Having tracked insurance AI adoption metrics across 20+ carrier earnings calls and vendor reports since 2025, the conference consensus that architecture, not algorithms, is the bottleneck confirms a pattern visible in the gap between pilot results and enterprise-wide productivity data. Celent’s third annual GenAI survey found that 48% of global insurers now run generative AI in production, yet Conning’s 2025 C-suite survey revealed that only 7% have scaled AI across their entire enterprise. That 41-point chasm does not reflect a shortage of capable models. It reflects a data infrastructure problem that every speaker at Insurtech Insights 2026, from Allianz’s CTO to Snapsheet’s CEO, identified as the real constraint.
This article synthesizes the conference’s central data architecture thesis with the specific metrics, award decisions, and competitive dynamics that emerged across two days of programming, and examines what the shift means for actuarial workflows that depend on the same fragmented data foundations carriers are now scrambling to fix.
The 80% Problem: Adjusters as Switchboard Operators
The conference’s most frequently cited statistic came from the “Claims Without Friction” panel, where Andy Cohen, CEO of Snapsheet, described a claims operation where adjusters spend 80% of their time functioning as “switchboard operators,” shuttling data between disconnected systems rather than performing the judgment work they were hired to do. The panel, which included Pete Piotrowski of Hippo, Michelle Raue of Raue Strategic Advisory, and James Benham of the InsurTech Geek Podcast, agreed on a shared goal: automate the data-routing tedium thoroughly enough to return adjusters to old-school claims work focused on understanding the claim, understanding the claimant, and making the judgment calls only experience can support.
Cohen’s framing was deliberate. The ambition, he said, is to make adjusters “superhuman,” not to replace them. That framing matters because it positions AI investment as a productivity tool rather than a headcount reduction strategy, a distinction that carries weight with regulators increasingly scrutinizing AI’s impact on insurance employment. The 80% figure also reframes the ROI calculation for claims AI: the prize is not a faster algorithm but the recovery of four out of every five hours currently wasted on data logistics.
The stat aligns with Microsoft’s February 2026 analysis of insurance AI bottlenecks, which found that adjusters typically require one to three days just to gather, read, and interpret documents before beginning substantive claims work. In a year when more than 30 million personal auto claims were filed in the U.S. alone, that document-gathering lag translates into billions of dollars in loss adjustment expense. Sedgwick’s Sidekick Agent, developed in collaboration with Microsoft, demonstrated that targeted AI intervention can improve claims processing efficiency by more than 30%, but that improvement requires clean data flowing through connected systems. When the systems are disconnected, even the best AI agent spends its compute cycles on data extraction rather than decision support.
From tracking cross-industry claims AI data, the pattern is consistent: carriers that achieved the largest cycle-time reductions (75% in some deployed lines) all completed data infrastructure consolidation before deploying AI. The carriers that bolted AI onto fragmented systems saw single-digit improvements that plateaued quickly.
Dream Big or Go Home: Allianz’s CTO Sets the Tone
Christian Freytag, Group CTO and Head of Group Technology and Data at Allianz, opened the conference on June 3 with a keynote titled “The AI-Defined Insurer: Rewriting the Rules of Risk, Data, and Competitive Advantage.” His opening three words set the tone for the next two days: “Dream big or go home.” Sharing the main stage with Mike Ram, Anthropic’s Head of Insurance, Freytag outlined where AI in insurance actually stands, what separates leaders from laggards, and why the competitive question is no longer whether to adopt AI but how fast an organization can execute.
The keynote’s subtext carried a clear message for mid-market carriers. Allianz operates with 156,000 employees globally and has registered more than 900 AI use cases under its partnership with Anthropic, which was formalized in January 2026. When a carrier of that scale declares that the gap between leaders and laggards is widening, it signals that the technology advantage enjoyed by top-five global carriers may be compounding rather than diffusing. Mid-market carriers without comparable technology budgets face a strategic choice: invest heavily in data infrastructure now, or accept a growing competitive disadvantage as carriers like Allianz deploy AI across hundreds of use cases simultaneously.
Freytag’s presence alongside Anthropic’s Ram was itself a signal. Three years ago, the opening keynote at a major insurtech conference would have featured a consulting firm partner or a venture-backed startup founder. In 2026, the opening slot went to a carrier CTO and a foundation model lab executive, reflecting the structural shift from consulting-intermediated AI procurement to direct lab-to-carrier partnerships that we analyzed in our pre-conference coverage.
Anthropic and OpenAI Share the Stage: What Dual Keynotes Signal
For the first time at a major insurance industry conference, both Anthropic and OpenAI delivered keynote-level presentations in the same week. Mike Ram, Anthropic’s Head of Insurance, co-keynoted the opening session on June 3. Bastiaan de Goei, Industry Marketing Leader at OpenAI, took the stage on June 4 alongside Patrick Miller, Head of Data and AI at Newfront, for a dialogue that brought the AI conversation directly into the insurance industry’s central forum.
