A London-based waste-removal operator partnered with us to deploy a single AI agent that takes inbound and outbound phone calls, replies on WhatsApp, and updates the CRM — qualifying, pricing, and booking leads around the clock, and only forwarding to a human operator when the situation genuinely calls for one.

Key Results

80% of calls handled end-to-end with no human involvement

24/7 coverage across voice calls, WhatsApp, and CRM

~3 min average handle time from answer to priced offer

What Is a Multi-Channel AI Sales Agent?

An AI voice agent is an intelligent phone assistant that can answer calls, qualify leads, book appointments, handle customer support, and automate sales conversations 24/7. Unlike traditional IVR menus or voicemail systems, AI voice agents understand natural speech, respond in real time, and provide a human-like customer experience.

Businesses use AI phone agents, AI receptionists, and voice AI for sales to reduce missed calls, improve response times, and increase conversions. Connected to your CRM and business tools, an AI voice agent can update records automatically and transfer important calls to your team when needed.

Executive Summary

Our client, a licensed waste-carrier operating across London and the surrounding area, was generating strong paid-search lead volume but losing bookings to slower response times and inconsistent quoting. Leads arriving outside office hours waited hours for a call-back; pricing varied depending on which agent picked up; and CRM hygiene was uneven because every stage update was a manual step.

We designed and deployed an AI sales agent that now handles roughly 80% of inbound and outbound calls end-to-end — qualifying the job, producing a priced offer, sending a payment link, and leaving the CRM in the right state. The remaining ~20% of calls, where the situation is complex or sensitive, are handed off to a human operator with full context attached. Crucially, most of the agent’s behavioural specification was not written by hand: we processed historical recordings of the client’s best-performing human agent and used that corpus to automatically discover the call structure, objection-handling patterns, and pricing logic that made those calls convert — reducing the effort the business itself had to invest in the build.

Business Context & the Challenge

About the Client

Industry	Waste removal & clearance — consumer and light-commercial
Market	London and surrounding areas, UK
Business Profile	Licensed, fully insured SMB waste-carrier
Lead Source	Primarily paid search, with high volume of same-day enquiries
Engagement	Design, build, and rollout of a multi-channel AI sales agent

The Four Problems Costing Bookings

The client’s paid-search funnel was healthy at the top and leaky in the middle. Four specific problems were costing bookings:

• Speed-to-lead. Enquiries arrived at all hours. A lead that waited until the next morning for a call-back had often already booked with a competitor.
• Pricing consistency. Quotes depend on a range of factors about the job and the property. Different human agents interpreted the rate card differently, and mistakes in either direction — over-quoting or under-quoting — cost the business money.
• Channel fragmentation. Some clients want a phone call, many prefer WhatsApp, and most expect a payment link they can tap at their convenience. The team was shuttling information between phone, WhatsApp, and the CRM manually.
• CRM discipline. Pipeline stages, follow-up tasks, job details, and preferred collection dates were captured inconsistently, which broke reporting and made reliable follow-up difficult.

The objective was clear: every enquiry should receive an immediate, correctly-priced, channel-appropriate response, with the CRM always left in a usable state — and without expanding the human team.

The Solution

A Single AI Agent Across Voice, WhatsApp, and CRM

The AI sales agent we built operates as one behavioural unit across three integrated channels. It is not a chatbot bolted onto a call centre and a separate voice system — it is the same agent, applying the same logic and maintaining the same tone, regardless of how the customer chooses to communicate:

• Cellular voice calls. The agent places outbound calls to newly-submitted leads within minutes, and takes inbound calls on the company’s line. It qualifies, prices, and closes in a single conversation where possible.
• WhatsApp chat. The same agent replies in WhatsApp threads, receiving photos, answering questions, and sending the offer and payment link — keeping the conversation on whichever channel the client prefers.
• CRM actions. The agent writes directly into the CRM: updating the lead, recording preferences, creating follow-up tasks, and dispatching templated offers with payment links. Everything a human operator would type gets typed.

Roughly 80% of calls are processed entirely by the agent with no human involvement. The remaining ~20% — complicated jobs, sensitive situations, unusual commercial requests, escalations, or cases where the client simply prefers to speak with a person — are forwarded to a human operator with the full conversation and CRM context already populated, so the operator picks up informed rather than starting from zero.

How We Reduced the Client’s Build Effort

One of the most time-consuming parts of deploying an AI sales agent is normally writing the specification: the call structure, the qualifying questions, the objection handling, the tone, the edge cases. Asking a busy SMB to produce all of that in document form is a known blocker.

Instead, we took a large set of historical call recordings from the client’s best-performing human agent and ran them through an automated discovery pipeline. The pipeline:

• transcribed the calls, 
• extracted the recurring call structure, 
• identified the qualifying questions that actually moved the price, 
• surfaced the phrases that worked in objection handling, and 
• reconstructed the implicit pricing logic that the top agent was applying. 

That output became the first draft of the behavioural specification, which we then refined with the client in a small number of focused sessions.

The practical effect is that the client spent hours, not weeks, on specification work — and the agent sounds like their best person because it was trained on that person’s actual calls.

