of the technology.
Build versus buy
ManyPets starts with customer value rather than supplier relationships when deciding whether to build or partner. The company keeps core capabilities in-house, including policy administration, claims systems and AI model training and deployment.
AWS provides hosting and Google supports the data platform-services du Toit says are not worth replicating internally. But competitive systems stay under the company's control, preserving data ownership and long-term sustainability.
The process-first lesson
Du Toit offers a direct warning: AI does not fix broken processes. "If you bolt AI onto a badly designed workflow, you can actually make things considerably worse," he said.
ManyPets focuses first on workflow redesign, sometimes modestly, before introducing automation. The gains often came from cleaning up processes rather than grand reinvention. Internal resistance faded quickly once teams saw that AI removed repetitive, unpopular work.
Organization and training
The company uses cross-functional teams bringing together product, data and engineering around clear ownership and measurable outcomes. Training varies by role-how a back-end engineer uses generative AI differs from how a product manager, marketer or analyst uses it.
Du Toit identifies focus as the most important management lesson from the past two years. The business has no shortage of ideas, but too many potential uses for data and AI becomes a problem. Projects performed best when anchored to clearly defined metrics from the start.
What's next
Du Toit expects three developments over the next three to five years. Personal AI assistants that understand individual users and their documents will become standard. Automated model development-where AI systems iterate through modeling ideas continuously-will reduce manual work. Business users will get richer, real-time insight through smarter decision support.
Large language models like ChatGPT, Claude and Gemini could become meaningful distribution channels, he said, as general insurance distribution currently relies heavily on aggregators and Google direct marketing.
In pet insurance specifically, he expects a more connected model linking insurance, advice, triage, claims and treatment into a single customer journey.
Regulation as opportunity
Du Toit is relaxed about the Competition and Markets Authority review into veterinary pricing. ManyPets welcomes moves that make pricing clearer for consumers and more structured for insurers.
The company currently extracts cost information from veterinary invoices, a difficult process for benchmarking and cost management. Cleaner, more structured data from greater transparency would strengthen existing pricing and claims capabilities rather than disrupt them.
For more on how insurers are implementing AI capabilities, see AI for Insurance and Generative AI and LLM.
Your membership also unlocks:
