Startup Spotlight: Solva builds claims AI that checks policy terms against evidence

Why it matters

Solva is targeting a high-stakes, document-heavy workflow where speed only counts if insurers can trace a recommendation back to the policy and evidence. Its founders bring relevant operating experience, but customer-level results and the scope of adoption remain the clearest tests of the pitch.

Herman Baverud Olsson, who led AI and machine-learning work at BCG X and helped build pet insurer Lassie, co-founded Solva with Linus Malmen, a fintech founder, and Sorena Amini, an operator with experience in insurance and technology-enabled services. Their bet is that claims work needs software that can read a policy, connect it to a file of evidence and show its reasoning, rather than another general-purpose chatbot.

Solva describes its product as domain-specific AI for insurers handling complex claims. Its system is designed to identify missing evidence, compare claim details with policy terms, flag potential fraud or policy breaches, and produce recommendations linked to source material. Solva says its software can run alongside existing claims systems, leaving insurers with their current core platforms while automating or assisting parts of the work around them. Y Combinator's company profile lists Solva in its Summer 2025 batch and describes it as active; the listing is a profile of an existing company, not a new launch.

Three founders, three parts of the problem

Olsson's route to Solva runs through insurance operations, financial services and applied AI. Before co-founding Solva, he worked at Morgan Stanley and led AI and machine-learning transformations for financial-services clients at BCG X. He then helped build Lassie, a Swedish pet insurer. That combination gives him a direct line to two challenges Solva is selling against: the practical friction of insurance work and the difficulty of putting machine-learning systems into consequential workflows.

Public figures for Lassie's scale vary across Solva and Y Combinator materials. A Y Combinator launch profile says Olsson was part of scaling Lassie to $50 million. Solva's current founder page says $100 million, while the current YC profile says $120 million in ARR and $120 million raised. Those statements do not define their measures or time periods consistently, so they should be read as career claims rather than comparable, independently verified figures. The variation is a useful reminder that founder bios often compress a company's trajectory into a single number.

Malmen brings experience building financial infrastructure. Solva and YC's profile describe him as a co-founder of Zaver, a European fintech company that processed more than EUR1 billion annually under an EBA-CRD license and served clients including Porsche, Volkswagen and Loewe. That background is relevant to a product that must connect decisions to evidence and operate within an insurer's controls. In this kind of software, the hard part is rarely producing fluent text. It is fitting the system into a workflow where policy language, claim documents, exceptions and audit requirements all matter at once.

Amini brings operating and financial-services experience. Solva and YC's profile say she helped scale KeyTo from $60 million to $250 million while at Fidelio Capital, and worked on insurance and InsurTech at Citi before taking growth and product roles at Talentech, formerly ReachMee. Together, the three founders make a more specific pitch than a generic claim to AI expertise: one has built an insurer, one has built financial technology, and one has worked on scaling technology-led businesses and advising insurance companies.

The claim file is the product surface

Solva's launch materials call the product "Harvey for insurance" and describe verifiable AI agents that automate claims tasks and help prevent wrongful payouts. In practice, its stated workflow is about assembling and checking information: determine what evidence is missing, identify where the facts may conflict with policy terms, flag issues that merit further review, and return recommendations with citations to source material. Solva's product description positions the software as a system for complex claims rather than an open-ended assistant.

That focus reflects a real product constraint. A claim decision can depend on several documents, coverage conditions, internal guidelines and exceptions. A system that summarizes one document may save time, but an insurer needs to know which source supports a recommendation and where a human adjuster should examine the file. Solva's materials emphasize cited recommendations and auditability, and its CTO described the company's approach to supporting complex claims at a Google Cloud Summit event in May 2026. The commercial distinction is practical: a useful tool must fit the existing claims process, while the insurer remains responsible for the decision and its consequences.

