Custom AI assistant to answer your visitors
A custom assistant answers your visitors' questions from your own information, not from general knowledge. FastSolve builds it and sets its limits, so that it helps instead of inventing.
You can see the result before discussing it: the link below is available online, with no sign-up required.
The same questions, every day
Four situations come up before almost every assistant project.
- Three quarters of the messages you receive ask one of the same five questions, and the answer is already on your website.
- Enquiries arrive in the evening and at weekends. Somebody waiting for an answer until Monday has often written elsewhere in the meantime.
- Your visitors cannot find information that is there, because it sits in the middle of a page they do not read.
- You receive off-topic enquiries that take as long to handle as the real ones.
An assistant is not there to look modern. It exists to answer the repetitive part immediately, and to pass on only what genuinely needs your attention.
Four things that make an assistant useful
The difference between a useful assistant and an irritating gadget comes down to these four points.
1. It only knows your information
The assistant works from your text, your prices, your terms, your hours. It does not improvise from something it read elsewhere. That limit is what makes it reliable, and it is the first setting of the project.
Typical case. A supplier had tried a generic assistant that promised lead times he did not offer. Its replacement quotes only the times written in his terms.
2. It says when it does not know
This is the most important behaviour, and the rarest. Faced with a question outside its scope, a good assistant admits it and offers to pass the message on, rather than producing a plausible and wrong answer.
Typical case. A tradesman feared an assistant would commit him on a quote. The assistant systematically refers pricing to a human exchange, and says so plainly.
3. It qualifies before passing on
Instead of an empty message, you receive a described need, with the context, what the person is looking for and by when. You reply to enquiries that have been prepared.
Typical case. A company received one-line messages with no context. The assistant now collects the trade, the need and the deadline before passing on, which removes the first round trip.
4. It stays under your control
You see what was asked and what was answered, and you correct the frame when an answer does not suit you. An assistant is tuned over time, it is not delivered once and for all.
Typical case. A company discovered, reading back the conversations, that its visitors were mostly after information absent from the site. The page was added, and the assistant now uses it.
What an assistant will not do
It answers what is written. It does not replace your judgement.
- It does not replace an adviser. On a specific case, a dispute or a negotiation, it must hand over, and it is set up to do so.
- It does not guess what you have not written. The quality of the answers depends directly on the quality of the information supplied, which makes scoping decisive.
- It does not commit you. Firm prices, contractual deadlines, commercial promises: those subjects are deliberately referred back to you.
- It is not free to run. Every conversation has a cost, small but real, paid to the model provider. That is estimated during scoping and appears on the quote.
The gain is in the repetitive questions and in the quality of the enquiries you receive, not in the decisions, which stay yours.
Frequently asked questions
Will it make up answers?
That is the main risk and it is dealt with technically. The assistant is constrained to draw on your documents rather than on its general knowledge, and it is set up to acknowledge when it does not know. We test that behaviour before delivery, with deliberately off-topic questions. No assistant is infallible, which is why sensitive subjects are referred back to you by design.
How much does a custom assistant cost?
There are two distinct amounts. The build, which depends on the volume of information to take in and the number of behaviours to set. And the running, billed by the model provider per conversation, generally modest but proportional to use. Both are costed during scoping and appear separately on the quote. As a guide, excluding VAT, the build most often falls between 2,000 and 6,000 euros. Running costs generally amount to a few euros to a few dozen euros per month, depending on usage.
What information does it draw on?
Whatever you supply: pages from your site, internal documents, questions already received, terms of sale. An assistant is worth what those sources are worth, and it is often a chance to update texts that had fallen behind.
What happens to the conversations?
They are available to you, and that is one of the most useful parts: they show what your visitors are really after. Retention and access are decided during scoping, all the more so as those exchanges can contain personal data.
Can it do more than answer?
Yes. It can pass on a qualified enquiry, point to the right page, or trigger an action in a tool you use. Every capability added needs its own guard rails, so it is decided explicitly.
Should I open it to logged-in clients?
That is possible, inside a client portal, where it can then answer about the logged-in person's file. It calls for more care, since it touches private data, and is handled as a separate project.
Tell us the five questions you are asked endlessly
See how a project runs, from scoping to delivery
See all FastSolve services