Key takeaways
- Chat suits considered, text-friendly enquiries; voice suits urgency and callers who will not fill in a form.
- Voice is materially more expensive to run, because it bills per minute rather than per message.
- Both need the same foundation: your content, structured well enough to answer from.
- Disclosure that the caller is speaking to an AI should be configured by default.
- If your enquiry volume is low, a fast human response beats either of them.
They solve different problems
A chatbot answers questions in text on your website, app or WhatsApp. A voice agent holds a spoken conversation on the phone, inbound or outbound. Both use the same underlying language models, but the operational fit is quite different and the choice usually follows from how your customers already behave.
The strongest signal is your existing enquiry mix. If most enquiries arrive as calls, a chatbot will not capture them — those callers are choosing the phone deliberately. If most arrive by form or WhatsApp, a voice agent is solving a problem you do not have.
Where chat wins
Chat is better for anything the user needs to read carefully: specifications, pricing structures, availability, policy details. It leaves a transcript the user can scroll back through, it handles links and images, and it works in a noisy room or a shared office.
It is also far cheaper at volume. A chatbot conversation costs a fraction of a phone minute, and it handles many conversations concurrently without any additional infrastructure.
And it is asynchronous. A user can start a conversation, leave, and come back — which suits considered purchases where nobody decides in one sitting.
Where voice wins
Voice wins on urgency and on reach. Some customers will not use a form or a chat window under any circumstances, and for many businesses those customers are a large share of revenue. A phone number that is always answered captures enquiries a chat widget never sees.
Outbound is the other strong case. Speed of response to a new lead is one of the more reliable predictors of whether it converts, and a voice agent can call a form submission back within seconds rather than whenever someone works down the list.
Voice also handles the messy middle better than a form does. A caller who does not know what they need can be asked questions, and the agent can qualify them in one pass rather than through three rounds of email.
What both need before they work
Neither is a drop-in product. Both need grounding in your actual content — your pricing, your policies, your availability, your service definitions — so that they answer from your business rather than from generic knowledge. If that content does not exist in a usable form, that is the first piece of work regardless of which channel you pick.
Both need an escalation path. The failure mode that costs you customers is not the AI being wrong; it is the AI refusing to hand over. Sentiment, repeated confusion or a direct request for a person should trigger a transfer, and the human should receive the context so the customer does not start again.
Both need logging. The unanswered questions are the most valuable output in the first month, because they tell you exactly where your published information is inadequate.
The cost difference is real
Chat bills roughly per message and is inexpensive at almost any volume a small or mid-sized business generates. Voice bills per minute across speech recognition, the language model and speech synthesis, and a five-minute call is not a rounding error.
That changes where each makes sense. Voice pays for itself when the value of a captured call is high — a booked appointment, a qualified sales enquiry, an urgent service request. It rarely pays for itself for low-value, high-volume queries, which is exactly what chat is good at.
A common sensible split: chat handles informational volume on the website, voice handles inbound calls and speed-to-lead callbacks.
When the answer is neither
If you get five enquiries a week, a person answering them well within an hour will outperform either system, and the build cost will not return. Automate when volume or coverage is the constraint, not because the technology is available.
If your enquiries are genuinely complex — bespoke quotations, sensitive circumstances, negotiation — a badly-scoped AI layer will frustrate the exact customers you most want. Route those to a person immediately and use the AI for the qualification step ahead of them, if at all.
Frequently asked questions
- Can an AI voice agent replace my reception team?
- It can cover the calls a reception team cannot — nights, weekends, and overflow when everyone is already on a call — and handle routine qualification and booking. Treating it as a full replacement usually degrades the experience for the callers who most need a person, so we scope it as coverage rather than replacement.
- Do we have to tell callers they are speaking to an AI?
- We configure disclosure at the start of every call by default. It is the honest baseline, and in a growing number of jurisdictions it is also a legal requirement.
- Can the chatbot and the voice agent share the same knowledge base?
- Yes, and they should. The retrieval layer is the expensive part to build and maintain, and running one source of truth across both channels avoids the situation where chat and phone give different answers.
- What happens when the AI does not know the answer?
- It should say so and hand off, not improvise. We configure escalation to email, WhatsApp or a phone transfer, and log the unanswered question so the gap in your content gets filled.
- Which is cheaper to run?
- Chat, considerably. It bills per message rather than per minute and handles concurrent conversations without extra infrastructure. Voice justifies its cost when each captured call is worth a lot — a booking, a qualified lead, an urgent job.
Work with us on this
We start every engagement with a written scope after a free discovery call.