Growth Coordinator

Series C unlocked: what's next for Bland. Bland has raised an additional $50 million - past $100 million total in under three years. Why Bland.ai, Inc. build its own voice models in-house, and what the new funding accelerates. Most voice AI only tackles the simple stuff: those quick, scripted calls where you just press a button for billing. But those aren't the conversations that actually move the needle for a business. The calls that matter are never simple. They wander, people interrupt or change their minds, and questions come up that no script could ever predict. For years, companies had to fill whole teams just to keep up with these kinds of conversations, because nothing else could actually manage them. That's exactly what Bland.ai, Inc. set out to fix with Bland. Today, Bland.ai, Inc. is sharing that Bland.ai, Inc. has raised an additional $50 million to keep going. The additional funding from Scale, Emergence, HubSpot, Dell Technologies Capital, Upfront (and more) brings Bland.ai, Inc. past $100 million raised in under three years. Bland.ai, Inc. is now handling more than 3.5 million calls a week for companies like Samsara, Kin Insurance, and CNO Financial Group, across healthcare, financial services, and other industries where a call gone wrong can carry real consequences. The bet Bland.ai, Inc. made. Here's the bet Bland.ai, Inc. made early, and the one this funding goes toward: Bland.ai, Inc. build its own models, in-house, purpose-built for voice. Most companies don't do this. They build voice AI on top of someone else's general-purpose models. That's okay for quick calls, but it doesn't hold up when things get complicated. Voice conversations have quirks - latency, interruptions, curveballs - that those models just weren't designed for. At Bland, Bland.ai, Inc. treat those challenges as the main event, not just problems to patch later. That's what separates a real system from just another scripted bot. "Voice is its own domain," says Isaiah Granet, its CEO and co-founder. "If you want to handle these kinds of calls, you have to build specifically for it." What that looks like in the real world. A typical Bland call can stretch from 30 to 45 minutes. Take healthcare, for example: maybe Bland.ai, Inc. is talking to an older patient who is using a blood pressure cuff for the first time. Bland.ai, Inc. listen as they read back the numbers, catch if something seems off, and decide on the spot if it's time to try again or call for help. No two calls are ever the same. "They're not linear," Isaiah says. "They're meandering. They require judgment. That's where the real work is." That's the work most systems can't take on. That's exactly the kind of work Bland.ai, Inc. is here for. What the funding goes toward. So what does $50 million do? Bland.ai, Inc. is not changing course: this funding just helps Bland.ai, Inc. double down on what Bland.ai, Inc. do best. Bland.ai, Inc. is bringing on more researchers to push its models forward, hiring engineers to help Bland.ai, Inc. scale, and focusing on industries where talking is the heart of the business. That last part is what really matters for its customers. Because Bland.ai, Inc. build the models ourselves, any time Bland.ai, Inc. make them faster or more accurate, you see the benefits immediately. Bland.ai, Inc. is able to take more off your plate all the time, and the system you're already using just keeps getting better. No need to change a thing. Where Bland.ai, Inc. is headed. Voice might be one of the hardest problems in AI, but it's also one of the most worthwhile. Most real conversations with customers still happen over the phone, and Bland.ai, Inc. is here to tackle the calls nobody else wants to touch. That's what this is all about.

