If your support team spends the day answering "where's my order", "do you ship to Gdańsk", "what are your opening hours" and "how do I return this" - you are paying skilled people to do work a machine handles in seconds. In most businesses I look at, 70-80% of incoming tickets are this kind of routine, and it does not need a human at all. An AI support agent built on your own knowledge base can close that 70-80% across website chat, WhatsApp and Telegram, and hand the genuinely hard cases to a person with the full conversation already attached. Below is the cost math, what the bot can and can't do, and how I actually build one.
Where the money really goes
Let's count a normal Polish support setup. A junior support person costs you, fully loaded (salary, ZUS, taxes, equipment, the manager's time), somewhere around 6000-8000 zł a month. A team of five is therefore 30 000-40 000 zł every single month, before you count holidays, sick leave and the constant turnover that means you are training someone new every quarter.
Now look at what that team actually does. Pull a week of tickets and tag them. Almost always it splits like this:
- 70-80% pure routine: order status, delivery, returns policy, prices, opening hours, "is X in stock", "how do I pay with BLIK".
- 15-20% needs a human but is still simple: a small complaint, a change to an existing order.
- 5-10% genuinely hard: an angry customer, a non-standard case, something that needs judgement.
You are paying full salary for all of it, but only the last 20-30% truly requires a trained person. The routine is the part you are overpaying for, and it is exactly the part a bot does well.
What an AI support agent actually does
This is not the old "press 1 for sales" menu bot. I build an agent that reads a real knowledge base - your FAQ, returns and delivery rules, product info, order data from the store or CRM - and answers in plain language, in Polish, English, Ukrainian or Russian, whichever the customer writes in.
What it handles on its own:
- Order status and tracking (pulled live from your store or InPost).
- Delivery times, shipping costs, Paczkomaty vs courier.
- Returns and complaints process, step by step.
- Product questions, availability, sizing, comparisons.
- Prices, promo codes, payment methods (BLIK, Przelewy24, card).
- Opening hours, address, booking a call or a slot.
When it hits something it can't answer confidently, it does not guess and it does not invent a policy. It escalates to a human - on WhatsApp, Telegram, email or your CRM - and passes the whole conversation along, so your person reads three lines of context and replies, instead of starting from "hi, how can I help". One agent works the same on the website widget, on WhatsApp and on Telegram, so you set it up once and it covers every channel.
The honest part: it augments, it doesn't replace everyone
I am not going to sell you a fantasy where you fire the whole team on Monday. A bot that tries to handle 100% of cases will lie to customers on the 5% it shouldn't touch, and that costs you more than the salary you saved.
What actually happens: the bot takes the routine 70-80%, and your people stop drowning. A five-person team often becomes a two- or three-person team handling only the cases that need a human - the complaints, the judgement calls, the upset customer who needs a real person. The remaining people do better work because they are not answering "where's my parcel" forty times a day. So the saving is real, but it comes from not growing the team and from shrinking it carefully, not from a clean sweep.
Quick math on the same five-person team. Say the bot closes 75% of tickets. You realistically go from five people to two, keeping the two best for the hard cases. That's roughly 18 000-24 000 zł a month off the payroll. The bot costs a one-off setup plus a small monthly fee (numbers below). Even on the cautious end it pays for itself inside the first two months, and after that it is pure saving every month - plus the team that's left is happier and stays longer.
How I build it and what it costs
The work is mostly about the knowledge base and the escalation rules, not the chat window.
- Knowledge base. We gather everything the bot needs to know - FAQ, policies, product data, common cases from your real ticket history. The better this is, the more the bot closes. This is where most of the effort goes.
- Connections. I wire it to your store or CRM so it can read live order data, and to the channels you use: website chat, WhatsApp, Telegram.
- Escalation logic. We define exactly when the bot hands off to a person, and where that lands. No confident guessing on policy or money.
- Tuning. For the first few weeks I watch where the bot trips, add to the base, and push the close rate up. A bot is not "buy and forget" - the first month of tuning is what gets you from 50% to 75%.
Cost: an AI agent like this runs from €900-2500 to set up, depending on how many channels and integrations you need, plus a small monthly fee for hosting, model tokens and keeping the knowledge base current. Set against 18 000-24 000 zł a month in salary, the math is not close. Full details are on my AI agents service page, and I give a fixed quote for your case within 24 hours.
A real example: for a client's leadbot project I built an agent that took the repetitive load off the team and saved them about 210 hours a month - time that used to vanish into copy-paste replies and now goes to actual work.
If your support team is buried in the same questions every day, send me your ticket volume and I'll tell you honestly how much of it a bot can take and what it would save you. Start on the AI agents service or just get in touch - the quote is free.
FAQ
How much can an AI agent really cut support costs? In most businesses 70-80% of tickets are routine, so a well-built agent closes that share and lets a five-person team shrink to two or three. On a Polish payroll that's typically 15 000-24 000 zł a month saved, against a setup from €900-2500 plus a small monthly fee.
Will the bot replace my whole support team? No, and you shouldn't aim for that. It takes the routine 70-80%, while people keep the complaints, the judgement calls and the sensitive cases. The team usually shrinks and stops growing rather than disappearing - the bot augments your people, it doesn't fully replace them.
What happens when the bot can't answer? It escalates to a human instead of guessing. The customer gets handed to your person on WhatsApp, Telegram or your CRM, with the whole conversation attached, so your team replies with full context in seconds.
Which channels does it work on? The same agent runs on your website chat, WhatsApp and Telegram at once. You set it up once and it answers customers wherever they write, in Polish, English, Ukrainian or Russian.
How long until it pays for itself? With routine-heavy support, usually inside the first two months. The saving on salaries from cutting even two positions far outweighs the setup and monthly cost. If your tickets are all complex one-offs, it pays back slower - and I'll tell you that upfront.
