Conversational AI case study: a voice agent that makes first contact with new leads in under a minute

The challenge: giving every new lead a timely first response

Before the build, sales teams worked with leads across different tools and their own phones. A first reply took 2 to 6 hours on average, longer when the team was busy, and around 15 to 30% of inbound leads had no meaningful contact before the next day. Some prospective customers lost interest or chose another provider while they waited.

Reps spent 2 to 4 hours a day entering data, moving leads from stage to stage, sending routine follow-up messages and searching previous conversations and call notes. The amount of detail in notes varied, which made a warm handoff to a colleague harder and limited managers' view of where prospective customers left the process or which channels brought well-matched leads.

The brief was to answer every new enquiry straight away, and to run the same process at peak times and after hours as at quiet ones.

What we built: conversational AI on one record per lead

One CRM: a Next.js front end on a Node.js back end, with an automation layer on top where the AI works. Twilio joins the four channels (calls, WhatsApp, SMS and email) in one messaging engine, so every call and message lands on the right lead's record without anyone filing it. Lead data and the history of every conversation sit in MongoDB and PostgreSQL.

The voice agent handles first contact. When a new lead comes in, it reaches out at once, asks the agreed questions, collects the details and leaves a short summary of the call on the record. Messages receive a reply within 10 to 20 seconds.

The front end runs on Vercel and the back end on AWS EC2 and Lambda. Critical communication events are retried, so a busy hour doesn't lose a message. Around the agent, the CRM gives managers live dashboards, pipeline and performance reports and lead scoring, and gives every lead one timeline holding each call, message and note.

Where the voice agent stops, and who takes over

The agent's part ends with the first contact. Everything after it is done by rules, by scheduled workflows or by a person:

The routing rules also distribute the work across the team. Because the agent has already gathered the basics, the rep's first conversation starts where the earlier call ended, giving reps more time for thoughtful sales conversations.

The summary also makes handoffs more considerate. Before, the detail available when a lead moved between reps depended on the notes captured at the time. Now the first call is summarised on the record, and the lead's timeline keeps every later call, message and note, so the next rep can continue without asking the prospective customer to repeat information.

Results

A lead record takes seconds to update because the system captures most of the data itself. The conversion figure is an estimate. It comes mostly from faster replies and more consistent follow-ups, which meant fewer prospective customers left between the enquiry and the first response. Managers can now see how the funnel performs, where the team needs support and which sources bring the best-matched leads.

Technologies and tools

Timeline

The build took nine months.

If your leads wait hours for a first reply, our page on AI voice agents sets out how we build the first call, and the one on AI sales agents covers the questions it asks and what it writes to your CRM.