Cresta Voice
Improve call handling with AI
Our chat platform success showed promising results: support agents handled higher volumes while sales agents increased conversions. However, chat represented less than 25% of contact centers—the vast majority operated via phone, creating a significant opportunity.
Vision
Empower voice agents with real-time AI support to boost confidence, reduce cognitive load, and improve call outcomes.
Problem to be solved:
Bridge performance gaps and boost consistency among voice agents.
Business need:
Understand the unique requirements of voice contact centers, representing 80% of the market.
My role:
End-to-end design leadership and adoption strategy.
Outcome:
Established strong foundation for future Cresta Voice iterations.
Learning:
Generated critical insights that became the groundwork for our voice agent solution.
Unlike chat agents, voice interactions required a completely different approach:
🏃♂️ No Processing Time
Callers expect immediate responses similar to face-to-face conversations. With agents handling one call at a time, they must respond instantly while trying to maximize call volume.
🗒️ No Conversation Reference
Without the ability to review conversation history, agents develop personal workarounds—handwritten notes, separate apps, or physical movement. This creates additional cognitive load and frustration when callers expect continuity across interactions.
☎️ Optional Tool Adoption
Our biggest challenge: while chat agents relied on Cresta as their primary workspace, voice agents could function with just a phone and CRM. We needed to demonstrate compelling value to drive adoption in an environment where our solution wasn't essential to basic workflow.
Key Problems
⏳ Extended Onboarding
The real-time nature of voice interactions allows minimal response delays. While holds are possible, they disrupt conversation flow. This high-pressure environment requires quick thinking and significantly extends new agent training time.
📚 Knowledge Access Bottlenecks
Agents struggled to quickly find current information, often resorting to messaging colleagues—wasting critical call time. This highlighted the need for instant, contextual knowledge delivery.
👂 Limited Conversation Memory
Without chat history or transcription tools, voice agents relied on personal note-taking to maintain context. This created additional cognitive load and made it difficult to track conversation details during and after calls.
These challenges demanded a solution specifically designed for the unique pressures of voice interactions.
What did we want to achieve?
Product goals
For the first iteration of our voice product, my focus was on two key objectives:
1. Actionable, Just-in-Time Support
Instead of forcing behavior changes, we prioritized delivering tangible assistance that complemented existing workflows. This approach aligned with Cresta's mission of enabling day-one expert performance while respecting voice agents' unique needs.
2. High Visibility Without Disruption
Since voice agents didn't depend on our platform (unlike chat agents), we designed Cresta to be consistently available yet unobtrusive—ready to assist without interfering with critical call moments.
First version
Final Design
Adaptive Interface
Created a sticky sidebar that remained visible without consuming excessive screen space. Beta testing revealed monitor size constraints, so we designed a minimal, modular interface with four key components—transcript, call flow, knowledge search, and help—all independently toggleable. This modular approach accommodated diverse needs: new agents kept the call flow visible for guidance while experienced agents customized their setup. The design also enabled precise usage tracking to inform future iterations.
Live Transcription
This feature gained immediate adoption, eliminating manual note-taking and providing reliable conversation references. Beyond improving real-time performance, transcripts simplified post-call documentation by preserving critical details that might otherwise be forgotten.
Intelligent Knowledge Access
After discovering agents resorted to Google searches and URL bookmarking due to poor CRM search capabilities, we integrated direct knowledge-base access within Cresta. We enhanced this with predictive auto-fetch that identified conversation keywords and proactively surfaced relevant articles—eliminating manual searching altogether.
Post-beta launch
Feedback and next steps
While transcription gained widespread adoption, other features showed limited uptake. Working closely with beta users revealed two critical improvement opportunities:
Enhanced Transcription Tools
Agents relied heavily on transcripts but found extracting key information from lengthy calls challenging. We planned to implement smart notes and automated summaries to highlight action items, customer issues, and resolutions—reducing cognitive load and streamlining post-call workflows.
Voice-Appropriate Coaching
We discovered that real-time Hints overwhelmed voice agents who needed to respond instantly. This led us to develop voice-specific approaches including asynchronous coaching, post-call analysis, and subtler real-time guidance—ensuring our behavior change features supported rather than disrupted the natural flow of calls.
These findings fundamentally shaped our development of voice-first solutions tailored to the unique demands of phone-based customer interactions.