Start with the call that cannot wait. If a caller needs a person to make a decision during the conversation, choose an arrangement that can reach that person. If a clear answer or a message for an agreed callback completes the receptionist's part of the job, AI can be worth trialling.
That decision comes before the provider shortlist. A pleasant voice cannot approve a refund, change a booking or promise a response that nobody in your business has agreed to deliver.
Use the worksheet below to turn your recurring calls into a brief: what needs to happen, who owns it and what the caller hears when that person or system is unavailable.
Choose an answering arrangement
There are several ways to put a receptionist between an unanswered ring and your team. This guide compares answering arrangements, including remote human services. Employing a receptionist brings different responsibilities, so a salary is not a useful stand-in for the cost of every human option.
Human answering gives the caller a person from the start. That person can work from your brief, take messages and, on a suitable plan, attempt a transfer. They still need permission to make decisions for you. ReceptionHQ's MyReceptionist service, for example, describes supplied business information, transfers and a message when nobody answers the transfer.
AI answering with a callback can suit routine questions and enquiries that your team will follow up. Sayora sends a summary by text, email or both. Sayora has no live transfer or staffed fallback, so someone asking for an immediate conversation with a person needs a different arrangement. The virtual receptionist page sets out that fit.
AI with an in-call human route adds a way for someone to take over during the conversation. Moneypenny describes escalation to its receptionists or the customer's own team. Confirm whose staff receive the call, their hours and the full charges for the selected plan. The word hybrid does not answer those questions by itself.
Write a brief for each kind of call
Work from the calls you actually receive. Group them by the outcome they need: a new enquiry, an appointment change, a progress update or a complaint, for example. Keep personal caller details out of the worksheet; you need the recurring job, not a copy of someone's private conversation.
Copy this block for each call type:
Caller need: What are they asking for?
Completed outcome: What must happen for that request to be dealt with?
Information and authority: Which approved facts, system access and permissions are needed?
Timing: Can the caller wait for a callback? If so, what expectation has the business agreed it can meet?
Owner and cover: Who takes the next action, when are they available and who covers their absence?
Failure path: What should the receptionist say and do if the answer, system or person is unavailable?
Candidate arrangement: Human answering, AI with a callback, or AI with a verified in-call human route. Why does it fit?
Evidence still needed: Which capability or trial result must be checked before you divert real calls?
Write an actual role against the owner field. “Someone will pick it up” leaves the most important part of the process undecided. If no one can own a promised next step, change the promise or the arrangement.
For a sector-specific brief, the estate agency guide separates viewing requests from tenant emergencies. The dealership and garage guide separates sales enquiries from workshop bookings and recovery calls.
Worked example: a new quote enquiry
This is an illustrative example. The caller wants to know whether the business covers their area, then asks for a quote. The owner handles quotes on working days, and the business has approved a next-working-day callback promise. A named deputy covers the owner's absence.
Completed outcome: answer the coverage question from approved information, capture the request and give the agreed next step. The quote itself comes from the owner or deputy.
Information and authority: the receptionist needs the service-area list and the questions that make a quote request usable. There is no permission to invent a price or promise that the work has been accepted.
Failure path: if an area is missing from the supplied information, record it as a question for the owner. If neither owner nor deputy can meet the callback promise, revise that promise before using this call flow.
Candidate arrangement: AI with a callback is worth trialling because the caller's request can wait for the authorised person. A human message-taking service is also an option. Give both the same brief and compare the usable enquiry and completed bill.
Evidence still needed: check an ordinary enquiry and one outside the coverage information. Inspect whether the message preserves the question, contact details and next action without adding an unsupported promise.
Worked example: changing an appointment
In this illustrative example, a bookings administrator can amend existing appointments during office hours. The caller accepts a callback before the original appointment takes place.
Completed outcome: identify the correct booking and make an authorised change, or explain clearly that a change has only been requested. Until someone confirms the amendment, the original appointment remains in place.
Information and authority: the service needs access to the actual booking, any required identity checks and permission to make that change. A product that creates new appointments may not support rescheduling existing ones. Sayora's published calendar booking should not be treated as evidence of that separate capability.
Failure path: if the receptionist cannot identify or amend the booking, pass the request to the administrator and explain what remains unconfirmed. If the appointment happens before the administrator returns, a later callback does not meet this example's timing requirement.
Candidate arrangement: AI with a callback can be considered while the administrator has enough time to act. Immediate rescheduling needs a service with the verified integration and permissions, or a route to someone who has them. Being human does not remove the need for system access.
Evidence still needed: use a test booking. Check what the caller is told and what the booking system actually records, including when the requested new time is unavailable.
Worked example: a disputed charge
Here the caller needs a decision about a charge during the conversation. The duty manager is available for the illustrated call and has authority to make that decision.
Completed outcome: reach the duty manager and let that person discuss the charge. Taking a message alone does not meet the stated requirement.
Information and authority: the receptionist needs the agreed transfer route and enough context to explain the request. Permission to answer the phone is not permission to negotiate a charge.
Failure path: if the manager becomes unavailable, explain that clearly. A later callback requires the caller's agreement and changes the original requirement. Do not promise an immediate decision that nobody can deliver.
Candidate arrangement: a human service with a verified transfer route, or an AI service with an appropriate in-call human route. A callback-only AI service does not fit this call as described.
Evidence still needed: test both an answered transfer and an unanswered one using invented details. Check who receives the context, what the caller hears and who owns any follow-up.
Check what human fallback actually means
A supplier's receptionist, a member of your team and a specialist authorised to make a decision are different people. Find out which one the service can reach. Then check whether that person is available during the hours you need.
Ask what triggers the handover, whether the caller has to repeat their explanation and what happens if the transfer is unanswered. Ask the supplier to show the selected plan's actual flow. A claim about another product or tier does not settle it.
The solicitors' guide applies that distinction to new enquiries and urgent legal calls.
You may choose different arrangements for different hours or call types: answering your own phone in the day and diverting it after hours, for example. Write down how those arrangements connect and who maintains them. Do not assume a product provides routing between them simply because the overall plan makes sense on paper.
Price the arrangement you have chosen
Give each candidate the same completed brief, along with your ordinary and busy-month usage. Include the required hours, call duration, transfers, human involvement and booking work. Ask for allowances, overage, setup, tax and minimum term alongside the subscription.
The answering-service cost guide works through comparable usage calculations. Use those calculations after you have checked that the plan can do the required work. A cheaper service that leaves every appointment change for an unavailable administrator has not solved that part of the problem.
Take the brief into a trial
Choose candidates that can meet the written requirements. Run the same scenarios using invented caller details and a test calendar. The trial checklist gives you a place to record the setup, expected result and actual result.
Check the answer, the final message, any calendar change and the next person's response. Repeat a failed scenario after correcting the instructions or configuration. A few successful calls demonstrate those calls; they do not establish a general accuracy rate.
If you already pay for an answering service, you don't have to cancel it first. The AI answering service page covers what changes when you switch, and how to try Sayora alongside the service you have.
Keep an arrangement when it completes the required work at a bill you can accept, with a clear owner when it cannot help. If a call still ends with nobody responsible for the next step, return to the worksheet before forwarding real enquiries.