Case Study: Klubi
Klubi is reinventing consórcios in Brazil with a fully digital experience that makes large purchases accessible without financing interest. By pairing an intuitive app with AI voice agents powered by Deepgram’s speech-to-text, Klubi is scaling voice-led growth to automate pre-sales at scale, increase qualification, and expand AI agents across sales, service, and post-sales workflows — while upholding strict operational controls.
Key Results at a Glance
Klubi rebuilt its funnel around AI voice, anchored by Deepgram Speech-to-Text (STT), to increase conversion, expand coverage, and improve consistency across thousands of interactions.
- ~80k total monthly customer interactions across pre-sales, post-sales, and support
- ~30k monthly AI-led voice interactions
- 90%+ of conversations handled end-to-end by AI agents
- Workforce recomposition: from ~40 human SDRs to AI agents operating continuously across voice and text
As a digital provider of a regulated, interest-free group purchasing model for high-ticket goods, Klubi serves consumers who often lack access to traditional credit.
Inbound demand is strong, but prior to AI adoption, early outreach faced familiar constraints:
- ~60% of leads unresponsive on first WhatsApp contact
- ~50 SDRs responsible for explaining the model, qualifying intent, and routing opportunities
Human-only pre-sales limited throughput and consistency, slowing response times for higher-intent prospects.
To overcome these constraints, Klubi re-architected its funnel around AI voice agents, using Deepgram's Nova-3 streaming API as the core speech-to-text layer to ensure accuracy, speed, and resilience under real-world Brazilian call conditions.
- Pre-sales automation at scale: AI agents initiate contact, educate, qualify, and hand off when appropriate
- High-fidelity understanding: reliable recognition of Brazilian Portuguese (pt-BR) and domain-specific terminology
- Noise robustness: sustained accuracy in everyday environments such as street-level or mobile calls
- Simple integration and rapid iteration: fast deployment and operational stability using Deepgram's Streaming STT
Deepgram’s STT foundation also supports service and post-sales workflows, with transcripts feeding monitoring and QA processes.
“We evaluated several speech-to-text providers, but Deepgram stood out immediately for Brazilian Portuguese. Other solutions struggled with domain-specific vocabulary—confusing car model names like 'Ônix' with common words like 'ônibus' [bus]. Deepgram got it right, even in noisy environments like traffic or subway calls. That accuracy was non-negotiable for us.”
Gabriel Abreu
Tech Lead, Klubi
Adopting Deepgram enabled Klubi to establish a voice-led operating model built for scale:
- Tens of thousands of monthly AI-led interactions across voice and text
- The majority of customer conversations automated without loss of control or standardization
- Consistent, steady conversational tone at scale, independent of volume or repetition
Behind the scenes, Klubi’s agents rely on Deepgram’s low-latency streaming and accurate transcription to drive real-time decisioning, with an architecture that can later adopt Deepgram’s Voice Agent API or Flux for even more natural, turn-based conversations.
“Integration was surprisingly straightforward. Our developers gave positive feedback from day one, and we rolled out to production quickly. Now our AI agents, powered by Deepgram, handle calls even when customers are on the street or in transit. That reliability lets us focus on building better experiences, not debugging transcription issues.”
Gabriel Abreu
Tech Lead, Klubi
Looking Ahead
With the automation foundation in place, Klubi has gradually evolved the quality of those interactions.
In voice channels, AI agents use regional accents to reduce the sense of distance often associated with automated calls in the first moments of contact. Rather than moving immediately into transactional intent, some conversations open with light, context-aware small talk, helping set a more natural tone before progressing.
Klubi has also introduced contextual continuity across interactions, allowing conversations to resume from prior touchpoints instead of restarting each time. This approach improves fluency and perceived attentiveness, particularly in longer or interrupted customer journeys.
By pairing disciplined agent design with accurate, noise-robust speech recognition, Klubi demonstrates how speech-to-text can serve as a durable foundation for scaling regulated financial operations — while still leaving room to evolve how natural, continuous, and familiar those conversations feel over time.
As Klubi continues to scale, the team is exploring additional capabilities from Deepgram's roadmap. The company is monitoring Deepgram Flux's expansion into multilingual support, particularly for Portuguese, to further improve accuracy and efficiency.
