A chatbot that makes up an answer is worse than no chatbot. I build support and sales bots that answer from your real docs, products and order status, know when to stop, and pass the chat to your team with the full context.
A bot wired straight to an AI model will answer confidently about a policy it has never seen. Customers believe it, and you inherit the promise. The fix is retrieval and a confidence threshold, not a longer prompt.
Yes! Any item, any time, no questions asked.
Made upUnopened items within 30 days. For opened ones, I'll check with the team.
From returns-policy.pdfThe bot can't answer and there's no route to a person, so the customer repeats themselves three times and leaves angry. Handover decides whether people trust the thing at all.
Sorry, I didn't get that. Can you rephrase?
LoopI'm passing this to a person now. They'll see this chat and your payment.
Handed overBuilt once, not three times. And wired into the tools your team already uses, because a bot that can't read order status can't answer the question people actually ask.
Reads live Shopify and 3PL data to answer "where is my order?" without a ticket ever being opened.
Drafts each reply for a person to approve. One client's first response went from 4 hours to 12 minutes.
Asks two or three questions, then recommends from your real catalogue, never an invented product.
Scores the enquiry, books the call and writes the CRM record before anyone opens their laptop.
Answers your team's questions from your own handbook and process docs, with the source linked.
Your person gets the transcript and the customer record together, not a cold "user needs help".
Tendabot is the AI support agent I built and run: one bot for your website, WhatsApp and email, with every chat in one inbox and a person able to take over any of them.
What people actually ask, and whether the content exists to answer it.
Which model, where the handover line sits, one fixed price.
Retrieval over your content. Escalation paths built before the happy path.
A person approves replies while we tune on real chats.
30 days of free fixes and a weekly look at what it got wrong.
Retrieval plus a threshold. The bot answers from your own documents, not from the model's memory, and every answer has to clear a confidence bar before it's shown. Below that bar it says it doesn't know and hands over.
The build is one fixed price, quoted after a free 30-minute call. Running costs are model usage, usually $30 to $200 a month depending on volume, and I size that against your real traffic before you commit.
The cheapest one that clears your quality bar. Claude for a nuanced support tone, a smaller model for high-volume sorting, open-source and self-hosted if data can't leave your servers. I have no reason to prefer one vendor.
For most builds, yes: that's how the APIs work. Anthropic and OpenAI both offer terms that keep your data out of training, and I set those up. For legal, medical or financial use, I can self-host an open model so nothing leaves your servers.
Yes, and I wouldn't build one that couldn't. Handover goes to Slack, your helpdesk or email with the full transcript and customer record attached, so nobody starts from zero.
A free 30-minute call. I'll tell you whether your content is ready to ground a bot, and what it would take if not.
Book a chatbot call
Hammad Shahzad