Operating Model

• Automatic handling for the standard path: qualification, pricing, offer, payment link, CRM update, follow-up scheduling.
• Automatic hand-off for everything else: clearly-defined trigger conditions route the lead to a human operator with full context, rather than the agent attempting work it shouldn’t.
• Consistent tone across channels: the same voice and register whether the client is on the phone or in WhatsApp.

Technology Stack

• ElevenLabs — natural, low-latency voice for the cellular channel.
• OpenAI — language model powering conversation, qualification, and pricing reasoning.
• LangChain — orchestration of the agent’s tools, channels, and conversational state.
• Python — the application layer, including CRM tool integrations and the automated recording-analysis pipeline used to bootstrap the specification.

Results and Business Impact

The deployment directly addressed each of the four breakages identified at the start of the engagement.

Operational Results

• 80% of calls fully automated. The agent handles the vast majority of conversations end-to-end, with human operators reserved for the complicated ~20% where their judgement actually adds value.
• Speed-to-lead collapsed to minutes. Every new lead gets a response attempt essentially immediately, 24/7 — including evenings, weekends, and overnight enquiries that previously sat in a queue.
• Consistent pricing on every call. Quotes no longer depend on which agent picked up; the same logic applies every time, in the client’s interest.

CRM & Reporting Results

• Clean, structured pipeline data. Every lead lands in a well-defined state with the correct follow-up task attached, making forecasting and reporting genuinely usable.
• Follow-ups never dropped. Warm leads no longer go silently cold in an inbox — every deferred conversation produces a scheduled task.
• Structured notes instead of free-text chaos. Subsequent humans get a clean summary of the job, price, and preferred date at a glance.

Commercial & Customer Experience Results

• Every enquiry gets a priced, bookable offer. Where possible, the agent closes with a payment link in the same conversation, turning an enquiry into a prepaid booking.
• Appropriate escalation rather than avoidance. Sensitive and complex cases reach a human operator quickly and with full context — improving both outcomes and the client’s experience of the brand.
• Human capacity redirected to high-value work. With the agent absorbing the repetitive qualification and quoting load, the team’s hours go to on-the-ground operations and the genuinely complex jobs that warrant a human touch.
• Minimal build effort required from the client. Because the behavioural specification was bootstrapped automatically from existing call recordings, the client invested a small fraction of the time a handwritten spec would have demanded.

In Practice

“The agent runs the same playbook our best person would — on every lead, at every hour, across calls and WhatsApp, with the CRM updated correctly every time. And the part we expected to be painful — writing down how we actually sell — barely happened, because it was reconstructed from our own recordings.”— Operations Lead, Client Company

Does This AI Sales Agent Approach Work for Other Industries?

Yes. The challenges this client faced — slow response times, missed calls, inconsistent pricing, fragmented communication channels, and poor CRM management — are common across many service-based businesses. A custom AI sales agent can solve these issues by handling enquiries, qualifying leads, booking appointments, and updating CRM records automatically.

This approach works especially well for:

• Beauty salons, barbershops, and clinics needing automated bookings and appointment reminders
• Restaurants, cafés, and hospitality businesses managing table reservations and customer enquiries
• Home services and trades such as plumbing, electrical, removals, and cleaning
• Field service and logistics companies handling urgent requests and scheduling jobs
• Property agencies and letting businesses managing high enquiry volume
• Any SMB using phone calls or WhatsApp for lead qualification and pricing quotes

A custom AI phone agent is especially valuable when the best sales knowledge exists in experienced staff members rather than written processes. The AI can learn those conversations, apply the same logic consistently, and scale customer service without adding headcount.

Ready to achieve similar results?

If your business depends on qualifying inbound leads quickly, pricing them correctly, and booking them across multiple channels without expanding your team, we can design and deploy a custom AI agent for your specific operation. 

In most cases, we can train the AI voice agent using your existing call recordings and real customer conversations, so there is no need to create complex scripts from scratch.

→ Schedule a Free Strategy Call

Let’s discuss how an AI agent grounded in your best people’s real conversations could transform your lead pipeline.

Frequently Asked Questions

How much does a custom AI sales agent cost?

Pricing depends on channels used (phone, WhatsApp, CRM), integrations, and conversation complexity. Most projects are scoped after a free strategy call where we review your lead flow and business needs.

How long does it take to launch an AI phone agent?

Most AI voice agent projects go live in 4 to 8 weeks. If you already have call recordings or scripts, deployment is often faster.

Can AI handle inbound sales calls?

Yes. A well-trained AI sales agent can answer calls, qualify leads, give pricing ranges, book appointments, and transfer complex cases to your team when needed.

Does an AI sales agent integrate with our CRM?

Yes. We connect AI agents with popular CRM systems so leads, notes, follow-ups, and bookings are updated automatically.

Can AI answer both calls and WhatsApp messages?

Yes. One multi-channel AI agent can manage phone calls and WhatsApp conversations with consistent tone and logic.

Does the AI sound human?

Modern voice AI sounds natural, fast, and professional which most callers cannot distinguish from a human agent in a standard sales conversation. Tone, pacing, and vocabulary can be configured to match your brand. 

What if the AI cannot help?

The call or chat is handed to a human team member with conversation history included, so customers do not need to repeat themselves.