Solva has also begun framing its ambition around quality assurance, alongside speeding up individual claims. In a Solva post about claims QA, Solva argues that insurers often review a small sample of claims after decisions have already been made. Its proposed alternative is to analyze more claims while there is still time to intervene. The post relays an example from an unnamed top-10 carrier that Solva says identified about $22 million in recoverable opportunity in California and Florida. That is Solva's account of a conversation, not a named customer case study or independently audited result. Still, it shows the economic thesis: software that catches a coverage or handling issue before payment could affect claim outcomes, rather than merely reducing the time spent on paperwork.

The promise requires a careful balance. Insurers have financial reasons to identify incorrect or excessive payouts, while policyholders need valid claims paid fairly and on time. Software that flags potential fraud or policy breaches can help teams scrutinize files, but the same capability can steer attention toward denials if its recommendations are treated as answers rather than prompts for investigation. Solva says its outputs cite source material; its event materials also describe human oversight. Those controls matter most when the software is wrong, the policy is ambiguous or the record is incomplete.

Metrics without the measurement detail

Solva's homepage advertises a 2.1% loss-ratio improvement, 18-times-faster quality-assurance audits and three-times-faster claim-cycle times. These are company-reported performance figures. The public materials do not specify the customer sample, measurement period, baseline, claim type or whether the loss-ratio figure means a relative percentage or percentage points. The figures describe a potential outcome, but they do not yet show how broadly it has been reproduced.

Customer proof is similarly limited in the public-facing materials. Solva's homepage includes short comments attributed to roles such as a specialty insurer's CTO, a nationwide third-party administrator's claims director and a P&C carrier's senior claims manager. The quotes are not paired with named organizations, deployment dates or detailed before-and-after results. That gives a reader some evidence that Solva is talking to industry practitioners, while leaving customer adoption and commercial scale difficult to assess from the materials alone.

Solva's trust center states that Solva Technology is SOC 2 Type II and ISO 27001:2022 certified. The company's privacy policy says it was "preparing a SOC 2 Type II report," so the two public pages give conflicting accounts of its SOC 2 status. Both describe ISO 27001 certification. These are company-reported compliance claims; the trust center says detailed documentation is available on request. For an insurer assessing a product that handles sensitive claim records, security documentation and the ability to evaluate controls are part of the sale, not housekeeping details.

From accelerator profile to enterprise sale

Y Combinator lists Solva's team size as 15, and Solva identifies offices in New York and Stockholm. Solva's materials say its founders have built or operated insurance businesses representing $100 million in gross written premium and EUR1 billion in payouts. Those figures, like the product metrics, are company claims. Their relevance is that they point to practical exposure to the work Solva wants to automate; they do not establish how many customers use the current product.

Solva's financing record also has more than one layer. Solva's June 2025 announcement confirms Y Combinator invested, without naming an amount. Dealroom's company entry records a $125,000 YC seed investment in June 2025 and a SEK57 million seed round in August 2026 involving SV Angel, Y Combinator and Paul Graham. The larger round is a database-reported entry, rather than a funding announcement in the company materials reviewed here. Solva's current website identifies Y Combinator and SV Angel among its backers; it does not publish a valuation or confirmed total raised.

Solva has described participation in industry forums with Talanx, HDI Group and Warta, and its CTO presented the product at a Google Cloud event. Those are signs of activity around the industry, not proof that those organizations are customers. For an enterprise software business, the consequential milestones will be named deployments, measurable results across a defined set of claims and evidence that the product can accommodate different policy rules and claims systems without a costly bespoke build each time.

That is a demanding path, and it is where Solva's founder mix earns its keep. Olsson has seen insurance operations from inside a carrier, Malmen has built financial technology, and Amini has worked on scaling and advising regulated businesses. Their stated product thesis is grounded in familiar operational friction: documents, exceptions, old systems and decisions that have to be defensible later. Solva is asking insurers to trust software with parts of that work. Whether it becomes an enduring infrastructure layer will depend on outcomes its public metrics have not yet documented in enough detail: fewer avoidable errors, faster handling and recommendations that claims professionals can verify and act on.