A chatbot that knows your business — not a generic bot with your logo on it
Website and internal chatbots grounded in your real documentation, with honest escalation to a human when the answer is not in your material.
The problem this solves
The off-the-shelf chatbot you tried confidently told a customer something untrue about your return policy, and you turned it off. That is the correct reaction — a bot that invents answers costs more in trust than it saves in tickets. The problem was never the model. It was that nothing connected the bot to your actual policies, and nothing told it what to do when it did not know.
What you get
- A chatbot grounded in your real content — policies, product data, past inquiries, PDFs, help docs
- Citations on every answer, so a customer or staff member can verify the source in one click
- An explicit "I do not know, here is a human" path instead of a confident guess
- Handoff to your existing inbox or phone line, with the full conversation attached
- A conversation log you can review to see what customers actually ask — usually the most valuable output
What you can count on
- Answers cite their source document, so nothing is unverifiable
- Escalation is the designed default for anything outside the knowledge base
- A hard monthly spend cap is set during setup, so no traffic spike can produce a surprise bill
Then $250 a month to actually run it
The build is the easy part. What goes wrong afterwards is that nobody maintains it — six months later it is out of date, quietly broken, or nobody remembers how to change it.
The monthly fee is what stops that happening. It is not a support contract you hope never to use; it is the part where the thing keeps getting better.
- Model usage costs, capped and included
- Monthly review of real conversations, with fixes for anything that drifted
- Knowledge base kept current as your policies and products change
- A report on what customers actually asked — often worth more than the deflection
- No minimum term. Cancel any time and you keep everything that was built.
How it works
-
Source the truth
I collect the documents that define correct answers and identify the gaps. Ninety percent of bad chatbot answers trace back to a policy that was never written down anywhere.
-
Ground and constrain
The bot retrieves from your material and is constrained to answer only from it. Questions outside that boundary route to a human rather than to the model’s imagination.
-
Adversarial testing
I deliberately try to make it lie — edge cases, hostile phrasing, questions with no good answer. You see the transcript of every failure and how it was fixed before launch.
-
Launch and watch
It goes live on a slice of traffic first. I review real conversations weekly for the first month and tighten anything that drifted.
AI Chatbots — questions I get asked
How is this different from the chatbot my website builder offers?
Built-in bots match keywords against a short FAQ list, or pass your question to a general model with no access to your business. Neither knows what your actual return window is. These bots retrieve from your real documents before answering and cite what they used — and when your documents do not cover the question, they say so and fetch a human instead of guessing.
Will it make things up?
It is constrained to answer from your material and to escalate when the material does not cover the question. That is a design decision, not a setting — I test it adversarially before launch specifically by trying to provoke invented answers, and you see those transcripts. No system is perfect, which is why every answer carries a citation the reader can check.
Are the running costs on top of the retainer?
No. Model usage for typical small-business traffic is genuinely small — usually single-digit dollars a month — and it is included in the retainer along with a hard spend cap. You get one predictable number, not a bill that moves with your traffic.
Can it handle bookings, orders or account lookups?
Yes, but that is a different build. Answering questions is retrieval; taking an action means connecting to your booking system or order database, and every action needs its own guardrails and confirmation step. I usually launch answers first, watch what people actually ask for, and add actions where the volume justifies it.
Often paired with
Websites
Design, build and ongoing management of a small business website — fast, mobile-first, built to be found locally, and kept current every month.
DetailsLead Generation
Sourcing, enrichment, scoring and personalized first-touch outreach — automated end to end, with deliverability treated as a first-class concern.
DetailsMarketing Agents
Agents that research, draft, schedule and measure your content and campaigns — tuned to your voice, with approval gates before anything publishes.
DetailsReady to talk about ai chatbots?
Book a free 30-minute call. I will look at your actual situation and give you both numbers in writing — or tell you it is not worth doing, which happens more often than you would expect.