AI receptionist

The 20 Requests Small Businesses Get Most by Phone

A catalogue of twenty calls small businesses get again and again, with advice on which an AI receptionist can handle and which need a person.

Call editorial team · Published · 5 min read

Open any small business's call log and a pattern appears. The calls are different people, but the questions repeat. Opening hours, directions, prices, booking, delivery, complaints. Once you know the pattern, you can decide which calls an AI receptionist should handle, which it should take messages for, and which should go to a person.

A small shop owner on the phone behind a counter with shelves in the background
Illustrative image

The twenty are grouped into five families: information, bookings, orders and delivery, problems, and sales enquiries. Seeing them in families makes it easier to decide a policy for each family instead of for each question, which keeps your instructions short and consistent.

The list below is a general catalogue, drawn from the kinds of calls that appear across shops, clinics, trades and services. Your own business will not match it exactly. Use it as a prompt to examine your own calls. For industry-specific examples, see AI phone reception by industry.

Information requests

These are the easiest to handle. The answer is a fact, it does not change often, and it can be written down. They are the best starting point for an AI receptionist.

Booking and scheduling requests

Requests to arrange, change or cancel something. The AI can take details, but a person or your booking system should confirm. See handling appointment requests with an AI receptionist.

Because availability changes and mistakes cost money, treat these as requests. State clearly what happens next and when.

Orders, delivery and service status

Callers ask where their order or repair is. The answer lives in another system, so an AI that cannot see it must hand over.

Unless your AI is connected to the right system, the best it can do is state the policy and take a message with the reference number. Be honest about the limit: "I can't see your order, but I can take your reference and ask the team to call you."

Problems, complaints and emotion

These calls matter most and tolerate mistakes least. Callers are often upset. An AI can acknowledge and take details, but the resolution needs a person.

How an AI should treat problem calls

  • AcknowledgeA short, human sentence that shows the call was heard.
  • CaptureName, number, a clear summary and the urgency.
  • RoutePass urgent matters to a person quickly or tell the caller how to reach help.
  • Promise carefullyState what will happen next, not what the outcome will be.

Do not let the AI argue, explain away or promise refunds. For difficult conversations, de-escalating difficult calls has a human approach worth copying in the AI's tone.

Sales and new enquiries

These are valuable and time-sensitive. Callers who are comparing options usually ring several businesses and go with whoever answers helpfully first. The AI's role is to capture the lead properly and promise a prompt reply. See lead routing for what happens next.

Other calls that are not customers

Beyond the twenty, every business also gets sales calls to itself, wrong numbers and calls from suppliers and partners. Not every call deserves a full conversation. Decide how the AI should handle cold sales calls politely and briefly. Telling them how to send information by email is often enough.

Sort your twenty into three groups

Once you have your list, sort each request by what the AI should do. This is the most useful part of the exercise.

GroupExamplesWhat the AI does
AnswerHours, directions, standard policiesGives a short answer from sources
Take a requestBooking, quote, order statusCollects details and explains next steps
Hand overComplaints, urgent problems, sensitive topicsAcknowledges, takes details and routes to a person

Build your own list

  1. CollectGather a month of call notes or call history.
  2. GroupPut similar calls together and count them.
  3. RankOrder the groups by frequency.
  4. AssignChoose answer, request or handover for each.
  5. WriteCreate sources only for the top ten.

Do not try to cover all twenty on day one. Ten covered well beats twenty covered badly. After a few weeks, read what callers actually asked and adjust. See writing the FAQ your AI receptionist answers from.

How the mix changes through the year

The twenty requests are not evenly spread. A garden centre gets stock questions in spring. A heating engineer gets urgent calls in the first cold week. A tax adviser gets filing questions before deadlines. A restaurant gets bookings on the days before holidays. Treat your list as a living document and revisit it each season. Add temporary answers for events such as sales, closures and deadlines, and remove them afterwards so callers do not meet stale information.

A good habit is to keep a short "this week" note for the agent: any change to hours, any known delay, any special offer you want mentioned. A small team can maintain this in a few minutes on a Monday and avoid most embarrassing out-of-date answers.

A hypothetical worked example

Imagine a hypothetical independent opticians with three staff. After a month of noting calls, the owner finds that opening hours and directions make up a large part of the calls, with appointment requests next, then questions about glasses repairs, then occasional complaints about a late order. The owner sorts them: hours and directions go to the answer group, appointments and repair enquiries go to the request group, and complaints go to the handover group.

The first version of the agent covers only those three groups. It answers the hours and location questions, takes a name, number and preferred day for appointments and repairs, and takes a message for anything else. After a fortnight the owner reads fifteen transcripts, notices a recurring question about contact lens deliveries, and adds a short answer. That is the whole cycle: observe, sort, write, test, review.

Mapping request types in Call

In Call, call history lets you filter and read real calls, and AI call history shows each conversation, so you can see which requests recur. Question-and-answer sources cover the informational calls, the agent's instructions explain how to take a request, and a person-facing route such as voicemail or a callback request handles the rest. CRM contacts hold the history for callers who return. Use these together to match your own top request types to the right treatment.

Frequently asked questions

Are these the right twenty for my business?

Probably not exactly. They are common patterns. Check your own call history and replace any that do not appear with ones that do.

Which should I automate first?

Start with factual information: hours, location and standard policies. They are low risk and quick to write.

What about calls in other languages?

Check how your system handles languages, test it with real speakers and decide how callers can reach a person if needed. Never assume it will cope with every language.

How many calls will the AI handle on its own?

It depends on your mix. A business with many factual calls will see more resolved by the AI; one with mostly problems and quotes will see more passed to people.

What if a caller asks something outside my list?

The agent should say it cannot confirm that and take a message. Review these calls and add common ones to your sources.

Should it handle complaints?

It should take the details and pass them on. Resolving a complaint needs judgement and authority that a phone assistant should not pretend to have.

Match your common call types to an AI agent in Call

In the demo console you can try incoming calls in a simulation.

Related guides