You spend money on a website. You work on Google rankings. You ask happy customers for reviews. Maybe you run ads, sponsor a local team, or put your number on every truck.
Then someone calls.
They hear a menu. Or a voicemail greeting. Or the phone keeps ringing because your team is already helping someone else.
What happens next deserves the same attention as the work that brought them to you.
I think of Cheers and the feeling of walking into a familiar place where someone pays attention. That’s a useful ambition for a small business.
On the phone, it starts with a welcome. But a friendly voice alone won’t answer a warranty question or sort out a booking. The caller still needs someone to understand the request, give a correct answer and explain what happens next.
For a small business owner, this is worth checking before spending more to attract calls. Your phone system may already hold clues about where people give up and what your team needs to help them.
What I saw in two businesses
I recently looked at phone data for two clients. The roughly 30–40% dropped-call figure around the automated phone menu, or IVR, refers to all inbound calls captured at the IVR.
These are observations from two businesses, not an industry benchmark. They do not establish why callers hung up or whether a sale was lost.
My read was that many callers wanted to speak with someone right away. They had picked up the phone for a reason, and leaving a message or working through a menu wasn't what they wanted.
That's my interpretation. A call log can show where a call ended. Matching later calls may show whether someone called back. Neither tells us, on its own, why they hung up or whether they would have bought anything.
In one client conversation, we talked about how much effort it takes to get those calls in the first place. Watching people drop out after all that work is a reason to look closer.
Voicemail asks the customer to do more work
When someone reaches voicemail, you're asking them to explain their problem, leave their details, and wait. They also have to be available when you call back.
Sometimes that works. Sometimes they move on.
In a 2025 CallRail survey of 1,000 US consumers, 42% said they leave voicemail when a business doesn't answer. Another 21% said they immediately call another business.
It's a vendor survey about what people say they do. It isn't a measured missed-call rate, and it doesn't tell us how your customers behave. But it gives owners a reason to question the assumption that callers will always leave a message.
For a plumber out on a job or a shop with a busy counter, answering every call personally may be hard. The next step is to find out which calls need a person, which need a simple answer, and which need a clear callback plan.
Start with the call journey
Start by finding out what your phone system can tell you. Your VoIP provider can confirm what data your plan makes available. Call logs, recordings, and transcripts depend on the service, plan, and settings.
Not sure where to start? You don’t have to figure it out alone. Salient AI can help you review the data, spot where callers get stuck and decide what to improve first.
Then pick a defined period and follow genuine inbound calls from start to finish. Remove obvious spam. Count each call once, even if it rings several employees before someone answers.
Be careful with the labels. In Amazon Connect's metric definitions, an agent non-response can be counted more than once for the same contact, and that contact may later be handled by another agent. Its abandoned-contact metric refers to customers disconnecting while queued before connecting to an agent. Ask your own provider what each label counts, including whether an automated answer is distinguished from a conversation with a person.
For each call, ask:
- Did it arrive during open hours or after closing?
- Did the caller reach a person, a menu, a queue, or voicemail?
- Where did the call end?
- Was a message left or a callback requested?
- Did someone follow up, and was the request resolved?
Keep the stages separate so one call doesn't become several failures in your report. A caller who hangs up and gets a useful callback is a different outcome from someone nobody contacts.
You don't need a fancy dashboard to start. You need a clear view of where the conversation stops.
Decide whether the menu is earning its place
An IVR can help route calls. It can also make a small business feel harder to reach than it needs to be.
There is older research worth knowing about. In a 2019 Vonage survey of 2,010 US respondents, people estimated abandoning an average of 27% of their calls to businesses because they reached an IVR. This was a vendor-sponsored survey based on people's memories, not measured call logs. It is a reason to investigate your own menu, not a current benchmark to apply to your business.
Call your own number as a customer. Listen to every option. Try asking for help with something that doesn't fit neatly into a department.
How long does it take to reach someone? Is there an easy way to ask for a person? Does the menu send people to a team that's actually available?
