How I built an AI phone receptionist for Czech tradespeople - first paying customers
I noticed a simple but painful problem: small Czech businesses - plumbers,
roofers, electricians - miss dozens of calls every week. They're on the
roof, under the sink, or driving between jobs. The phone rings. Nobody answers.
The customer calls a competitor. Most customers don't call twice.
So I built KraftunAI - an AI voice receptionist that:
- Answers every call 24/7 (including nights and weekends)
- Books appointments directly into Google Calendar
- Sends a summary after every call
The whole thing runs on a local Czech +420 number so it feels natural to callers.
Setup takes 1-2 weeks. No software to install on the client side.
The hardest part wasn't the tech - it was convincing traditional Czech
tradespeople to trust an AI on their phone line. The demo call helped a lot.
We let them call the number and experience it themselves before signing up.
Currently at early customers, iterating on the product based on real feedback.
Happy to answer questions about voice AI, the Czech market, or the sales process
for "offline" businesses.
Congrats on the first paying customers, Dmytro!
Solving a leaky bucket problem for offline tradespeople is a brilliant niche they are notoriously hard to reach but have high intent.
I love the "experience the demo call first" sales strategy. Traditional businesses need to touch and feel the product before trusting it.
As an automation and backend developer, I’m curious about your client acquisition workflow. Since tradespeople are mostly offline, are you manually cold-calling them, or have you automated the pipeline by scraping targeted local business data (like Google Maps/local Czech directories) to find high-intent prospects who have lower ratings due to "unanswered calls"?
This has massive scalability potential. Upvoted, and wishing you more growth! Drop me a message if you ever want to brainstorm automating the outbound data pipelines for this.
Thanks a lot for the kind words and the upvote! 🙏
Right now my main acquisition channel is good old cold calling - it works surprisingly well for tradespeople since they actually pick up the phone (when they're not on a roof 😄). Scraping structured data for outbound is something I've thought about, but I'm still at the stage where manual calls give me the most direct feedback.
On the inbound side I'm actively investing in SEO for KraftunAI - trying to capture people who are already searching for solutions.
Manual cold calling is definitely unbeatable for that raw feedback loop, especially with tradespeople who value that direct human touch.
Since you're already investing in SEO and thinking about structured data, have you considered a hybrid approach? Instead of just 'cold' scraping, you could build a trigger-based pipeline for example, scraping business registries or Google Maps for newly registered tradespeople in the Czech Republic and hitting them up right when they are setting up their operations.
That way, your cold calls are 'warm' because you know they are in the setup phase and likely missing calls.
I’ve built similar high-speed Python pipelines for lead hunting in the Real Estate/E-commerce niches if you ever want to brainstorm how to automate that data gathering part so you can focus strictly on the calling/closing, I’d be happy to share some logic.
Good luck with the SEO climb, that’s a long game but worth it!
Smart positioning move with the +420 local number — exact kind of detail that decides trust in offline B2B. International voice AI competitors won't bother with country-specific infrastructure, which gives you a 2-3 year window.
The harder question is what the moat looks like after that. "AI receptionist for tradespeople" is rapidly saturating globally (Sameday, Goodcall, Bland, PolyAI) and Czech localization is replicable.
The pattern we see at Hivemind across geography-first wedges: second-order moat usually has to be vertical depth, not language. Plumber call patterns aren't roofer call patterns aren't electrician dispatch logic. Most generic AI receptionists fail tradespeople because they're built generic.
Going deep on one trade (workflow, objections, booking logic, after-hours rules) is probably the next defensible layer. Curious which trade is pulling hardest right now — that's usually where the second wedge lives.
Totally agree - and this is something I've been thinking about too.
Localization buys time, but it's not a moat. The real depth has to come from vertical specificity. I already have two clear candidates I'm evaluating for that next layer, and the data from early calls is helping me prioritize.
Appreciate the framing - useful signal on where to focus.
Glad it landed. Quick framework for picking between the two candidates from your call data:
Call volume isn't the most important signal. Look at average call duration, missed-call recovery rate, and willingness-to-pay signals (do they ask price first, or describe problem first?). Longest average call duration usually has the deepest workflow to specialize for. Highest missed-call cost usually has the highest WTP.
If you want a second pair of eyes on the sequencing call: myosin.xyz/hivemind. Either way, rooting for KraftunAI.
What model did you use for the voice of the assistant?
We're using ElevenLabs for the voice.