Guava’s $12.5M Seed: What LA Businesses Should Ask Before AI Answers the Phone

Los Angeles has a new voice AI company with a different pitch than most of the "AI receptionist" tools flooding small-business inboxes. Guava, an LA-based developer platform for building voice agents, announced a $12.5 million seed round on Nov. 2 led by Acrew Capital, with Flourish Ventures, WillowTree Ventures, Clocktower Ventures and pre-seed backer Commerce Ventures joining, according to the company's Business Wire announcement. Axios Pro first reported the round.

Guava sells to enterprises and business process outsourcers, not to the taco shop on Sunset. But the way it is building — and the reasons investors gave for backing it — are a useful checklist for any LA founder or operator deciding whether to let software answer the phone.

What Guava actually built

Most voice agents on the market today are assembled from parts: one vendor for speech recognition, another for the language model, a third for the synthetic voice, all steered mainly by written prompts. Guava says it runs its own speech recognition, language model and speech synthesis as a single stack, and lets developers control agent behavior with code and business rules instead of prompt tinkering. Developers can plug it into their own products through Python, TypeScript or Claude Code.

The technology comes out of Gridspace, a voice AI lab whose co-founders, Anthony Scodary and Nico Benitez, are also Guava co-founders. The company says its models were trained on more than 10 billion minutes of live conversations in regulated call centers. CEO Omri Dahan, whom the release describes as a former Marqeta revenue chief, and chief revenue officer Mike Piechna, who Acrew says brings more than 15 years of early-stage sales experience, round out the founding team, per the release and Acrew's investment note.

Guava's first public model, Daytona, launched in September. The company says it already supports more than a dozen enterprise customers and tens of thousands of calls per day, and that enterprise clients typically take their first live regulated call within 14 days of kickoff. There is also a free tier that the company says can produce a working agent in under five minutes. The new money goes to model and product development and a first go-to-market team; FinTech Global adds that the roadmap includes edge-device and on-premises deployments and "full-duplex" agents that can listen and talk at the same time.

Why "high-stakes calls" is the real story

Acrew framed the bet around a simple observation: a lot of important customer work — scheduling, live support, insurance and financial questions — still happens by phone, and in regulated settings those calls have to be accurate, secure and auditable. That is a higher bar than a chatbot on a website.

For LA businesses, that framing matters even if you never touch Guava's SDK. Dental and medical offices, insurance agencies, property managers, auto shops, salons and law firms across the region field the same repetitive calls every day. The temptation is to bolt on the cheapest AI answering tool available. Guava's pitch is a reminder of what can go wrong when the voice layer is a black box: wrong appointment times, invented policy answers, or no clean record of what the agent said.

A buyer's checklist for LA operators

Before you let any voice agent talk to your customers, ask the vendor:

1. Who controls what the agent can say? Look for tools that pull answers from your own systems and rules, not just a long prompt someone typed once.
2. What happens when it doesn't know? A good agent hands off to a human or takes a message. A bad one guesses.
3. Can you audit calls? You want recordings or transcripts you can review, especially for anything touching health, money or contracts.
4. What are the consent rules? California is an all-party consent state for recording confidential calls, so get legal advice on disclosures before you record or transcribe customer conversations.
5. How fast is a real pilot? Guava cites roughly two weeks for enterprise customers. Small businesses should expect a vendor to prove value on a narrow use case — say, appointment confirmations — within a month.
6. What does it cost per minute at your volume? Ask for pricing at your real call count, not a demo tier.

The founder takeaway

For LA startup founders, Guava is also a signal about where seed money is going in AI: toward teams that own a hard technical layer and sell into regulated, high-volume workflows, rather than thin wrappers on someone else's model. The company paired a decade of voice research with an operator who has scaled fintech revenue — the kind of founder mix investors keep rewarding.

If you are building on top of voice AI, the bar is rising. Reliability, control and auditability are becoming the features customers ask about first. If you are buying it, start small, keep a human in the loop and insist on the same standards Guava says its enterprise customers demand.