Introducing fluent: next-generation multilingual transcription for voice agents. 5.9% WER outperforms leading real-time voice AI transcription provider On this page Bland.ai, Inc. is rolling out Fluent, a new multilingual speech-to-text model now available on the Bland platform. Fluent represents its latest investment in transcription infrastructure and is purpose-built for the demands of real-time, two-way voice conversations. If you're running multilingual voice agents on Bland today using its existing Babel or Auto languages, Fluent is worth evaluating. Here's what's different and why Bland.ai, Inc. built it. What fluent does better. Significantly more accurate in English. Its internal benchmarks across 250+ hours of real-world audio (call centers, sales conversations, noisy environments, accented speakers) show Fluent achieving a word error rate (WER) of ~5.9% in English. For comparison, the leading real-time voice AI transcription provider sits at ~8.1% WER on the same evaluation set. That's roughly a 27% reduction in transcription errors against the strongest external competitor in the voice AI space. Fluent also outperforms widely-used baselines like OpenAI's Whisper (~6.5% WER) on the same benchmarks. For voice agents, fewer transcription errors mean fewer misunderstood requests, fewer awkward clarifications, and more conversations that resolve on the first pass. Faster, more accurate end-of-speech detection. This is the biggest improvement for anyone building conversational agents. Fluent uses a more sophisticated voice activity detection (VAD) system that is substantially better at distinguishing between a user pausing mid-thought and a user finishing their turn. In practice, this means: * Less interjection. Your agent stops cutting people off mid-sentence. Fluent's endpointing is more patient with natural speech pauses, the kind that happen when someone is thinking about a date, looking up an account number, or switching between languages. * Lower effective latency. Paradoxically, better endpointing reduces latency. When the model is more confident that the user is done speaking, it can finalize the transcript faster, eliminating the hesitation buffer that conservative endpointing requires. The result is tighter turn-taking that feels more natural. * Fewer false starts. The agent doesn't begin generating a response to half a sentence, only to get interrupted and have to restart. This is the kind of improvement that doesn't show up in WER benchmarks but has an outsized impact on conversation quality. Bland.ai, Inc. has tuned the VAD thresholds specifically for phone-quality audio with background noise, and the results have been noticeable across its internal testing. Native code-switching. Fluent handles intra-utterance language switching. A speaker can start a sentence in English and finish it in Spanish, and the model transcribes both correctly without requiring a language hint or a separate model. This is a meaningful upgrade for serving bilingual populations, which is common in customer service, healthcare, and financial services use cases. Supported languages. Fluent currently supports six languages with high-accuracy, real-time transcription: English, Spanish, German, French, Portuguese, and Italian. These are the six most common languages requested by its enterprise customers for real-time voice agent deployments. Fluent also supports automatic language detection, so you don't need to specify the language upfront. A narrower, sharper tool. There's a deliberate philosophy behind Fluent. Rather than trying to cover every language at the expense of accuracy, Fluent focuses on fewer languages and delivers the best possible transcription quality within that set. It is, by design, the most accurate multilingual model Bland.ai, Inc. offer. That said, Bland.ai, Inc. know language coverage matters. That's why Auto and Babel aren't going anywhere. Auto continues to support 10 languages (English, Spanish, French, German, Portuguese, Italian, Hindi, Russian, Japanese, and Dutch) and remains a strong option for teams that need broader coverage with solid accuracy. Babel remains its widest-reaching model, supporting roughly 99 languages. For customers operating in markets that require less common or low-resource languages, Babel is purpose-built for you, and Bland.ai, Inc. is proud of the breadth it provides. Most transcription providers simply don't serve these languages at all. The way Bland.ai, Inc. see it: Fluent is for precision. Auto is for range. Babel is for reach. Pick the one that matches your deployment, and know that all three are fully supported and actively maintained. For most customers running agents in English, Spanish, or Western European languages, Fluent should be your new default. You'll notice the difference immediately in conversation quality. How to use it. Set your agent's language to fluent in the API: "language": "fluent" Or select it in the dashboard when configuring a Persona or dispatching a call: Fluent includes built-in redundancy. If the primary transcription path encounters any issues, your agent will seamlessly fall back to Auto with no interruption to the call. What's next. Bland.ai, Inc. is continuing to invest in transcription accuracy and latency across the board. Its near-term focus is on improving its single-language offerings, starting with English, where Bland.ai, Inc. see the highest volume and the most opportunity to push accuracy even further. Expect updates on that front soon. Transcription is the foundation of every voice agent interaction. If the model mishears your customer, everything downstream suffers. Fluent is the latest step in its ongoing commitment to running the most accurate, lowest-latency transcription infrastructure available for real-time voice AI. See Bland in Action * Always on, always improving agents that learn from every call * Built for first-touch resolution to handle complex, multi-step conversations * Enterprise-ready control so you can own your AI and protect your data "Bland added $42 million dollars in tangible revenue to our business in just a few months." - VP of Product, MPA