Voice is the most natural human interface. It removes friction, builds trust, and enables real-time, nuanced interaction, which is critical in complex and highly regulated industries. Here at Frontline, we believe voice AI marks a paradigm shift in how customers will interact with businesses. We’ve already backed Telnyx, Regal, and Tucuvi in this space, and are now excited to welcome Avallon to the Frontline portfolio, too.
The Evolution of Voice
For most of the 2010s, “voice” in finance meant rigid IVR trees, high transfer rates, and siloed call-center tooling. Word error rates were too high for complex, regulated conversations, assistants were FAQ-bound, and anything beyond simple queries fell back to humans. Voice was a cost center — useful for scale, limited for intelligence.
Over the last decade, that has fundamentally changed. Continuous advances in automatic speech recognition (ASR) and natural language understanding (NLU) reduced word-error rates from double digits to near human parity (~5% in benchmark tasks). Today’s real-time LLM-powered systems go much further than transcribing: they infer intent, detect sentiment, and respond with contextual nuance within milliseconds. Voice interactions have evolved from “press 1 for balance” to natural, dynamic conversations that can execute secure, end-to-end tasks.
Wave 1: Reliable self-service: Improved ASR/NLU moved voice from basic triage to reliable service and enabled banks to deploy assistants that handled complete customer requests (PIN resets, card replacement, statements, etc.), cutting handle times and improving first-call resolution. Voice became the front door for routine service.
Wave 2: Voice as workflow: Modern voice agents capture structured data, validate policy details, trigger back-end workflows, and automatically write accurate CRM or Third-Party Administrator (TPA) notes. It’s not just the conversation, but the entire workflow that’s supported.
The convergence of accuracy, latency, and reasoning marks a new era in voice capabilities, where financial institutions like banks and insurers can move beyond call centers to embed voice across the entire stack, from automating servicing, claims intake, compliance review, and relationship management:
- Claims and servicing: Voice AI can collect structured data from spoken narratives, auto-fill claim forms, and trigger triage steps in minutes instead of hours, cutting “First Notification of Loss” (FNOL) times by up to 60%.
- Wealth management: Advisors use voice assistants to summarize client meetings, extract action items, and automatically update CRMs.
- Banking and credit: Banks like NatWest and Bank of America are using voice-enabled virtual assistants for personalized product discovery, not just support.
- Compliance and audit: Voice systems can flag disclosure gaps or risk phrases during customer conversations, turning a traditionally manual QA process into an automated one.
Beyond improving efficiency for the company, voice AI is also more accessible for customers, who may be visually or physically impaired, have low literacy levels, or speak a different language from the vendor.
In an industry where trust and complexity matter most, we’re seeing financial institutions reimagine how they speak with their customers, literally. Early adopters are already reporting 20-40% reductions in average handle time and 25-30% improvements in first-call resolution. That impact is translating into spend too, with the market for conversational AI projected to reach ~$41B by 2030 at ~24% CAGR. The data is clear: customers prefer to talk, and the institutions that choose to listen intelligently will win.
As generative AI moves into regulated environments, winners will combine conversational design with governance, explainability, and domain expertise. The right platform architecture includes:
- Role‑based permissions for actions that change money, entitlements, or customer data
- 100% interaction scoring for accuracy, completeness, and tone, with automated QA and audit trails
- Locally trained, domain‑specific speech models that handle accents, jargon, and policy terms
- Redaction, retention, and supervision policies consistent with regulator expectations
- Grounded reasoning on current policy and product states to avoid drift
It’s this next wave of opportunity in voice that we are most excited about at Frontline. Specifically, we’ve been deep-diving on the insurance industry, which is now primed for Voice AI because of three converging trends:
- Model maturity: Speech-to-text, intent recognition, and LLM reasoning have crossed critical accuracy thresholds (financial services-grade assistants handling billions of interactions with >98% answer success) making automation viable even in high-stakes domains.
- Regulatory openness: Supervisors are increasingly comfortable with explainable AI systems that enhance transparency and auditability.
- Customer expectation: Consumers now expect instant, human-like support, without the frustration of rigid IVR trees (reflected in rising digital engagement and assistant usage across banks).
As we’ve heard repeatedly in diligence calls, real-world adoption depends not on flashy demos, but on error rates and explainability. For example, a single wrong digit in a car insurance claim or an incorrect name in a KYC process can derail entire workflows.
That’s why domain-specific, locally trained speech models, especially in highly complex financial processes such as claims processing, are likely to outcompete generalist systems over the next few years.
We expect the global speech AI market to bifurcate:
- Horizontal infrastructure providers powering global developer ecosystems, and
- Vertical voice intelligence platforms built for regulated domains such as finance, healthcare, and government.
Introducing Avallon
Founded by Cornelius Schramm and Bryan Guin , Avallon is a company that is well-positioned to build a generational company within the Voice AI market for financial services — starting with insurance and claims operations, where friction and manual processing are still the norm. Today, we’re delighted to share that we’ve led the team’s $4.6m Seed round.
Avallon’s platform offers a more scalable and efficient claims handling process for insurance carriers and TPAs, most of which depend on legacy IVR systems and call centers, without sacrificing human warmth or regulatory rigor. Its AI voice agents address operational inefficiencies and automate operational tasks across the claims lifecycle from intake to resolution, including:
- managing intake, status and billing questions by phone, email or file upload
- tracking case status, contacting employers, providers, repair shops and injured workers
- summarizing and extracting information from critical insurance documents such as PDFs, invoices and medical reports
- analyzing policy terms and flagging exposures
In early deployments, Avallon has already shown measurable improvements in:
- Customer satisfaction through shorter wait times and higher first-call resolution
- Agent productivity via automated documentation and post-call summaries
- Operational efficiency by turning unstructured voice data into actionable insights
As one of their early customers put it: “It’s the first time we’ve seen an AI that actually understands what our customers are saying, not just what they’re typing.”
Avallon’s product strategy leverages Voice AI to gain a deeper understanding of claim-specific context, positioning the company as a contender to become the future system of record. We believe this is a rare and disruptive opportunity: by owning both the interface and the data model over time, they have the potential to reshape how critical insurance interactions are captured, processed, and acted upon, creating significant long-term strategic value.
Philipp and I first met with founders Cornelius and Bryan in their hacker house in San Francisco, and were quickly impressed by their rare mix of regulated-industry experience and deep knowledge on speech technology.
The pair met while studying computer science and machine learning at Cornell University. Since then, Cornelius scaled the US expansion efforts for FINN as an automation engineer, where he built the fleet operations platform and experienced the manual, paperwork-driven pain points of the insurance industry firsthand. Bryan, meanwhile, worked as software engineer at Agentive where he built AI systems for auditors and led product-consulting teams at EY advising Fortune 500 clients on AI.
They’re joined by Moritz (ex-Taktile) and Leander (ex-FINN) to build Avallon out of the heart of financial services – New York City. Avallon is hiring software and systems engineers. Learn more about open positions here.
For us, Avallon represents a bet on the next operating layer for financial services, and we’re proud to partner with them on their mission to make financial institutions speak and listen more intelligently.
If you’re building a voice AI product with global ambitions, or know a team that is, please get in touch.
