Start here
Your website goes silent at 6 p.m.
Most consultancy websites go silent after hours. The contact form sits there. The phone number links to a voicemail. The visitor who lands at 8 p.m. wanting to know whether you can solve their problem fills in a text box and hopes someone replies before they move on.
This is not a technology problem. It is a business process problem dressed up as a web design problem, and it is losing firms qualified leads every week without anyone noticing. It is the same feeling many Managing Partners and COOs describe looking at their contact page: the sense that their website is working against them while competitors move faster, and nobody has had time to fix it.
The leak
The lead leakage nobody is watching
39% of website conversations happen outside regular working hours, when there is nobody available to respond. Most firms handle this with a contact form or a basic chatbot. Both are too slow and too limited.
Speed matters more than most firms realise. Research shows that websites waiting five minutes to respond increase the risk of losing that visitor by ten times. Wait ten minutes and that multiplier reaches one hundred.
The firm that responds first, with something useful, wins the conversation. The one still relying on office hours alone does not.
The villain here is not the contact form. It is the mix of AI hype and generic vendor advice that has firms waiting for the perfect AI strategy while a simple, deployable tool sits ready to use.
The tool
What ElevenLabs conversational AI actually does
ElevenLabs is best known for voice generation: over 5,000 voice options, 70-plus languages, and some of the best voice quality available. But the part that matters for professional services is the conversational agent product.
The agent can operate as a voice assistant or a text chatbot embedded directly in your website. It answers questions from your own content, qualifies enquiries, and connects into your existing tools via webhooks. The two most useful integrations are calendar booking — so a qualified visitor can schedule a call without any human involvement — and webhooks, so conversations can trigger actions in whatever stack you already run.
This is the Agents layer of the Knowledge–Workflow–Agent (KWA) model in practice: AI doing real work in production, handling the top of your funnel while your team sleeps.
Build it
The agent in five steps
Setting this up is not a six-month AI project. It is five steps.
- Write a proper system prompt. This is where most implementations fail. A vague prompt produces a vague chatbot. Tell the agent who it represents, what it can answer, what it should never try to answer, and when it should suggest booking a call. Be specific about firm size, current AI usage, and the problem the visitor is trying to solve. Everything downstream depends on it.
- Choose a voice. ElevenLabs offers over 5,000 options including custom voice clones. Pick one that fits your brand’s register — a management consultancy sounds different from a creative agency. This is a brand decision, not a technical one.
- Build the knowledge base. Upload your services pages, case studies, pricing guide, and FAQs. ElevenLabs uses retrieval-augmented generation (RAG) to index this so the agent answers from your material rather than inventing. Non-enterprise accounts support up to 20 megabytes or 300,000 characters — enough for most consultancies. Document-heavy firms, such as legal practices, should check the enterprise tier.
- Connect calendar booking. Using a webhook, connect the agent to your calendar tool so it can check availability and book discovery calls directly. The agent collects the visitor’s name, email, preferred time, and time zone, then handles the booking without human involvement.
- Embed the code. ElevenLabs provides a snippet. Add it to your site and the agent goes live.
See it run
What the agent does under pressure
A live stress test across five realistic scenarios shows where the value is.
- Basic service questions. The agent answers clearly from the knowledge base without hallucinating, explaining what the firm does and does not do.
- Lead qualification. This is where the business case sits. A visitor saying “we’ve been hearing a lot about AI but don’t know where to start” gets a structured response: the agent asks about firm context, identifies a relevant starting point, and moves the conversation forward.
- Out-of-scope questions. “Can you train our team on Excel?” gets a clear no and a brief explanation of what falls within scope. Boundaries matter.
- Booking a call. The agent collects the required details, confirms spelling, checks availability, and books a slot. The visitor receives a confirmation email. Nobody in the office was involved.
- Adversarial questions. Challenged directly about data harvesting, the agent responds with transparency rather than deflection, pointing to published resources on AI privacy and bias.
With some prompt refinement, the agent can respond more like a senior salesperson on a first call: tighter answers, less filler, more forward momentum.
The caveats
The honest limits before you go live
- Knowledge base quality is everything. If the content is not in the knowledge base, the agent will either hallucinate or give a generic response. Invest time upfront writing clear, specific content. Garbage in, garbage out applies here more than anywhere else.
- Voice quality is excellent; conversation intelligence varies. The underlying language model drives dialogue quality. ElevenLabs lets you choose the model — better models cost more per minute. Test before you commit.
- This is not a replacement for human follow-up. The chatbot qualifies and books. A person still shows up and closes the deal. Do not treat it otherwise.
- GDPR and regulated industries. Conversations are processed through ElevenLabs’ servers. For most consultancies that is fine. For financial services, legal, or healthcare, review the data processing agreements first; higher-tier plans offer enhanced compliance options.
The call
When it makes sense, and when it does not
Add a voice chatbot if your website gets traffic but your contact form rarely converts, you are losing leads after hours or over weekends, your pre-sales questions are repetitive and answerable from existing content, or you want to show a forward-thinking approach without committing to a lengthy AI project.
Hold off if your website gets fewer than ten visitors a day — fix traffic first. Hold off if your services are not documented anywhere; the chatbot needs content to work from. And hold off if you have no process for following up on the leads it qualifies. A chatbot that books meetings nobody attends is worse than no chatbot at all.
The numbers
What it actually costs
ElevenLabs pricing starts at $0.10 per minute on a monthly plan, dropping to around $0.08 per minute on an annual business plan.
For a consultancy handling twenty conversations a day at three minutes each — roughly 1,800 minutes a month — that is approximately $180 on a monthly plan, or around $1,700 per year on an annual one.
For context: a single Google Ads lead in business services averages over $100. One qualified booking from the chatbot covers a month’s cost. The marginal economics are straightforward.
Where it fits
The next step, not the whole journey
Most firms reaching this stage have completed the first two steps of the 5 Steps for AI Leadership framework: they have Aligned on a written AI position and they are Activating AI tools in daily work. A voice chatbot is a practical next step toward Amplifying that work — taking an AI capability and making it visible to clients, not just internal to the team.
It is also a useful reference point for your Governance step. Before going live, confirm your data handling SOP, decide what the agent can and cannot say about pricing, and document the workflow from chatbot booking to human follow-up.
The bottom line
The technology is ready; are you?
Most consultancy websites are silent after 6 p.m. A voice AI chatbot costs less than a single Google Ads lead and operates around the clock. The technology is ready. The real question is whether your knowledge base, your processes, and your follow-up workflow are ready with it. If they are, the setup is five steps and a few hours of work. If they are not, start with the documentation.