Voice AI Will Change How We Bank (February 2025 Fintech Newsletter)

Voice AI will change how we bank

James da Costa, Angela Strange

When Waymo launched in San Francisco in August 2023, few bystanders expected it to take a significant portion of the ride-sharing market. Fifteen months later, Waymo represents 22% of all SF ride-shares — equal to that of 13-year-old Lyft. Similarly, many industry players didn’t initially expect autonomous agents, let alone autonomous voice agents, to gain traction in banking. Yet, just like with Waymo, we anticipate that before long, we will begin to see a world where businesses and consumers can sit back and have their finances run on autopilot.

Financial services make up 25% of total spend on all global contact centers and over $100 billion in annual business process outsourcing (BPO) spend, whether it be verifying consumers with voice-based biometric authentication or managing complex customer issues via call centers. Even before the launch of ChatGPT, many banks were experimenting with voice automation: Bank of America’s voice agent Erica had 1 billion interactions by 2022.

Voice AI has not only reached par with humans; in some cases, it’s better

As our colleagues have written, voice AI is at an exciting inflection point. Voice agents allow you to scale to peak demand, staff phones 24/7, communicate in your customer’s preferred language, and operate at a fraction of the cost of human employees. Consider an immigrant from the Philippines with an auto loan who is working 9 a.m to 5 p.m. every weekday. Now, instead of needing to take time off work to get her questions answered, she can converse with an infinitely patient AI agent in her native Tagalog over the weekend. Early data suggests that some customers prefer speaking with an AI than with a human: imagine always getting the same customer service agent, with perfect memory of your every interaction every time you call.

Which brings us to the question: Why now? In 2024, a number of technical advances at the AI voice model layer paved the way for models to reach minimum thresholds in latency and emotion for wider adoption, leading to the emergence of exciting new AI applications. For example, ChatGPT Advanced Voice Mode added internet search, while Google launched Gemini 2.0 with native audio. This gave rise to a plethora of new voice agents and AI scribes across verticals.

Building on these core model improvements, several players have emerged that abstract away infrastructure requirements and provide tooling to get voice agents up and running quickly and reliably. Platforms like ElevenLab’s conversational AI, Vapi, Retell, and Bland all manage the orchestration of third-party speech-to-text (STT) models, LLMs, and text-to-speech (TTS) models. Yet, as startup and enterprise use cases reach scale, an increasing number of players are moving to their own infrastructure stacks across model orchestration (e.g., Pipecat), telephony (e.g., telynx), evaluations and testing (e.g., Coval), and custom databases for knowledge retrieval (e.g., Pinecone).

What we believe is required for voice agents to win in financial services

For voice agents in finance to become successful, they will need to navigate industry-specific nuances, such as:

• Compliance and regulatory requirements: Like their human counterparts, voice agents in finance need to adhere to strict regulatory requirements, including limitations on investment advice. Due to the potential for financial crime and fraud, agents will also need to build in additional or multi-step customer verification flows, especially as voice cloning becomes easier.
• Custom integrations and deployments: Teams building agents in finance will likely need to navigate on-prem deployments and access data from a range of legacy systems of record that require custom integrations.
• Domain-specific knowledge: Agents will need to be trained on a corpus of industry-specific terms and jargon, and have breadth across a large range of financial products (e.g., checking, savings, investments, credit, and mortgages).

When to Hire a CFO

Brian Roberts

After achieving product-market fit, early stage companies hire a finance leader who is tactically strong and capable of wearing multiple hats. Then, as the business scales and complexity mushrooms, startups often graduate to recruiting a head of finance. But at what stage of a company is a proper CFO needed? In our current era, there’s no predictable timeline for an IPO. Companies remain private longer for any number of reasons that make good business sense.

As the former CFO of Splunk, Lyft, and OpenSea, a16z General Partner Brian Roberts is frequently asked by founders about the next step—when to hire a CFO. He says companies and boards should start their decision-making process with a candid evaluation of their organization’s finance needs and continue this assessment on a semi-annual basis, at the very least.

Read more>>

More From the a16z Fintech Team

Recent M&A Deals and Market Intel

Monzo is reportedly holding preliminary discussions with bankers as it seeks to get “IPO ready” by the end of this year. Monzo’s CEO is advocating for a U.S. listing, while the company’s board is leaning towards a listing in the U.K.

Tabby is reportedly in talks with HSBC, JP Morgan, and Morgan Stanley for a public offering in Saudi Arabia in late 2025 or 2026.

HiBob announced its acquisition of Mosaic, a cloud-based FP&A platform, on February 13. The acquisition enables HiBob to offer a fully integrated solution that helps mid-market companies gain real-time insights into workforce productivity, employee engagement, and financial outcomes.

Rapyd is in talks to raise $300 million from investors at a $3.5 billion valuation. It intends to use the funding in part to acquire a payment processing startup.

Worldpay announced its acquisition of Ravelin, a next-gen fraud prediction and prevention solution for e-commerce merchants, on February 4. The acquisition complements and enhances Worldpay’s portfolio of value-added solutions. It is Worldpay’s first acquisition since spinning off from FIS last year.