The simultaneous presence of both foundation model labs reflects the competitive dynamics we have been tracking across carrier AI deployments. OpenAI captured an outsized share of early carrier AI partnerships, with State Farm joining its Frontier platform, Travelers deploying its Realtime API for agentic claims calls, and multiple other top-20 carriers running OpenAI models in production. Anthropic has built its insurance position through a different route: the Allianz global partnership across 156,000 employees, Travelers’ parallel deployment of Claude for its 10,000-person engineering staff, and HUB International’s rollout of Claude across 20,000+ broker employees.
The competitive significance extends beyond speaker slots. When two foundation model labs are both investing dedicated insurance vertical teams and keynoting the industry’s largest conference, it validates insurance as a priority revenue vertical for AI platforms. For carriers, the dual-vendor dynamic creates leverage that did not exist when OpenAI was the de facto sole option. For actuaries evaluating vendor concentration risk under ASOP No. 56 model governance requirements, the emergence of a genuine two-vendor market is a material risk mitigation development.
The conference programming also featured life insurance AI leadership that broadened the conversation beyond P&C. Deepa Soni, Chief Information Officer at New York Life, and Laura Money, Chief Information and Technology Innovation Officer at Sun Life, both presented during the two-day event. Their participation signals that life carriers, which have traditionally lagged P&C in AI adoption, are accelerating their investment timelines. The combination of P&C, life, and specialty lines perspectives across 400+ speakers made the 2026 conference the broadest cross-sector AI programming the insurance industry has staged.
The Data Foundation Thesis: 28 Domains, One Missing Layer
The conference’s analytical center was the Thursday underwriting track session titled “Data Foundations to Decision Power: Building a Single Source of Truth to Unlock AI’s Impact.” Featuring Samrat Dua of Swiss Re Reinsurance, Suraj Tiwari of AXA XL, and Mark Blake of Stibo Systems, the panel described an exercise in which AI capabilities were mapped across 28 operational domains, from underwriting triage to claims adjudication to distribution analytics.
The key finding: the organizations that advanced fastest deliberately prioritized moderate- and lower-risk use cases first, building the delivery track record and data governance infrastructure needed before moving into higher-stakes financial and compliance applications. That sequencing matters because it contradicts the common assumption that carriers should target the highest-value use case first. Instead, the panel argued, the data foundation must be proven in lower-stakes environments before the organization (or its regulators) will trust AI outputs in reserving, pricing, or solvency calculations.
Blake introduced the concept of the “Golden Record,” a single, consolidated, and trustworthy version of each customer’s data with clear validation rules applied consistently. He described the current state at many carriers as a “Frankenstein customer” problem: the same policyholder exists across five or six disconnected systems with conflicting addresses, coverage histories, and claims records. When an AI model queries that data, it inherits the inconsistencies. The model’s output may be technically correct given the inputs, but the inputs themselves are unreliable because no single source of truth exists.
For actuaries, the Golden Record concept maps directly to the data quality requirements embedded in ASOP No. 23 (Data Quality) and ASOP No. 56 (Modeling). An AI model that consumes fragmented customer data to generate pricing indications or reserve estimates produces work product that the actuary must validate under both standards. If the underlying data lacks a single source of truth, the actuary’s reliance on that data requires explicit disclosure, and the model’s output carries a qualification that undermines its utility. The conference’s data foundation thesis is, at its core, an ASOP No. 23 compliance argument translated into technology strategy language.
For organizations still processing legacy documents written more than a decade ago, speakers were unambiguous: the architecture is the problem, not the AI. No amount of model sophistication compensates for a data layer that forces every query to reconcile conflicting records across disconnected systems.
Quantexa’s Award Validates the Data-Layer Thesis
The conference’s Insurtech Impact Award went to Quantexa, the connected data intelligence platform, in recognition of “demonstrated business value, solution uniqueness, and scope of real-world impact across the insurance sector.” The award was selected through carrier and partner nominations evaluated against four criteria: business process goal clarity, measurable financial impact, solution uniqueness in the market, and breadth of impact across users and teams.
Quantexa’s win is significant precisely because it is a data infrastructure company, not an AI model company. Its platform works by connecting disparate internal and external data points to create a single, contextualized view of customers and risk. In a conference where every keynote emphasized that AI is only as powerful as the data foundation beneath it, the top award went to a vendor that builds data foundations rather than AI models. The selection committee effectively ratified the conference’s central thesis through its award decision.
MJP Insurance received an honorable mention for achieving 84% efficiency gains versus market averages, a result that the awards committee noted required both technology implementation and operational redesign. The dual recognition of Quantexa (data infrastructure) and MJP (operational efficiency) reinforced the conference’s recurring message: technology without process change produces limited results, and AI without data infrastructure produces unreliable results.
