Loading...
Discussion
What actually determines whether a chatbot sticks — lessons from deploying voice AI in rural SD
I've deployed three AI voice and chat agents for clients and my own practice in Mitchell, SD, and the thing that kills most chatbot projects isn't the technology — it's the handoff design.
Most people ask "which platform should I use?" when they should be asking "what happens when it can't answer, and does my client trust it enough not to hang up?"
A few patterns I've seen hold up in real deployments:
Voice vs. chat matters more than people think. For agriculture, trades, and field service clients, voice agents outperform chat widgets by a wide margin — thumbs are busy, and typing a question to a bot on a tractor isn't happening. For professional service firms (CPAs, law offices), chat is preferred because it's perceived as more private and deliberate.
Knowledge base quality is the actual product. I rebuilt a client agent's knowledge base from scratch after launch because the initial answers were plausible but wrong in the details — the kind of wrong that takes 60 seconds to detect but erodes trust permanently. The platform barely mattered after that rebuild; the answers did.
The "I'll transfer you" moment defines the agent's reputation. Agents that confidently say "I don't have that, let me get you someone who does" outperform agents that hallucinate an answer every time. Escalation design is not a fallback — it's a feature.
Curious what others are seeing. For those building for SMBs specifically: are you finding clients want a branded voice/persona, or do they just want the function and don't care about the character of the agent?
Most people ask "which platform should I use?" when they should be asking "what happens when it can't answer, and does my client trust it enough not to hang up?"
A few patterns I've seen hold up in real deployments:
Voice vs. chat matters more than people think. For agriculture, trades, and field service clients, voice agents outperform chat widgets by a wide margin — thumbs are busy, and typing a question to a bot on a tractor isn't happening. For professional service firms (CPAs, law offices), chat is preferred because it's perceived as more private and deliberate.
Knowledge base quality is the actual product. I rebuilt a client agent's knowledge base from scratch after launch because the initial answers were plausible but wrong in the details — the kind of wrong that takes 60 seconds to detect but erodes trust permanently. The platform barely mattered after that rebuild; the answers did.
The "I'll transfer you" moment defines the agent's reputation. Agents that confidently say "I don't have that, let me get you someone who does" outperform agents that hallucinate an answer every time. Escalation design is not a fallback — it's a feature.
Curious what others are seeing. For those building for SMBs specifically: are you finding clients want a branded voice/persona, or do they just want the function and don't care about the character of the agent?
I have a guy that I taught who just finished training an agent on glass glazing! It’s freaking insane. Very niche industry, and he has that thing to a point where his employees can pretty much call in for any specific question and its customers facing!! 😆😅
Love this — niche vertical training is where the real magic happens.
We're running something similar out here in rural South Dakota. Built an orchestrated voice agent for farmers — you call in, and it pulls live weather, commodity market prices, and then a Claude-powered layer interprets those signals together. Not just "here's the rain forecast" — it's connecting the dots: planting windows, input costs, chemical pricing trends, crop insurance products, and which USDA/FSA programs apply to your situation right now.
The goal was pretty simple: a guy who's never set foot on a farm should be able to pick up the phone, get briefed in 90 seconds, and hold a real conversation with a producer. The knowledge base does the heavy lifting.
That's the thing about these vertical agents — they don't replace expertise, they distribute it. A glass glazing crew in the field, a farmer calling from the cab of a combine — same principle, different dirt.
We're running something similar out here in rural South Dakota. Built an orchestrated voice agent for farmers — you call in, and it pulls live weather, commodity market prices, and then a Claude-powered layer interprets those signals together. Not just "here's the rain forecast" — it's connecting the dots: planting windows, input costs, chemical pricing trends, crop insurance products, and which USDA/FSA programs apply to your situation right now.
The goal was pretty simple: a guy who's never set foot on a farm should be able to pick up the phone, get briefed in 90 seconds, and hold a real conversation with a producer. The knowledge base does the heavy lifting.
That's the thing about these vertical agents — they don't replace expertise, they distribute it. A glass glazing crew in the field, a farmer calling from the cab of a combine — same principle, different dirt.
That just made me think of something.... back in the day, I would actually call the bank every morning. They had this agent technically years ago that would answer and give you the temp, weather, and all that for the day going into the next night I would use that to make sure I was dressed right for school! Crazy now that I think about that a town called Mountain View Arkansas there was like 2000 people!