"It will replace your manager" is the kind of claim that makes me ask for numbers. Because the answer depends entirely on what that manager spends their day doing. If eighty percent of it goes on "how much is it", "are you open on Saturday" and "where is my order", then yes, it will replace that and pay for itself in a couple of months. If they are negotiating and closing five-thousand-euro deals, then no, and it should not.
I build AI agents and I run the payback numbers before starting, because a client deserves to know what they are buying. Here is the method and the honest limits.
What a bot closes completely
Here it outperforms a person and costs pennies:
- Routine questions about prices, services, availability, terms, opening hours, address, delivery, returns.
- Order or booking status, once connected to your database or CRM.
- First-pass qualification - what they need, what budget, what timeline, whether they are a fit at all.
- Booking an appointment with a live check of free slots.
- Capturing an enquiry into the CRM with a notification to the right person.
- Nights, weekends and lunch breaks - here it has a monopoly.
Across my projects that is 50-70% of all first-line contacts, and in large-catalogue online stores it reaches 80%.
What it closes partially
Here the bot prepares the ground and a human decides:
- Non-standard quotes where scope has to be judged.
- Complaint intake: gather the facts and hand over, but apologising and resolving stays human.
- Complex selection where experience matters more than spec sheets.
- Anything that is an exception to the rules.
What it will not replace
This is where I usually talk clients out of it:
- Negotiation and objection handling when the decision carries real money.
- Relationships with regulars and partners.
- Accountability. A promise that could be expensive should come from a person.
- Expertise that is not written down. A bot answers from your knowledge base, not from ten years of accumulated instinct.
If your sales run on personal relationships, a bot is an assistant, not a replacement. That is fine: it clears the routine so your people do what you hired them for.
The math on a concrete example
A service business in Poland: 500 contacts a month across the website, phone and messengers, two people on the front line.
Current cost. Full employer cost per person, taxes included, is roughly €1,800-2,300 a month. Call it €2,000. So €4,000 a month total. Coverage: Mon-Fri, 9:00-17:00, which is about 170 working hours out of the 720 in a month.
What the bot does. The contact logs show 60% are routine: prices, timelines, availability, status, booking. The bot closes those outright. That frees up roughly 0.8 of a full-time role, so about €1,600 a month in direct time savings.
What gets added on top. Typically 30-40% of contacts arrive outside working hours: evenings, nights and weekends. They used to be answered the next day, and some of those people had already gone to whoever replied first. If 175 contacts arrive out of hours and you were losing even 15% of them, you recover around 26 enquiries a month. At a €300 average order and a 30% close rate, that is another €2,300 in revenue.
What the bot costs. Implementation with a knowledge base, CRM and calendar integration: €3,000 one-off. Model usage and hosting at that volume: €100-200 a month.
Result. The savings plus recovered enquiries cover the build in 2-3 months; after that it is €150 a month against €1,600 of freed-up time.
An important caveat: the people usually do not go anywhere. They get moved onto work nobody had time for - repeat sales, working the existing base, handling complex enquiries properly. In my experience that produces more money than cutting headcount.
When a bot will not pay back
Saying this plainly so you do not waste money.
- Under 100 contacts a month. The time saved is too small to cover the build in a sensible timeframe. Fixing your website form and notifications is cheaper.
- Every enquiry is bespoke. Project-based B2B where each request needs a real estimate. A bot helps with qualification, but there is no front-line saving.
- No knowledge base. If prices, terms and rules live in the owner's head, the bot has nothing to answer from. That needs writing down first - which, incidentally, is worth doing anyway.
- Broken processes. Automating chaos gets you faster chaos. I wrote about this in what to automate first.
What matters more than the savings
Often the main effect is not on the payroll line.
Response speed. A customer who messaged three companies buys from whichever answered first. A bot answers in seconds, every time.
Consistent quality. Your manager on Friday evening and on Monday morning are two different people. A bot is not.
Nothing gets lost. Every contact lands in the CRM with the full conversation. Missed calls and unnoticed Instagram messages are their own painful topic, covered in how not to lose customers to missed calls.
Data. After a month you have an exact list of what customers actually ask. That usually reshapes both the website and the price list.
What it looks like in practice
The working setup I build: a knowledge-base bot trained on your own materials answers on the site and in messengers, qualifies, books or captures the enquiry, writes everything to the CRM and pings the right person in Telegram. Anything complex or unusual goes straight to a human - escalation rules are configured separately, and there is a dedicated piece on that.
Timeline: 2-3 weeks from kickoff to a working bot. Cost: from €1,200 for a simple scenario to €7,000 for a complex multi-channel build with integrations - detail in how much an AI chatbot costs. Approach and examples are on the AI agents service page.
If you want the numbers run against your own figures, message me with your contact volume, average order value and opening hours. I will do the payback calculation and tell you honestly if it does not add up.
FAQ
What share of enquiries does an AI agent really close without a human? On the front line, usually 50-70%, and up to 80% in large-catalogue online stores. That covers routine questions: prices, services, availability, terms, hours, order status, bookings. The exact share depends on how complete your materials are and how repetitive the enquiries. You can estimate it in advance: take a month of chats and calls and count how many would have been closed by a ready answer from your price list and FAQ.
How much does a bot save in money terms? It is calculated from freed-up front-line time plus enquiries recovered outside working hours. A typical calculation for a service business with 500 contacts a month: roughly 0.8 of a role freed up (about €1,600 at a €2,000 full employer cost) plus the enquiries previously lost at night and on weekends. With a €3,000 build and €100-200 a month running cost, payback lands in 2-3 months.
Will I have to let staff go? Usually no, and I do not recommend building the project around that. In practice people get moved onto work nobody had time for: repeat sales, working the existing customer base, complex enquiries, service quality. That earns more than the payroll saving, and the team stops resisting the rollout. Cutting headcount only makes sense where the front line was staffed for peak load.
Below what volume is a bot pointless? The rough line is under 100 contacts a month. The time saved is too small to justify the build within a sensible period, and it makes more sense to first fix your website form, notifications and response speed. The exception is a business with high-value enquiries, where even a few recovered contacts a month cover the cost, or a company whose traffic peaks at night and on weekends.
What will a bot definitely not do? Run a negotiation where you need to read the other person and haggle. Take responsibility for a promise that carries real cost. Answer something that is not in your knowledge base - and that is correct behaviour, because the alternative of an invented answer is far more dangerous. Defuse a conflict with an unhappy customer: it can collect the facts, but apologising and resolving has to be human.
