Nine in the evening, a visitor is on your site wondering whether you have a slot on Thursday, whether the property on Storgatan has a lift, or when her car will be ready. Nobody answers. By the next day she has found someone else.
Most questions a chatbot gets are the same five: what does it cost, how long does it take, do you have availability, what is included, how is my case going. The last two have one thing in common – the answer is not on the website. It is in a system.
Three things the chatbot does every day, without anyone sitting at it.
From your own information: your website, your documents, what you typed in yourself. Not from the internet at large.
Pulls current data out of your systems and takes the visitor on to the right next step.
When a question needs a person it collects contact details and emails you straight away, with the whole conversation.
You describe how you work. The platform does the rest.
Give it your address and the platform reads the site itself: sitemap, subpages, up to 500 pages. It respects your robots.txt. You can add documents, price lists and your own questions and answers.
What the chatbot should help with, what tone it should take, how much it may sell, and what it must never answer. You approve the instruction before anything is published.
Paste it before </body>, or use our WordPress plugin. Then it is live.
The questions worth the most are usually the ones the website cannot answer.
A website is static. Your systems know what actually applies right now: which slots are free, what is in stock, how far a case has got, what details belong to a particular property.
Svea Interact connects to those systems. When a visitor asks something that needs a real lookup, the chatbot does it mid-conversation and answers with what is true now – not with a link to a page where she has to go looking herself.
Free slots from the booking system, booked right in the chat.
Details about a listing from the property system, book a viewing, take a registration of interest.
Status on a job in progress, free workshop slots, a price for a routine repair.
Take a request for a quote with a photo, register it in your system, book a site visit.
Current assignments, the status of an application, a registration of interest.
Free slots, contract details, changes to a booked visit.
Bokadirekt: services, free slots and booking right in the chat. Email via Gmail, Microsoft 365 and IMAP.
We connect them for you. If the system has an API we set the connection up; you do not need to know how an API works. Tell us which system you run and we will say plainly what it takes.
The platform works out how your system behaves. We set up the connection. You approve it before it goes live.
If you have technical documentation, pointing at it is enough. The platform reads it itself.
It works out how sign-in works and which lookups are possible, then suggests which ones are worth connecting.
You see exactly what the chatbot would answer before anything goes live.
And we keep it alive. If the sign-in expires it renews automatically, and if the system stops responding we notice.
Up to five systems per chatbot, with twelve lookups in each
An ordinary chatbot answers and stops there. Svea Interact knows what you want to achieve and steers the conversation there, without becoming pushy.
You set your goals in priority order: a booked appointment, a request for a quote, a registration of interest, collected contact details. The chatbot recognises what the visitor is after and steers towards the right goal.
It does so in steps. First it answers the question. Then, once the visitor shows interest, it suggests the next step. If the visitor says no it backs off – and does not ask again until there is a reason to.
You rank what matters most. The chatbot knows which goal applies when several are possible.
Five steps from curious visitor to customer. The chatbot does not skip steps and does not push someone who is not ready.
If the visitor mentions her car, her address or the service she is after, the chatbot remembers it through the whole conversation, and fills it in for her on the form.
If it sounds urgent the chatbot puts the sales step aside and helps straight away.
The chatbot asks once. If it gets a no it returns to answering questions, and only raises it again if the visitor shows fresh interest.
And when the material runs out it says so, instead of inventing something.
The chatbot draws its answers from your own material: your website, your uploaded documents, and the free text you type in yourself. It does not search the internet. What you typed by hand carries the most weight – if the chatbot finds both a page on the site and your own correction, it picks your correction.
You choose whether the chatbot may only use your material, or whether it may also answer with general knowledge on questions that fall outside it.
The chatbot can only link to addresses it actually knows from your material. Links to other domains are stripped automatically, before the answer reaches the visitor.
Every passage the chatbot has learned can be changed or removed in the dashboard. You do not have to re-crawl the whole site to correct one price.
Change something on the website and you press refresh, and the chatbot rereads the page. Nothing changes behind your back.
You get an email straight away, with the whole conversation.
If the visitor asks to speak to someone, or the question moves into an area you told the chatbot to stay away from, it stops and collects contact details instead.
You get an email immediately, whatever notification settings you chose otherwise. That alert always gets through – a visitor who asked for help should not end up in an inbox nobody watches.
Handover is built in and needs no setup – it triggers when the visitor asks to speak to someone, or when the question touches a topic you have flagged.
A leaking pipe, a damaged bumper, a damp patch in a bathroom. The visitor takes a photo on her phone right in the chat, and the chatbot sees what is in the picture and can answer on it.
The picture travels with the case when it is handed over, and can be attached to the form automatically, so the request for a quote arrives complete the first time.
Up to two pictures per message, max 2 MB each
You explain in ordinary words what went wrong. The chatbot shows how it would answer instead, and only saves once you say yes.
If you see an answer that is wrong, you just say so. The platform works out whether it is a factual error, an answer that should be phrased differently, or the chatbot being too sales-driven – and shows you the result before anything changes.
Nothing is written to the chatbot’s knowledge without you confirming it first.
The chatbot keeps its own list of questions it could not answer. Real visitor questions sit at the top, most recent first, and a question that recurs is counted up rather than listed twice. That list is what makes the chatbot better, and it is filled in by your visitors, not by us.
One line of code, or our WordPress plugin.
The chatbot requires your site to run on HTTPS
Not just how many wrote, but how many became customers. Every visitor is followed through four steps: saw the chat, opened it, used it, converted.
| Metric | Description |
|---|---|
| Conversion funnel | Seen, opened, used, converted – unique visitors |
| Per goal | How many conversations matched the goal, how many moved forward, how many converted |
| What visitors clicked | Split by type, so you see what the chat actually drives |
| Collected leads | With the whole conversation behind them |
| Unanswered questions | What visitors asked that the chatbot could not answer |
| Mobile versus desktop | Where your visitors actually are |
| Per chat and per channel | If you run several chats, or the email assistant too |
We show what share of conversations were handled without being passed to a person. That is not the same as the answer being right – we would rather say so plainly than pretend otherwise.
Named and checkable, not a badge.
Your database sits on a server in Stockholm. Conversations, leads and the knowledge base are stored there.
The AI answers themselves are generated at our model provider. Your customer data is never used to train language models.
So you know before you decide.