AI chatbots for customer enquiries: useful or a wall

Aug 27
3 min read

Updated: 4 days ago

Introduction

Every small business is currently being sold a chatbot. The pitch is that it handles enquiries around the clock so you do not have to.

Sometimes that is true and genuinely valuable. Just as often the result is a customer trying to reach a human, failing, and going to a competitor. The difference is not the technology but a handful of design decisions.

1. AI chatbots for customer enquiries succeed on the repetitive questions

Be honest about what the tool is actually good at.

Opening hours, location, parking, prices, whether you cover an area, what a service includes, delivery times, whether something is in stock. These are asked constantly, are tedious to answer, and are answered perfectly well by software.

2. Always leave an obvious route to a person

This single decision determines whether customers experience the bot as help or obstruction.

A visible option to reach a human, offered early and not buried after five failed attempts. A customer with an unusual or urgent problem needs a person, and preventing that is how businesses lose exactly the enquiries that matter most.

3. Feed it your actual information, not generic content

A chatbot is only as good as what it has been given.

Your real prices, policies, service areas, hours and processes. A bot answering from generic material will invent plausible-sounding details, and a confidently wrong answer about your returns policy is considerably worse than no answer.

4. Decide what it must never do

Set boundaries before it goes live, because the failures are predictable.

Quoting prices on bespoke work, making promises about timescales, giving advice with safety, legal, medical or financial consequences, or handling complaints. In each case the correct behaviour is to take details and pass it to a human.

5. Capture the enquiry even when it cannot answer

Every conversation should end with something you can act on.

A name and contact detail, and a summary of what they wanted, passed to you. A bot that reaches its limit and simply apologises has lost a lead that would otherwise have arrived as an email.

6. Set expectations about what it is

Pretending the bot is a person backfires when it inevitably fails.

Saying plainly that it is an automated assistant, and that a person is available, makes customers more tolerant of its limitations. People forgive a machine for not understanding and resent being deceived about which they were talking to.

7. Read the transcripts, because they are valuable

The conversation log is a free record of what your customers actually want to know.

Recurring questions belong on your website. Questions the bot handles badly need better information. Confusion about a policy suggests the policy is unclear. Most businesses install a bot and never read a single transcript.

8. Consider whether it is the right problem to solve

Not every business needs one, and some need something else.

If your enquiries are mostly urgent, complex, emotional or high-value, a person answering the phone will serve you better. The tool suits high volumes of simple, repetitive questions, and installing one because it is available is not a reason.

9. Measure resolution and escalation, not conversation count

The vendor's dashboard will report activity, which is not the same as value.

Track the proportion of conversations resolved without a human, how many escalate, how many produce a contactable lead, and any complaints about the bot itself. A high conversation count with low resolution means it is intercepting enquiries rather than handling them.

Conclusion

Use a chatbot for the repetitive questions and design it so a customer can always escape it, because the failure mode is becoming a wall rather than a help.

Feed it your real prices, policies and processes, define what it must never attempt, capture contact details even when it cannot answer, be honest that it is automated, read the transcripts as free customer research, consider whether your enquiry mix actually suits one, and measure resolution and escalation rather than conversation volume.