What does an AI-enabled website actually do?
Not a chat bubble. A page that answers from your own material, qualifies, books, and knows when to stop.
The short answer
An AI-enabled website answers a visitor’s question from the business’s own material — prices, policies, stock, past enquiries — then qualifies what they need, books the call or takes the order, and hands over to a person when it should. The difference from a chatbot is what sits behind it: a general model with no access to your content will answer confidently and wrongly about your lead times and your terms, which is worse than no chat at all. A useful one is grounded in your documents, cites where an answer came from, refuses what it cannot support, and escalates anything worth money. Most of its value is earned outside office hours, on traffic you have already paid for.
What is actually different from a chatbot?
The old scripted bot matched keywords to canned replies, and everybody learned within two clicks that it could not help. The new failure is the opposite and more expensive: a general model connected to nothing, answering fluently and inventing your delivery times.
What makes one useful is unglamorous. It reads from your material rather than from the internet. It says where an answer came from. It refuses when your material does not cover the question, and takes a message instead of guessing. It has a defined boundary — the things it must never handle alone — agreed before it goes live.
The single most useful question to ask any vendor demonstrating one: "ask it something it should not know, in front of me."
What does it do in a normal week?
- Answers the same fifteen questions it gets asked constantly — lead times, coverage, sizes, compatibility, what happens after an order — in seconds, at any hour.
- Qualifies while it answers: quantity, timeline, location, budget band, asked in the order a person would ask them.
- Books the call straight into the calendar with the context attached, so nobody repeats themselves.
- Captures what it could not answer, which quietly becomes a list of what your website fails to explain.
- Hands complaints, negotiations and anything unusual to a human immediately, with the transcript.
Where does the value actually come from?
Rarely from novelty. It comes from three unremarkable places: enquiries that arrive when nobody is working, questions that were costing a person an hour a day, and the traffic you already pay for converting slightly better because somebody answered before the visitor left.
That last one is why the homepage is usually where it earns first. The traffic is bought and paid for; the marginal cost of answering it properly is close to zero.
It is also measurable in a way most website changes are not: conversations held, questions unanswered, enquiries captured outside office hours, meetings booked without a human. If a vendor cannot show you those four numbers after a month, they have not instrumented it.
What does it cost to run?
Two costs, and only one of them is obvious. The model and infrastructure spend scales with use — for a normal business website it is small, but it is not zero, and a system nobody modelled can cost more per month than the build did. Ask for the monthly figure at the volume you actually expect, and what happens if usage triples.
The second cost is attention. Something has to watch whether answers are still correct after your prices change, your policy is updated, or the model underneath is retired by its provider. A system nobody is watching in month three is the normal way these fail.
We bill model and infrastructure costs at cost, with no markup, under a ceiling agreed before anything goes live. Any vendor unwilling to do the same is taking a margin on your usage.
When is it the wrong thing to build?
- When the traffic is not there. A site with forty visitors a month does not have a response-time problem; it has a demand problem, and an agent on it is decoration.
- When the honest answer to most questions is "it depends on a survey" — construction, bespoke manufacturing, complex insurance. The agent can qualify and route, but it should not pretend to quote.
- When the underlying content does not exist. If your prices and policies live in somebody’s head, the first job is writing them down, and that job is worth doing whether or not anything is built on top.
- When the real problem is the offer. If the page is not converting because the proposition is wrong, a faster answer will not fix it — and we would rather say so.
How long does one take to build?
A working proof of concept on your own homepage takes about two days — we build them free, because arguing about a live thing is faster than arguing about a proposal.
A production version is usually one to two weeks. What takes the time is not the model: it is gathering the material it should answer from, agreeing what it must never say, and connecting it to the calendar or CRM where the outcome has to land.
Questions this raises
Will it say something wrong about our prices?
Not if it is built properly: it answers from the material you give it and refuses what that material does not cover. The failure you are imagining comes from connecting a general model to a website with no grounding, which is a different product built badly. Ask to see the refusal behaviour before you see the happy path.
Do we need to write new content for it?
Usually not. Most businesses already have what it needs — a price list, a delivery policy, specification sheets, a year of email replies answering the same questions. If a topic genuinely is not written down anywhere, the agent will tell you, which is the cheapest content audit you will ever get.
Can it work on WhatsApp too?
Yes, and in India that is often where it matters more than the website. The same agent and the same material can run on WhatsApp, on a phone line, or both. The homepage is usually first because the traffic is already there and already paid for.
How do we know it is working?
Four numbers, from the first week: conversations held, questions it could not answer, enquiries captured outside office hours, and meetings booked without anyone intervening. The second number is the useful one — it is a list of what your site fails to explain, and it tends to be worth more than the agent.
This is what we do about it
More like this
What should you check before signing a software quote?
Five things that decide whether a build goes well. None of them is the price.
How fast can software actually be built with AI tooling?
Weeks instead of quarters is real. Which parts compress, which parts refuse to, and where the time goes instead.
Why do software firms not publish prices?
Mostly for bad reasons. There is one good one — and either way, here is how to get a real number out of them quickly.