Case study · Software & Internet · Sales automation
Colca AI Builds Voice AI SDR with ShellKode to drive 3.5x More Meetings
Colca AI · Sales automation
Colca AI's pipeline was limited by how many calls its reps could make. ShellKode enhanced Converse, its outbound sales product, with real-time voice AI on AWS, putting the team's best SDR on every call.
3.5x
More meetings booked
70%
Lower customer acquisition cost
80%
Faster ramp-up (onboarding)
~300–800 ms
Agent response time
The opportunity
A sales team that could reach thousands of prospects, but only call as many as its reps could dial.
Every sales team has one person who is brilliant on the phone. They open with something that matters to the prospect, handle pushback without missing a beat, and know exactly when to ask for the meeting. Colca AI's customers wanted that rep on every call. What they had instead was an outbound motion that could only grow as fast as they could hire.
Colca AI knew the problem firsthand. It had built its own pipeline through manual cold calls, with reps dialing and logging every conversation by hand. There were no transcripts, no summaries and no outcome tags. When a call ended, most of what was learned on it ended too.
More SDRs were never going to fix that. A different model would. Colca AI had that model in Converse, its outbound sales product, and ShellKode enhanced it with voice AI to deliver it at scale.
Reach set by the roster
How many prospects a team called each day came down to how many reps showed up.
One playbook, many versions
Every rep qualified and handled objections their own way, so good opportunities slipped through.
Calls that left no trace
Outcomes weren't captured, so handoffs to the sales team ran on memory.
Growth priced per hire
More calls meant more people, so pipeline grew only as fast as hiring did.
The solution
A voice agent that prepares, engages and follows through.
Converse is Colca AI's outbound sales product, now with voice AI. ShellKode, as engineering partner, enhanced it with real-time voice AI on ElevenLabs to run the full SDR workflow.
Before the call: know the prospect
Converse profiles each prospect by company, role, industry and use case, then opens with what matters to them, not a script.
During the call: a natural conversation
The prospect's speech is transcribed in real time by ElevenLabs speech-to-text (STT). Amazon Bedrock Agents handles dialogue, qualification and objection handling. Knowledge Bases retrieves accurate answers from the company's own content using keyword and semantic search on Amazon OpenSearch Serverless. ElevenLabs text-to-speech (TTS) then voices the reply, with responses in about 300–800 ms.
Human-like pacing
ShellKode added voice activity detection and contextual gap fillers. The prospect can confirm a meeting before the call ends.
A record of every call
Each call ends with a transcript, summary, sentiment read and outcome tag, from Interested to Meeting Booked, sent to downstream systems through Amazon API Gateway.
Built on AWS for security and scale
Serverless containers on Amazon ECS with AWS Fargate handle thousands of concurrent calls, while private networking and per-customer data separation keep information protected.
The impact
Now every prospect talks to the team's best SDR.
The quality of a conversation no longer depends on which rep is dialling. Every prospect gets a conversation that is prepared, relevant and consistent, and every call leaves sales knowing exactly what to do next.
3.5x more meetings booked
Converse qualifies prospects, handles objections and secures the discovery meeting during the call itself, across thousands of calls at once.
Every call opens with context
Converse profiles every prospect before dialing and opens with what matters to them, so more calls turn into meetings.
More replies, lower cost
Consistent, personalised outreach at scale lifted responses by 12% and lowered the cost of acquiring each customer.
Calls that feel like conversations
Prospects can interrupt and get a natural reply, so they stay engaged instead of hanging up on awkward pauses.
The stack
What it runs on.
- Voice and dialogue
- ElevenLabs speech-to-text and text-to-speech, Amazon Bedrock Agents
- Knowledge and retrieval
- Amazon Bedrock Knowledge Bases, Amazon OpenSearch Serverless
- Runtime and integration
- Amazon ECS with AWS Fargate, Amazon API Gateway
- Security
- Private networking, per-customer data separation
The road ahead
Built for one customer. Ready for many.
Converse was designed to serve many Colca AI customers from day one. Each customer sees only its own prospects, transcripts, summaries and knowledge, and that separation holds across every retrieval and workflow. So every new customer is an onboarding, not a rebuild.
- 01
Scaling tenant-level isolation to more customers
- 02
Richer call analytics for sales leaders
- 03
Deeper hand-offs into CRM workflows
Talk to us
Bring one sales motion. Leave with a voice agent that runs it.
30 minutes on one outbound workflow, whether that's prospecting, qualification or meeting booking, mapped to the architecture behind it. No slides.












