An assistant that answers from your content. Not from the internet.
It reads your product catalogue, your documentation, your policies and your FAQs, and answers the questions people actually ask. It shows where each answer came from. When your material does not cover the question, it says so and hands the conversation to a person instead of guessing.
In daily production for more than one customer, on a public advice site and on a corporate site
What it does for you
Your visitors ask in their own words. They get your answer, not a general one.
Grounded in your own material
Every answer is built from documents you supplied : the catalogue, the manuals, the terms, the help pages. Ask it something your content does not cover and it says so, instead of filling the gap with something that merely sounds right.
It shows what it answered from
Each answer points back at the page or document behind it. A visitor can check it, and whoever owns that content can correct the answer by correcting the source. Nothing is a black box that only we can explain.
It hands over instead of guessing
Some questions need a person : a complaint, an individual case, anything contractual. When the assistant cannot answer from your material, it stops and offers the route you nominate : a form, an address, a phone number, a ticket.
It lives in the website you already have
Added to an existing site as a widget in the corner or as a page of its own. No relaunch, no second portal to maintain, no migration of the content it reads.
The content stays yours
Hosted in Europe or inside your own network, depending on what your policy allows. Your documents remain your documents : they are not handed to anyone else as training material.
You find out what people actually ask
Every question is logged. Most customers tell us afterwards that this list, not the answering itself, was the thing that changed how they work.
How a conversation goes
Four steps, and one of them is knowing when to stop.
Step 01
Someone asks in their own words
No keyword syntax, no menu tree to climb. The question can be long, vague or badly phrased, the way real questions are. A visitor who would have given up on your search box gets an answer instead.
Step 02
It looks in your material
It searches what you gave it and nothing else : your catalogue, your documentation, your policies, your FAQs. Content you did not put in scope cannot turn up in an answer.
Step 03
It answers, and names the source
A plain-language answer to the question that was asked, with the page or document it came from attached. The reader can go and read the original, which is often the point : the assistant gets them to the right page.
Step 04
Or it hands the person over
If your material does not answer it, the assistant says so plainly and offers the human route. A visitor who is told “this needs a colleague, here is how to reach one” stays a customer. A visitor given a confident wrong answer does not.
The question log
The most valuable output is the list of questions.
After a few weeks you have something you never had before : what your customers ask, in their own words, in volume, rather than what you assumed they would ask. It shows the gaps in your documentation, the products people cannot find, the wording they use instead of yours, and the questions your team answers twice a day. Most customers change their own content because of it.
What the log tends to surface first
None of this requires a reporting project. The questions arrive as a plain list you can read, sort and export, and the ones with no good answer behind them are the interesting ones : each is a page somebody needs and cannot find. Reviewing that list every few weeks is usually the whole maintenance job.
You stay in control of the content
It knows what you give it, and nothing else.
Scope
What goes in
Product catalogue, manuals and datasheets, policies and terms, FAQs, help articles, the pages of your own site. If you can hand it a document, it can answer from it.
Scope
What stays out
Internal notes, drafts, prices you do not publish, anything a visitor should never be quoted. What is out of scope cannot appear in an answer, and you decide where that line runs.
Upkeep
When your content changes
Update the source and the assistant answers from the new version. There is one place where the truth lives, and it is yours : no second copy to keep in sync, no long retraining exercise for a corrected sentence.
Behaviour
What it will and will not discuss
The subject area is set by you. Outside it, the assistant declines politely and points back at what it can help with. It does not have opinions about your competitors, your prices or the news.
How it runs
Three hosting shapes, one behaviour.
Managed, hosted in Europe
We run it, you point your site at it. The fastest way to a working pilot.
- Data stays in the EU
- Updates and monitoring included
- No infrastructure work on your side
Your own cloud
Runs in your tenant, under your accounts, through your own security review.
- Your identity provider and single sign-on
- Your logging and retention rules
- Your procurement process, not ours
On-premise
Inside your network, for content that is not allowed to leave the building.
- No dependency on an external service at answer time
- Same behaviour as the managed shape
- Suitable where the documents are confidential
How to start
One content area, a few weeks, then decide.
A pilot on a single, well-bounded part of your content is enough to tell you whether this works for your visitors. You are not committing to a rollout to find that out, and the question log alone usually pays for the exercise.
First
Point us at the content
One area to begin with : a product range, a help centre, a set of policies. We tell you honestly whether the material is good enough to answer from before anything is set up.
Then
Run it where people actually are
On the live site, or behind a login for an internal audience. A pilot nobody visits proves nothing, so we would rather put it in front of real visitors on one section than hide it in a test page.
After that
Read the questions, then decide
Go through the log together, fix the content gaps it exposed, and decide whether to widen the scope, add a language, or stop. Whatever you decide, the improved content stays yours.
Composes with the rest of SYN IT
The assistant is one of the governed AI services we run, so it follows the same rules as the rest : scoped access, a record of what happened, hosting where your policy allows. Where an answer needs an action behind it, Agent Manager runs the automation and logs it. And if you are not yet sure the case is there, the Discover phase of our delivery model puts a number on it before anything is built.