What a Co-lega AI employee actually does

We get the same question on most first calls. "So is this a chatbot?" The honest answer is: it can be, if that's all you want. But that isn't what we sell, and it isn't what most of our customers end up with after week one.

Co-lega is an AI employee. We pick a role (Shop, Support, or Receptionist), we set it up against your site and your content, we test it, we put it live, and we tune it every month based on what your customers actually asked. The thing on your site is the visible part. The work that makes it useful happens off-screen.

This post walks through what the role covers, what it deliberately doesn't, what month two looks like, and the cases where we'd tell you to hire a person instead.

What the role covers

Concretely, on a customer site today, the AI employee does work in four areas:

• Answers product, pricing, and policy questions from a knowledge base we build from your site, your docs, and the catalogs we sync. When the answer isn't in the knowledge base, it doesn't guess.
• Qualifies inbound leads during and after business hours. Visitor asks about a custom job, the AI captures the brief, the contact details, and the urgency, and routes it to your team.
• Handles repeat questions so your team stops answering "do you ship to Slovenia" for the 400th time. Roughly half of inbound chats in a typical webshop are answerable from the FAQ.
• Escalates when it doesn't know. This is the most important behavior. The agent has an explicit "I don't know, but I'll get someone who does" path. It doesn't invent.

Three roles, same underlying engine, different defaults. We lead with the first one because it's where AI moves the needle fastest:

• Shop Employee for webshops and product businesses. Catalog-aware, recommends 2–3 products that fit a customer's stated need, explains delivery and returns from your policies, captures bulk and B2B leads, escalates safely when unsure.
• Support Employee for service businesses. Documentation-aware, escalation-rule-tuned, designed to deflect simple tickets and route hard ones to the right human.
• Receptionist for law, accounting, clinics, real estate. Calendar-aware (when integrated), screens inquiries, captures intent, books or routes.

What it doesn't cover

Equal time on this. The same agents:

• Don't replace your support team. They take the top of the ticket funnel. Anything substantive still goes to a human.
• Don't make commercial commitments. The agent doesn't quote prices that aren't in the catalog. It doesn't promise delivery windows the warehouse doesn't honor. When asked, it says "let me get someone who can confirm."
• Don't run on a free tier. There isn't a hobbyist plan. The work to set one up well isn't compatible with a $39/month price point.
• Don't speak every language well. English and Croatian are first-class today. Other languages work but get less tuning attention. If your site is in three languages and one of them is Vietnamese, ask us about it specifically.

How we set it up

The first week is a setup week, not a self-serve week.

1. Brief and scope. We agree on the role, the top 20 questions visitors actually ask, what counts as escalation, and the tone. This is one call, sometimes two.
2. Knowledge ingestion. We crawl your site, sync your catalog if you have one (WooCommerce or Shopify out of the box, custom by arrangement), and pull in any policy docs you send us.
3. Voice tuning. We give the agent your tone. The default is calm and direct. We adjust based on a handful of sample exchanges with you.
4. Test pass. Before go-live, we run a quality test: 30 to 50 scripted questions covering the cases that matter to you. Anything that fails gets fixed or explicitly marked as "should escalate."
5. Live. You drop a single line of script on your site. We monitor the first week closely. If we see something off, we tune it. You don't have to do anything.

Setup window is usually five to seven business days from kickoff to live, sometimes faster if your content is already clean.

What month two looks like

The thing that makes a Co-lega employee worth €200+ a month isn't the setup. It's that someone is watching.

Every month, you get a quality report: what visitors asked, what the agent answered well, what it missed, what it escalated, where we'd tighten the prompt or extend the knowledge base. On Growth and Managed plans we get on a call to walk through it. Then we make the changes.

This is the part DIY tools don't do. You can buy a chat widget for €39. What you can't buy at that price is someone who reads the transcripts the next morning.

When it isn't a fit

We disqualify customers regularly. We'd rather lose a sale than ship the wrong tool. Cases we'd tell you to hire a person instead:

• Regulated high-stakes advice. Medical diagnosis, legal opinions, financial planning. The agent should screen and route, not advise.
• Voice and phone. Our employees are text-first. If your main inbound channel is the phone, we aren't the right fit yet.
• Heavy negotiation flows. Custom enterprise deals where every conversation is bespoke. The agent will hand off, but if every conversation hands off, you didn't need the agent.
• Sub-10 conversations a month. If your traffic is low enough that you'd see every chat anyway, you're paying for monitoring you don't need. Hire later.

Pricing anchor

This sits in the €199 to €599+ band, per role per month. The right anchor isn't a chatbot. It's a part-time team member who covers the same ground at a fraction of the cost, can't take sick days, and gets a tuning pass every month from someone who watches every transcript.

If the price feels wrong, we're probably the wrong tool. The right customer for Co-lega already knows what missed leads after 4pm cost them.

See it live

The fastest way to evaluate is to use one. We have a live demo on a real customer site you can click through, ask questions in Croatian or English, and watch it escalate.

When you're ready to talk about your own site, request a demo and we'll get on a call.