An AI chatbot for a Slovak e-shop in 2026 costs €120-€480 one-time + €15-€80 monthly depending on type and conversation volume. A well-set-up AI bot can cover 60-75% of routine customer questions (stock availability, shipping, returns, invoicing), free up 8-12 hours per week for your team, and lift conversion rate by 18-32%. A badly set-up bot loses you both clients and money in the first month.
Three bot types and which to pick
Scripted bots (classic decision tree) are the oldest type, fixed branching logic, the client clicks a question, the bot sends a preset answer. AI bots use a language model (GPT-5, Claude Sonnet 4.7, Llama 4) and generate answers in real time based on your business context. Hybrid bots combine both, routine questions are handled by AI, edge cases escalate to a human.
For a Slovak e-shop with 100-2,000 daily visits, we usually recommend Hybrid. A scripted bot fails on questions you didn't define ahead of time; an AI bot without a human fallback occasionally "invents" product information. Hybrid catches both: 80% handled by AI, the rest goes to a human via Telegram or Slack notification.
Scripted bots fail on questions you didn't anticipate. AI bots occasionally invent. Hybrid catches both.
What the bot actually handles, and what it doesn't
A well-configured AI bot on a Slovak e-shop handles these question types without human intervention:
- "Do you have this product in stock?", the bot pings your stock API (Shoptet, Shopify, Pohoda) and returns the real number.
- "How much is shipping to Košice?", the bot knows your shipping table and returns price + estimated arrival.
- "How do I return an item?", the bot sends a short workflow + return form link.
- "I have an issue with an invoice", the bot grabs the order number and opens a ticket for accounting.
- "What size should I pick?", the bot combines your sizing table with a question about the customer's height/weight.
What the bot doesn't handle (and shouldn't try):
- Price negotiation, always human, otherwise the customer instantly loses trust.
- Complaints about complex issues (wrong delivery, damaged goods), a bot here just adds frustration.
- Communication with VIP clients, if a customer's volume is >€500/mo, they always deserve a human.
- Legal or medical questions, the risk of bad advice is too high.
Real operating costs in 2026
This is where you most often overpay. Slovak agencies sometimes sell an "AI chatbot for €4,800" while in fact integrating an open-source bot with no fine-tuning. DevNova pricing:
- Scripted bot setup: €120 one-time, €0/mo running cost (runs on your hosting).
- AI bot setup: €180 one-time + €15-40/mo depending on volume (1,000 conversations ≈ €18-25/mo via Vercel AI Gateway).
- Hybrid bot setup: €240 one-time + €25-60/mo.
- Custom training on your content (FAQ, products, brand voice): €120-360 one-time.
For context: AI inference in 2026, GPT-5 Mini ≈ $0.40/1M input + $1.60/1M output, Claude Haiku 4.7 ≈ $0.25/1M input + $1.25/1M output. A typical 4-6 turn conversation ≈ 800 tokens = €0.001-0.003. At 1,000 conversations/mo that's €1-3 raw AI cost + Vercel margin = €18-25/mo final bill.
Why 70% of Slovak AI chatbot implementations fail in month one
Most common failure modes (from audits we ran for 6 clients who came to us after a failed implementation by another agency):
- The bot doesn't know your product. If you didn't feed it your catalog, it doesn't know SKU, price, or availability. The customer asks about sneakers in 42 and the bot answers "We have various products." Customer leaves.
- The bot has no brand voice. AI without explicit instruction writes neutral and formal. For your restaurant where customers are addressed casually, a bot writing "Dear Sir" feels alien. Needs prompt engineering.
- The bot has no escalation to humans. The customer types "give me a person" in frustration and the bot replies "Happy to help!" Bad UX. Needs frustration detection + immediate handoff.
- The bot answers too verbosely. AI models tend to be wordy. Four paragraphs to "How much is shipping?" lose the customer. Needs a 1-2 sentence constraint.
Concrete example: AI bot for a Slovak handmade-jewelry e-shop
Anonymized client (handmade jewelry e-shop, 800 orders/month, average €85/order). Before implementation: 2 full-time customer support, 70% of questions same type (stock, sizing, shipping). After hybrid bot rollout:
- AI bot catches 72% of questions without a human.
