The owner of a family heating and cooling company with 14 technicians and four people answering phones spent a Sunday afternoon listening to 20 random recorded calls. On seven of them, a customer asked roughly what a job would cost, and the person answering said someone would call back with a price. In most of those cases, nobody did. Replacement systems and major repairs at his company run anywhere from a few hundred dollars to over $8,000. He did the math and stopped listening for the day.
Most small business owners have never heard their own phone calls. The ones who do usually find money leaking out in places they didn’t expect.
Start with the back office
The most practical uses of AI for business operations at a company this size are unglamorous. Reading supplier invoices as they arrive and matching them against purchase orders and job tickets, flagging the ones where a part was billed at a higher price than quoted. Turning a technician’s rushed notes from the field into a clean estimate. Summarizing each customer call into the job record so nobody has to ask the customer to repeat themselves. Sending polite reminder emails on invoices 15, 30, and 45 days overdue.
Before, the office manager at the heating company spent most of every Friday afternoon reconciling supplier invoices by hand. Now she reviews only the dozen or so the system flags.
That “review the exceptions” setup is the right way to start. Pick tasks that are repetitive, follow clear rules, and are easy to check. Let the software do the first pass and a person approve anything unusual. Hold off on letting AI send anything involving money to customers, such as estimates or collection notices, until you’ve watched it work for a few weeks. It will occasionally misread a part number or confuse two customers with the same last name, and the cost of that mistake lands on your reputation.
Most of these tools run $50 to $300 a month. Measure against the hours they actually return, not the hours promised in the demo.
What the recordings showed
Listening to random calls is a poor use of an owner’s time. Twenty calls out of 900 a month tells you something, but not much.
Transcription and call analysis tools change the sampling. Every call gets transcribed, and the system flags the ones worth hearing: a customer asked about price and no appointment was booked, someone mentioned a competitor, a caller sounded angry, a promised callback wasn’t logged. The owner now spends 30 minutes a week on flagged calls instead of an afternoon on random ones.
Coaching people, not scorecards
Finding the problem calls is half the job. Changing what happens on them is the other half.
Many methods for call center coaching were designed for operations with hundreds of agents, and they translate badly to a team of four. Detailed scorecards with 25 criteria, public leaderboards, and automated quality scores tied to performance reviews all tend to backfire in a small office. People learn to hit the checklist instead of booking the job, and a leaderboard with four names on it just embarrasses whoever is last.
What works better is simpler. A 20-minute weekly session with each person, listening to two of their calls together. Let them pick one they felt good about, and you pick one from the flagged list. Focus on a single behavior at a time. For the heating company, it was this: never end a pricing call without offering a specific appointment time, even if the customer isn’t ready to commit.
Role-play the hard calls that come up constantly: the price shopper calling five companies, the customer upset because a technician arrived late. Rotating who plays the customer keeps it from feeling like a test.
Use the software to find calls. Leave the coaching to people.
The numbers worth watching
Track the percentage of inbound calls that end in a booked appointment and the percentage of promised callbacks completed within two hours. Be wary of average call length. A longer call that books an $8,000 replacement beats a quick one that ends with “we’ll get back to you.”
Three months later, the heating company’s booking rate on pricing calls had climbed noticeably, and the owner credits the coaching more than the software. The tools didn’t teach anyone to ask for the appointment. What they did was point him toward the calls he needed to hear, and get his office manager her Friday afternoons back.
