AI Chatbots

Chatbots that know where answering ends

A bot that answers from your own content, and hands off cleanly the moment a question needs a judgement call or a system change.
Talk to us about chatbots
In short

An AI chatbot answers questions from your own content, for customers or for staff. Webcoda builds chatbots with a deliberate line between answering and acting: if a question needs a judgement call or a change to a system, the bot hands off cleanly rather than guessing.

Who this is for

Scenario 01
The same support questions, every day
A handful of questions make up most of your volume, and answering them is somebody's whole afternoon.
Scenario 02
A knowledge base nobody searches
The answers exist, in an intranet or a help centre, and staff ask a colleague instead because it is faster.
Scenario 03
A customer-facing bot giving wrong answers
Something is already deployed, it is confidently out of date, and nobody owns the content behind it.

What we deliver

  1. 01

    A bot trained on your content, not the open internet

    Answers come from sources you nominate, with the source shown, so a wrong answer is traceable to a page you can fix.

  2. 02

    A clear escalation path

    To a human, or to an agent, for anything past answering. Defined during design, not bolted on after the first complaint.

  3. 03

    Integration into your existing platform

    Your support desk, intranet or CMS. Conversations land where your team already works, with the context attached.

  4. 04

    Content freshness checks

    The failure mode of every chatbot is stale content. We build the review loop that catches it, and name who owns it.

Answering versus acting

A chatbot answers An agent acts
Retrieves and explains what your content already says. Takes a judgement call inside a process and does something about it.
Wrong answer means a bad experience and a correction. Wrong action means a changed record, and something to unwind.
Supervision is content review. Supervision is monitoring, permissions and an audit trail.
Right when the volume is questions. Right when the volume is decisions.

A chatbot that starts taking actions without you deciding to build that in is a chatbot doing an agent's job badly. If you need action, not just answers, that is a different service.

AI Agents

What we have built

Webcoda.ai site assistant

We built this for our own site, and it changed how people find things. Instead of digging through menus or a search box, visitors just ask. The assistant uses RAG to answer from our actual pages, so every answer is accurate, current and in our words, not a canned script.

Your customers get instant answers, you get fewer "where do I find" emails.

Chatbot-4

How an engagement runs

  1. Step 01

    Assess

    We read your last few hundred real questions. What people actually ask, in their words, and which of it your content can answer today.

  2. Step 02

    Prototype

    We build against that question set and score the answers. Accuracy measured on real questions, including the ones it should refuse.

  3. Step 03

    Integrate

    We deploy with the handoff wired in. Escalation to a person with the transcript attached, and source links on every answer.

  4. Step 04

    Scale

    We widen the scope as the content earns it. New topics get added when the underlying content is good enough, not when someone asks for a new tab.

Honest scoping

Shape
Content review first
If the answers are not written down anywhere, the first piece of work is content, not a bot.
Shape
One audience to start
Customers or staff, not both. The tone, the risk and the escalation path are different.
Ongoing
Running cost is real
Model usage plus content ownership. Small, but not nothing, and worth naming before launch.

Chatbots, answered

A chatbot answers questions from content. An agent takes a judgement call and acts on a system. Most organisations need the first one, and buy the second because it demos better.

It can, and that is the point where it stops being only a chatbot. Reading a record to answer a question is reasonable; changing one is agent work, with the supervision that implies.
It says so and hands off, with the transcript attached. A bot that guesses to avoid an awkward silence is worse than no bot, because it teaches people not to trust it.
The ones already in your stack, most often HubSpot, Microsoft 365 and SharePoint, or your CMS. We would rather add to the tool your team already watches than introduce another inbox.
Yes, though the accuracy depends on whether your source content exists in that language. Translating on the fly is possible and needs a review process before you rely on it.

Talk to us about chatbots

Bring your last hundred support questions. That conversation is more useful than a demo.

Contacts

For all your AI Integration, HubSpot, Umbraco, Web Development, Web Design, Intranets and digital marketing projects, connect with us. Tailored solutions for success.

Logo-hubspot Logo-umbraco Logo-B-Corp Logo-xperience Logo-microsoft