A shorter greeting, fewer choices, or better routing might fix the problem. AI is one option to test after you understand where callers are getting stuck.
Try AI where voicemail is falling short
An AI receptionist may be useful when nobody can answer. It can ask what the caller needs, collect contact details, and handle approved questions. Depending on the setup, it may route a call or help arrange a booking.
I'd start with a narrow job, such as after-hours enquiries or overflow calls. Give it clear rules about what it can answer, what needs staff review, and what to do when a person isn't available.
Before relying on it for live callers, test it like a customer:
- Ask for a human immediately.
- Call with background noise or an accent it may find difficult.
- Correct a name or phone number.
- Ask something outside its approved information.
- Try a transfer when nobody is available.
Check the actual setup rather than assuming every product works the same way. For example, CallRail documents how Voice Assist fits into its call flows.
Keep a human route available. In Gartner's 2026 customer survey, 87% said access to a human was essential when companies use generative AI for service. That survey covered customer service broadly, not just phone calls.
Measure whether callers get a useful answer, a completed booking, or a callback that happens. A higher answer rate by itself doesn't tell you whether service improved.
Listen to the calls your team does answer
Missed calls are only part of the picture. The calls you answer can reveal gaps in training and information.
Where recordings are allowed, review a sample. AI can help group calls into topics such as new enquiries, booking changes, product questions, warranty questions, and complaints. A person should check those categories before using them to make decisions.
Look for things you can actually hear:
- Was the greeting clear and welcoming?
- Did the employee understand the request?
- Was the answer accurate and based on current information?
- Did the caller know what would happen next?
- Did the conversation end with a clear, polite goodbye?
Sentiment tools can flag language for a closer look. I would treat those flags as prompts for human review, not proof of someone's emotions or a stand-alone score for an employee.
Repeated questions are especially useful. If staff keep putting callers on hold to find the same answer, you've found something to improve in your training or knowledge base.
Help the person answering the phone
Do your employees know the warranty period for every product you sell? What about someone who just started?
A new employee may have a great manner and still struggle with product details. Even experienced staff can't keep every update in their heads.
AI can help them find approved information while they work. Tools such as Google Agent Assist offer knowledge suggestions during customer conversations.
For a warranty question, the useful job is to retrieve the current policy, point out exclusions, and show the source. The employee checks that it applies and escalates anything unclear. The system shouldn't invent a promise to make the call go smoothly.
There's research behind this kind of support. A study published in the Quarterly Journal of Economics in 2025 examined a rollout covering 5,172 support agents and estimated that access to AI assistance increased resolved chats per hour by 15% on average. Gains were about 30% for less skilled and less experienced workers. The authors’ revised manuscript is available openly.
That was text chat at one business software company. It isn't a promise about small-business phone results. Experienced staff saw limited gains, with some signs of lower quality. It does suggest that helping newer staff find good answers is worth testing carefully.
Handle recordings with care
For Canadian businesses subject to PIPEDA, the Office of the Privacy Commissioner's call-recording guidance says to tell callers they are being recorded, explain why and obtain meaningful consent. Use recordings only for the stated purposes. If recording is not integral to the service, consider an alternative for someone who objects. Check which privacy rules apply to your business before recording or sharing calls with an AI provider.
Tell employees how recordings and AI reviews will be used, limit access, set retention periods, and review providers before sharing data. Ontario employers should also check whether electronic monitoring policy requirements apply.
Make the next call easier
Start with one week of call data and a small sample of permitted recordings. Find one problem, make one change, and check what happens next.
That might mean simplifying your menu, improving callbacks, testing an AI receptionist, or giving staff a better place to find answers.
That’s the part of the Cheers idea I’d bring to the phone: welcome people, listen properly and be useful. They don’t need the system to guess who they are. They need help with why they called.
You've worked hard to get someone to call. Make it easy for them to get the help they came for.
AI assistance disclosure: This article was prepared with AI assistance, using Stephen Lau's observations and the linked sources, and reviewed and approved by Stephen Lau. Illustrations are conceptual AI-generated scenes.