Quantexa’s Claims Accelerator, now available on the Guidewire Marketplace, brings real-time decision intelligence directly into the claims workflow. For carriers running Guidewire ClaimCenter, the integration eliminates one layer of the data-shuttling problem that Cohen’s 80% statistic describes: instead of adjusters querying separate systems and manually reconciling results, the platform delivers a connected view within the claims system the adjuster already uses.
The Capital Cushion: $100 Billion Funds the Transition
The “Funding the Future of Insurance” panel, featuring Risha Mahadeo-Brown of Arch Insurance Group, Max Aronchick of Guy Carpenter, Mark Elliott of Hagerty, and Matthew Jones of MS Transverse Insurance Group, addressed the capital dynamics enabling the current wave of insurtech investment. The key figure: roughly $100 billion in cumulative industry earnings over three years has created the capital cushion that allows carriers and reinsurers to fund AI partnerships without straining surplus adequacy.
That capital position is reflected in the funding data. Gallagher Re’s quarterly insurtech reports show that global insurtech funding rose 19.5% during 2025 to $5.08 billion, the first annual increase since 2021. Re/insurers made more private technology investments into insurtechs in 2025, at 162 deals, than any other year on record. And in Q1 2026, AI-focused companies captured 95.2% of all insurtech funding, a sharp jump from 77.9% in Q4 2025, pulling in $1.55 billion across 68 deals at an average deal size of $25.79 million.
Jones offered the panel’s sharpest observation on capital discipline: “If your hunch is that the right thing is to grow at all costs, we should probably get out of insurance.” The comment drew knowing nods from an audience that has watched multiple insurtech startups fail by prioritizing growth over underwriting discipline. The capital flowing into insurance AI in 2026 is qualitatively different from the 2020-2021 insurtech funding bubble because it is flowing through re/insurer balance sheets rather than venture capital, and the investment criteria prioritize measurable operational efficiency over top-line growth.
Interest rates were identified as the single most impactful variable for capital behavior over the next 18 to 24 months. If rates remain elevated, investment income continues to subsidize technology investment even when underwriting margins compress. If rates fall, the capital cushion shrinks and technology spending faces the same scrutiny that all discretionary expenses face in a soft-cycle environment. The panel noted that ILS and sidecar structures are continuing to expand as investors favor the flexibility these vehicles provide for deploying capital into insurtech-adjacent opportunities.
The Adoption-Scale Gap: 63% Use AI, 7% Scale It
The conference’s data architecture thesis gains force when measured against adoption data. Conning’s 2025 C-suite survey found that LLM adoption among U.S. insurers jumped from 18% to 63% in a single year, with 90% of respondents in some stage of generative AI evaluation and 55% in early or full adoption. Yet only 7% reported scaling AI across their entire enterprise. That 63-to-7 gap, visible in every industry survey from Conning to Celent to Capgemini’s 344-executive study, defines the problem that Insurtech Insights 2026 confronted head-on.
The gap is not a model problem. Foundation models from Anthropic, OpenAI, and Google are all capable of processing insurance documents, generating policy summaries, and supporting underwriting decisions. The gap is an infrastructure problem. Carriers that successfully scaled AI, the 7%, share a common characteristic: they invested in data unification before deploying AI models. Carriers that remain stuck in pilot mode, the other 56%, share a different characteristic: they deployed AI models on top of fragmented data and found that the models could not reliably consume inputs from disconnected systems.
McKinsey’s April 2026 analysis of agentic AI in insurance core system modernization quantified the productivity gains available when the infrastructure problem is solved. Discovery and reverse engineering of legacy policy logic showed 20% to 50% productivity improvement. Data mapping, conversion, and quality work showed 15% to 90% improvement. Testing, reconciliation, and defect cycle compression showed the highest gains at 15% to 90%. Crucially, McKinsey noted that once core agents are built and governed, their marginal cost of reuse sharply declines, meaning the initial infrastructure investment compounds as each subsequent AI deployment inherits the same clean data layer.
Kristoffer Lundberg, CEO of Insurtech Insights, framed the conference’s conclusion in architectural terms: “The staircase has been built. The insurers who move with speed, trust, and the right controls in place are already climbing it.” The metaphor is apt precisely because it implies sequential steps. You cannot climb to the third step without having built the first two. And in insurance AI, the first step is data infrastructure.
The Distribution Layer: AI Moves Beyond Carrier Operations
While the data architecture thesis dominated the underwriting and claims tracks, the distribution track revealed a parallel shift. Agentic AI product launches are migrating from carrier-internal operations to producer-facing tools, and the conference programming reflected that migration. Rob Schimek, CEO of Bolttech, and Dawn Miller, Chief Commercial Officer and CEO of Lloyd’s Americas, both addressed how AI is reshaping distribution economics.
