Using AI to Reply to Customer Enquiries: A Small Business Owner’s Experience

A customer asks about hiring ten bikes in the Lake District. Before you can quote, you need to know which bikes they want, how long they need them for and what delivery involves. Writing the reply is only part of the job.

Running Bike Hire UK and Waterside Bike Hire, I used to handle that process manually: read the email, pick out the requirements, open the relevant pricing documents, check the details and write everything back into a response. Some replies took up to 20 minutes.

I now use an AI email assistant to help with those enquiries. It is instructed to use designated business documents, ask questions when information is missing and prepare a draft for me to review. I still decide what gets sent.

The difficult part was moving between the email and the information needed to answer it. A reply might depend on bike types, hire duration, event details and delivery charges. Checking each detail meant going back and forth between documents before putting the answer into words.

For a small business owner, that work competes with serving customers and running the operation. An email assistant is useful when it reduces that repeated checking and drafting while leaving business decisions with the owner.

My setup has three main inputs: the customer's enquiry, designated Google Drive knowledge documents and a tone-of-voice guide.

The knowledge documents hold the relevant business information, including pricing and delivery details. The tone guide describes how I want replies written. My instruction is to use those approved sources for business facts and prices, rather than searching online or filling gaps with a guess.

When the information is sufficient, the assistant prepares a draft. When something is missing, it asks me a focused question or drafts a qualifying reply to the customer, depending on who needs to supply the answer.

This is an illustrative enquiry based on the sort of request I handle, rather than a published customer email:

“Could you provide ten bikes for a trip in the Lake District?”

That is enough to start a conversation, but it is not enough to calculate a reliable quote. The customer has not said which bike type they need or how long the hire will last.

The assistant's first job is to draft a reply asking for those missing details.

Once those details are available, the assistant can consult the designated documents for the applicable hire prices and the delivery charge for the area. If the documents do not cover the particular request, it needs to ask me rather than invent a charge.

The tone-of-voice document guides the wording so the response sounds closer to how I want the business to communicate. I then check the draft before sending it.

This sequence matters: clarify the request, consult the approved information, prepare the reply and get a human review. Generating a polished answer before the requirements are clear would simply make an unreliable quote look convincing.

Some enquiries previously took me up to 20 minutes to respond to because of all the document checking involved.

With the current setup, answering a question from the assistant generally takes me around 20–30 seconds when it needs my input. Once the email is drafted, my final skim and send usually takes under ten seconds.

Those are my estimates of my own active time. They exclude the assistant's processing time, any wait for a customer to answer and more complicated cases that need further checking. They are not a measured average or a promise of the same result for another business.

The practical benefit is that I spend less time assembling routine replies and more time providing the information and judgement that need to come from me.

One important lesson was being specific about where the assistant should get its information.

For my business, “find the price” is too open-ended. The instruction needs to identify the relevant knowledge documents and make clear that prices and business details should come from those documents only.

That also creates a responsibility for me: the documents need to stay current. If a price changes but the source document does not, the assistant may still produce an outdated answer. Restricting sources improves control; it does not make checking unnecessary.

The assistant prepares replies; I approve and send them. A tone guide and a set of source documents do not guarantee that every draft will be correct.

An unusual hire, an unclear delivery request or a promise about availability can need a closer look. My quick skim describes routine drafts, not how every enquiry should be reviewed. Anything involving an uncertain price, commitment or exception needs the appropriate check.

Choose one repeatable type of enquiry and gather the information needed to answer it. Make sure that information is clear and current before asking an assistant to use it.

Define which sources it can use, what it must ask when details are missing and who approves the reply. Add a short tone guide with examples of language you would actually use.

Try the process with fictional or anonymised enquiries first. Include incomplete requests to see whether the assistant asks sensible questions rather than guessing. Correct the instructions when it gets something wrong, and compare the time spent reviewing with the time you previously spent preparing replies.

For me, customer enquiries provide a practical example of how AI can help a small business: retrieving relevant information, identifying gaps and drafting a response while I stay responsible for the final email.

At Saunton AI, I help business owners identify where that kind of support could fit their own work. If customer enquiries or repeated admin are taking up your time, get in touch to discuss a sensible place to start. I am based in North Devon and work with small businesses across the UK.