Skip to content

Finance and costs20:05

Website chatbot for business: at what query volume is it actually profitable

We calculate the break-even threshold for a chatbot based on the real cost of a manual response. We provide deployment, subscription, and ongoing labor costs for a company handling 15–30 queries a day.

AI-generated image.

Listen to the article summary

Synthetic voice.
0:00
0:00

An open guide based on our direct experience. We provide direction and specific figures, leaving the decision entirely up to you. A service company from the Mazowieckie region, 22 staff members, three in customer support. Every day, between a dozen and thirty repetitive inquiries arrive via web forms, email, and Messenger: opening hours, directions, basic service pricing, required documentation. The two most experienced team members spent 45–60 minutes each day answering them; some messages were left until the next day because no one had the capacity in the afternoon. This is the baseline, not an exception: in many smaller service businesses, a repetitive query takes 4–8 minutes, and the labor cost per hour is 45–70 PLN at total employer cost, based on GUS and ZUS records: average wages and contributions.

AI-generated image.

How they calculated this before implementation

Before deploying a chatbot, the company logged every incoming query for two weeks. It turned out that 60% of messages were variations of twelve questions: opening hours, pricing, service scope, lead time. Handling a single query manually cost 3.5–9 PLN in labor, depending on the person and the day. At fifteen inquiries per day (about 450 per month), the value of time spent on replies was 1,575–4,050 PLN per month. This is the benchmark against which subscription and bot maintenance costs must be evaluated. Without this baseline measurement, every deployment is guesswork.

Cost ranges at Polish labor costs

In Poland, the cost of launching and maintaining a bot depends more heavily on knowledge base quality and system integrations than on raw technology. A rule-based chatbot, meaning a decision tree of questions and answers, costs 0–5,000 PLN when built in-house on a SaaS platform, and 5,000–20,000 PLN when configured by a Polish agency. An AI bot integrated with the company knowledge base (RAG) typically costs 8,000–25,000 PLN for a mid-sized business; full integration with a CRM or booking system API raises the initial investment to 20,000–70,000 PLN. Typical subscriptions? Up to 100 queries per month fit into a free plan or up to 100 PLN; 101–500 queries usually run 100–300 PLN; at 501–2,000 queries per month, plans cost 300–800 PLN. The LLM API cost for a short answer drawn from corporate knowledge is 0.005–0.05 grosz per query: for a month with 1,000 queries, this adds 5–50 PLN. These thresholds reflect actual price lists and implementation standards: Dialogflow CX pricing and OpenAI API pricing show that per-query costs remain negligible until dealing with extensive, multi-turn responses.

The break-even threshold with a labor cost of 45–70 PLN per hour and 4–8 minutes per answer: replying manually to one inquiry costs 3.5–9 PLN. A chatbot costing 300 PLN per month breaks even at 35–85 queries per month. At 15 inquiries per day (about 450 per month), net monthly savings after deducting the subscription amount to 1,300–3,700 PLN. This is the baseline: if queries arrive after hours and the business actively loses customers as a result, the threshold drops even further. Still, 35 queries per month for a 300 PLN bot represents a realistic operational threshold, not a marketing promise.

AI-generated image.

Calculation across four inquiry tiers

For a company in this scenario (15–30 inquiries daily, mostly repetitive), the numbers break down as follows: 100 inquiries per month mean wasted time worth 350–900 PLN; 500 inquiries mean 1,750–4,500 PLN. A chatbot with a 300 PLN subscription and 30 PLN API usage: a fixed cost of 330 PLN. Net savings: 20–570 PLN at 100 inquiries, 1,420–4,170 PLN at 500 inquiries. This math can only be challenged by proving that changing the process does not free up time, but merely shifts it; more on that below. At 50 inquiries per month, no scenario works out; our implementation practice recommendation is to stay on a free SaaS tier or skip automation entirely and improve the website FAQ.

What customer support does after implementation

Time savings make sense only when responsibilities are redistributed. In the case described, one team member took over the bot for 1–3 hours per week: a Monday session reviewing conversations, refining responses, and updating the FAQ database; once a week analyzing escalated queries and unresolved questions. Support staff no longer answer repetitive questions resolved by the bot; instead, they monitor, test, and eliminate errors. In addition, there is an information obligation under Art. 50 of the AI Act to disclose that the user is interacting with an AI system, not a human; mandatory from 2 August 2026: regulation text. Add GDPR compliance: a data processing agreement with the bot provider and a customer privacy notice regarding chat processing. Without these two documents, the cost estimate is incomplete.

Who this scenario makes sense for

This implementation pays off in service companies where inquiries are over 60% repetitive and arrive via websites, forms, email, or Messenger. For an accounting firm handling deadlines and documents, a repair service fielding pricing questions, or a clinic booking appointments, the 450 inquiries per month scenario holds true. There is no economic rationale for a corporate litigation law firm, a custom machinery manufacturer, or a B2B sales rep where every conversation is unique. The key factor is repeatability, not pure volume.

One final point: ROI in this case was calculated from a second measurement sample, ten weeks after launch. The bot resolved 64% of repetitive questions without human involvement, savings reached 1,550 PLN per month net, and the full implementation (4,800 PLN) broke even in month four. This is not brochure marketing using vendor numbers; this is our own verified measurement. This is how every project should be approached: measure, do not guess.

Source materials

Sources consulted during research. The text above is our own; these third parties are not responsible for its content and have not authorised it.

No-obligation call

Schedule a workshop with Paul

Thirty minutes or a full workshop, choose what suits you. Book a slot directly in the calendar and get instant confirmation.

Paul Lazniak

HEXART Founder