The distribution implications matter because the data architecture problem is, if anything, worse in the distribution channel than inside carriers. Independent agents and brokers operate across dozens of carrier systems, each with its own data formats, submission requirements, and quoting interfaces. The 80% data-shuttling problem that Cohen described for adjusters has an analog in distribution, where producers spend the majority of their time re-keying the same information into different carrier portals rather than advising clients.
Bolttech’s embedded insurance platform and the emerging wave of AI-powered submission tools are attempting to solve this problem from the distribution side. But the conference consensus applied equally: AI tools that sit on top of fragmented carrier APIs inherit the inconsistencies of those APIs. Until carriers expose clean, standardized data through modern integration layers, distribution AI tools will face the same architecture bottleneck that claims AI faces internally.
Actuarial Implications
The conference’s data architecture thesis creates specific consequences for actuarial practice across multiple workflows.
Data quality disclosures under ASOP No. 23. Actuaries relying on carrier data for reserving, pricing, or experience studies must assess whether the underlying systems provide a single source of truth or a reconciled approximation from multiple disconnected sources. The conference’s recurring finding that most carriers still operate with fragmented customer and claims data means that actuarial data requests may return outputs that reflect the limitations of the data architecture rather than the true policyholder experience. When data systems produce a “Frankenstein customer” as Blake described, the actuary’s data quality analysis under ASOP No. 23 must account for that fragmentation.
Model validation under ASOP No. 56. AI models deployed on fragmented data produce outputs that reflect data quality as much as model quality. An actuary validating an AI-driven pricing model or claims triage algorithm must distinguish between model error (the algorithm produces incorrect results given clean inputs) and data error (the algorithm produces correct results given unreliable inputs). The conference’s emphasis on data foundations as the binding constraint suggests that many model validation failures in practice may be data validation failures misattributed to the model.
Reserve development patterns. Carriers that complete data infrastructure modernization will exhibit different loss development patterns than carriers still running on fragmented systems. Faster FNOL intake, cleaner claims records, and automated routing will compress early development factors and reduce late-period adverse development driven by data-quality-related reopened claims. Actuaries building loss development triangles from industry data that blends carriers at different stages of infrastructure modernization should expect increasing heterogeneity in development patterns, making pooled industry factors less reliable than carrier-specific analysis.
Expense ratio assumptions. The $100 billion capital cushion funding insurtech investment will eventually translate into lower expense ratios at carriers that successfully deploy AI on modern infrastructure. Actuaries projecting expense ratios for pricing or competitive benchmarking should model a bifurcated trajectory: carriers with modern data infrastructure will see accelerating expense improvements as each new AI deployment leverages the existing data layer, while carriers on legacy infrastructure will see flat or increasing technology expenses as they invest in data modernization without yet realizing AI-driven productivity gains.
Vendor concentration risk. The dual Anthropic-OpenAI keynote presence confirms that the insurance industry is consolidating around two primary foundation model providers. Actuaries evaluating operational risk, particularly for appointed actuary opinions or enterprise risk management frameworks, should assess the systemic exposure created when multiple carriers in the same market deploy the same foundation model on the same vendor’s platform. A model failure or data breach at either provider could simultaneously affect actuarial workflows at dozens of carriers.
Why This Matters
Insurtech Insights 2026 marks the point where the insurance industry’s AI conversation shifted from model capability to infrastructure readiness. The conference did not debate whether AI works in insurance. It debated whether insurance data architecture can support AI at enterprise scale. That shift matters because it redefines what carriers need to invest in and how long the payoff timeline extends.
The 80% adjuster data-shuttling statistic, the Quantexa award for data infrastructure rather than AI models, the 28-domain mapping exercise that prioritized low-risk use cases as infrastructure proving grounds, and the 63-to-7 adoption-to-scale gap all converge on the same conclusion: the bottleneck is below the model layer. Carriers that recognize this and invest in data unification now will be the carriers climbing Lundberg’s staircase in 2027 and 2028. Carriers that continue layering AI on fragmented data will continue producing 42% non-measurement rates and single-digit productivity improvements that plateau after the pilot phase.
For actuaries, the infrastructure shift creates both opportunity and obligation. The opportunity: as data quality improves at carriers investing in modern architecture, the inputs to actuarial models become more reliable, reserve estimates become more stable, and pricing indications become more granular. The obligation: actuaries must now assess not just the AI model consuming the data but the data infrastructure producing the inputs. A clean model on dirty data produces dirty outputs, and the conference’s central message was that most of the industry’s data remains dirty.
Insurtech Insights 2027 is already scheduled to return to the Javits Center. If the 2026 conference defined the problem, the 2027 conference will measure whether carriers acted on it. The staircase has been built. The question is how many carriers are willing to climb.