- Support team shrank from 2 to 1 person, the freed person moved to social media + outreach.
- Conversion rate rose from 1.4% to 1.9%, because a new buyer gets an answer in 8 seconds instead of 35 minutes (until support responds).
- Returns dropped 28%, because the bot proactively shares shipping info and arrival ETA.
Costs: €240 setup + €45/mo AI runtime + €29/mo care plan (Marketing tier) = €74/mo after setup is paid off. Payback: saved 0.5 FTE salary + higher conversion. Payback in 6 weeks.
Who an AI chatbot doesn't make sense for
Be realistic. An AI chatbot doesn't make sense if:
- You get fewer than 50 daily visits, the bot doesn't amortize, customer support via WhatsApp is enough.
- You sell complex B2B services (audit, law, architecture), the client goes to first-call, the bot can't help.
- Your customers are senior (60+), UX acceptance of chatbots in this group is under 20%, better invest in a phone line.
- You sell luxury, at €5,000+ products the customer wants a human touch from second one.
WhatsApp Business integration, when it's worth it
WhatsApp Business API + AI chatbot is a combo that exploded in 2025. For the Slovak market it makes sense if: you have at least 30% returning customers, your products need post-purchase support (tracking, sizing, manuals), and your customers are on WhatsApp daily.
WhatsApp Business API cost: $0.005-0.025 per conversation (via Meta) + setup at our end €240. Monthly run at 500 conversations ≈ €15-25. Benefits: higher engagement (95% open rate vs email's 24%), direct mobile-pay integration (Slovak GoPay or Stripe Apple Pay).
Closing: three rules for an AI chatbot on a Slovak e-shop
- Never ship a bot live without 100+ internal training conversations, the bot must answer your top 30 questions perfectly.
- Always set up a human fallback via "give me a person" or frustration detection, without it you'll lose 8-15% of conversations.
- Review conversation logs monthly, the bot must learn from new questions it didn't know a month ago.
To start, we recommend a Hybrid Bot setup at €240 + AI Gateway. Fill out /brief, we'll come back within 24 hours with an integration plan for your specific e-shop.
Frequently asked questions
Q: Does the AI bot understand Slovak and Czech?
Yes, GPT-5 and Claude Sonnet 4.7 understand Slovak and Czech excellently, grammatically and semantically. Better than most English-only models from 2023. For professional terminology (medical, legal, technical) we recommend adding custom training on your own vocabulary, one-time €120-240.
Q: Can the AI bot access Shoptet / Shopify stock?
Yes. Shoptet has a REST API (requires an admin API key), Shopify has a GraphQL Admin API. Our integration calls these APIs in real time on every stock-related question. Latency 200-400ms (the customer sees a typing indicator during this). For Pohoda integration we use an XML feed with a 5-minute cache.
Q: How is this different from GPT-5 via my own OpenAI account?
Three differences. First, we route through Vercel AI Gateway, that's 50ms+ latency savings via edge inference. Second, we have prompt caching on by default, for repeated questions about your catalog, 70-90% cost savings. Third, we have fallback to another model (Claude, Llama) if the primary fails, your API never goes unanswered.
Q: Do I own the bot conversation data?
Always. Conversations log to your database (Postgres or Sanity depending on setup). We don't access logs without your consent. For GDPR compliance, we have automatic 30-day retention + right-to-erasure (the customer types "delete my data" and the bot authorizes the action).
Q: What if the AI bot gives a customer wrong information?
We implement three-tier protection. First, AI has a system prompt with "DO NOT INVENT product info, if unsure, say so and escalate." Second, AI always verifies stock data through your API (not from memory). Third, critical questions (price, availability, shipping) are always escalated for verification if the bot has less than 90% confidence.
About the author
Tair Khamitov, founder of DevNova in Bratislava. Since October 2025 he has delivered AI chatbots for Slovak e-shops and service businesses, including integrations with Shoptet, Shopify, Pohoda, Money S3. More at /services/chatboty or b2b@devnova.eu.