AI chatbots in customer service: what works and what backfires

What works

FAQ resolution. Bots that know your hours, prices, services, and policies inside-out. Customers get answers at 11pm. You don’t handle every “what time do you close?” query manually.

Lead qualification. Bots that ask the right questions before handing off to a human. “Are you renting or owning? Single or multi-property? Roughly what’s your budget?” By the time the bot transfers to your sales team, the lead is qualified.

Appointment booking. Bots that handle the “when can I come in?” conversation, check the calendar, book the slot, send the confirmation. End to end without a human touching it.

Status checks. Bots that look up your existing booking, your order status, your account balance. Faster than any phone call.

What backfires

Bots pretending to be human. Customers find out. Trust evaporates. Always be upfront: “Hi, I’m the AI assistant for [Business]. I can answer common questions or connect you to a human.”

Bots without human escalation. Every bot conversation needs a clear “talk to a human” option. Bots that trap customers in loops are universally hated.

Generic LLM with no business context. Bots that don’t actually know your business hallucinate confidently. The customer gets told the wrong price, then arrives expecting it. We’ve had to clean up after several of these.

Bots over-aggressive on lead capture. “Before I help, what’s your name, email, phone, business size?” Customers leave. Earn the contact info by being useful first.

What we build (L2Chat)

L2Chat is our AI chat product. It’s a website chatbot that knows your specific business: your services, prices, hours, policies, FAQs. Customers get accurate answers, you get qualified leads, the bot doesn’t make stuff up.

Key design choices: (1) bot is honest about being a bot, (2) human handoff is one click away, (3) bot only answers from your verified content (no LLM hallucinations), (4) lead capture happens after value, not before.

From $39/mo, 14-day free trial. Most clients see meaningful conversation volume within the first week.

When NOT to use a chatbot

You can’t describe your business in writing. If you can’t document your services, prices, and policies clearly, you can’t train a chatbot on them. Fix the documentation first.

You don’t have inbound enquiry volume. A chatbot solves “too many enquiries to handle.” If you have 5 enquiries a week, just answer them yourself.

Your buyers want phone first. Some industries (high-trust trades, legal services for older demographics) skew strongly to phone. A chatbot that’s ignored is wasted budget.

You can’t handle the leads it creates. A bot that books appointments doesn’t help if you can’t fulfil them.

How to evaluate one

Try it yourself before signing up. Ask 5 questions a real customer would ask. Try to break it (ask off-topic questions, ask weird edge cases). See how the human handoff works. Test on mobile.

If the bot fails any of those tests, it’ll fail with your customers.