Picture a cold Tuesday night in December. At 10:40 p.m., a woman stands outside her own front door with a bag of groceries, a phone at nine percent, and no keys. She calls the small family locksmith whose van she has seen around the neighborhood. The phone rings and rings. Both locksmiths are out on other calls, one rekeying a rental between tenants and the other helping a driver locked out of a car at a gas station. Nobody answers. She calls the next number she finds.

The next morning, the owner listens to the voicemail and winces. It’s not the first one like it this year. It’s not even the first one this week.

This is an illustrative scene. The locksmith in this piece is an invented example business, not a client story, and nothing here describes a real result. But the problem is familiar to almost any small service business. The work happens in the field, the phone rings in the office, and the people who could answer it are holding a drill, a pick, or a steering wheel.

December is a good month to look back at that problem honestly. This post uses a simple year-end exercise called Start, Stop, Keep. What should the business start doing with AI in 2027? What should it stop doing by hand? And what must it keep in the hands of people, no matter how good the tools get?

Why reflect in December

Most small businesses don’t get a quiet week. But December often brings a slightly different rhythm: a few slower days between the rush of holiday lockouts and the January wave of new locks for new homes, new tenants, and new starts. That’s a good time to look at the year with clear eyes.

Start, Stop, Keep is easy to remember and hard to fake. You don’t need a consultant, a whiteboard, or a retreat. You need the past year’s calls, messages, and calendar, and an honest hour with the people who do the work.

The rules are simple:

• Start: one or two things you’ll begin doing next year.
• Stop: one or two habits that cost more than they give.
• Keep: the things that make your business trustworthy, which no tool should change.

For AI, the third list matters most. It’s tempting to write a long Start list full of new tools. The better question is what you’d never hand to software, and why.

Meet the shop (an invented example)

Our example is a two-van residential and automotive locksmith run by Marisol and her brother Dev. They’ve been in business long enough that some customers call them by name. They rekey homes for new owners, install deadbolts, cut keys, open cars, and handle the occasional commercial job for a few small offices in town.

Marisol runs the office side, which mostly means her phone. Dev is in the field most days. When both are on jobs, calls go to voicemail. Messages come in through the website contact form, a few through text, and some through social media. Marisol answers them at night after dinner, typing the same answers over and over: yes, we rekey; yes, we come to you; no, we don’t sell safes; please call for an estimate.

Looking back at the year, Marisol sees three patterns:

1. Missed calls during the busiest hours. When everyone is on a job, the phone goes unanswered at exactly the moment people need help.
2. Repeated questions. A large share of messages ask the same few things: hours, service area, whether they do cars, whether they can come today.
3. Notes that live in her head. Callbacks, follow-ups, and estimates are tracked on sticky notes and memory.

None of this is unusual. And none of it is a reason to hand the business over to software. It’s a reason to sort the work into what a tool can carry and what a person must hold.

Start: using AI for simple tasks

The calendar line for this month says it plainly: start using AI for simple tasks. The hard part is deciding what “simple” means.

Here’s a test Marisol can use. A task is simple enough for AI when:

• The answer is already written down. Hours, service area, the list of services, and what the shop doesn’t do.
• The answer is the same for everyone. It doesn’t depend on who’s asking or on a judgment call.
• A mistake is easy to fix. If the tool gets it slightly wrong, someone can correct it without harm.
• A person can see what happened. There’s a record, summary, or transcript to review later.

By that test, a lot of Marisol’s evening typing qualifies. Here’s what the shop might start with in 2027.

Answering common questions on the website and by text. An AI chat receptionist can answer from the shop’s own information, such as its website, a short list of services, and basic business details like address and hours. When it doesn’t have enough information, it can say so and collect the person’s contact details so Marisol or Dev can follow up.

Answering the phone when nobody can. An AI voice receptionist can pick up when both locksmiths are busy, answer common questions from the same information, and take a message when it can’t help. It can transfer calls to set numbers based on the caller’s need and the time of day, and after-hours behavior can be set to transfer or take a message. If a call is missed, a text can go back to the caller so they know someone saw it.

Booking routine appointments. Non-urgent jobs, such as rekeying after a move, can be booked into a connected calendar. Marisol decides which job types can be booked this way and which need a conversation first.

Summaries instead of sticky notes. Conversations can land in one shared inbox, with summaries and follow-up tasks, so callbacks don’t depend on memory.

Each of these is a simple task in the sense that matters. The information is already known, the answer doesn’t need judgment, and someone can see what happened.

What “simple” does not mean

Simple doesn’t mean unimportant. A lockout at 10:40 p.m. is a big deal to the person standing in the cold. The AI isn’t solving her problem. It’s making sure her call doesn’t disappear while the people who can solve it are busy.

Simple also doesn’t mean everything. The test above rules out a lot. An estimate depends on the lock, the door, the car, and the time of night. Whether to open a door depends on who’s asking and whether they have the right to be there. Whether a situation is an emergency depends on details a person should hear.

That’s why the Start list should be short. Marisol picks two things for January: the after-hours phone answer with a message and a text back, and website chat that answers the five most common questions. Everything else waits until those two are working and reviewed.

The U.S. Small Business Administration’s guide to AI for small business gives similar advice. It suggests starting small and testing tools to see whether they add value. It also notes that AI can help with customer service, such as a website chatbot that answers common questions or phone service that routes calls to the right place.

Stop: doing everything by hand

The second line of the calendar says stop doing everything by hand. That’s not the same as “stop doing things by hand.” Some things should always be done by hand. The goal is to stop spending human time on work that doesn’t need a human.

Looking back at the year, Marisol’s Stop list might include:

• Retyping the same five answers every night. If the hours, service area, and service list are written down once and kept current, she doesn’t have to type them forty times a month.
• Tracking callbacks on sticky notes. A shared inbox with follow-up tasks means Dev can see what’s waiting without calling his sister from a driveway.
• Letting every missed call go to a generic voicemail. A short answer and a text back tells people they were heard.
• Answering messages from five different places. Phone, text, website chat, and social messages can land in one shared inbox, with handoff to the team.

There’s also a quieter habit to stop: assuming the owner has to be the answer to every question. Marisol has been the switchboard for years. Some of that is care, and some of it is just habit. Letting a tool carry the routine answers gives her back time for the conversations that need her.

Keep: people in charge

The last line of the calendar is the one that matters most: keep people in charge.

For a locksmith, this isn’t abstract. A locksmith’s job is to let the right person in and keep the wrong person out. That judgment can’t be handed to software, and it shouldn’t be described as if it could.

Here’s what stays with Marisol and Dev in 2027, no matter what the tools can do.

Verifying who’s at the door. Consumer protection guidance on hiring a locksmith, such as this fact sheet from Wisconsin’s consumer protection agency, tells customers what to expect from a reputable locksmith: an estimate before work begins, clear information about extra fees, and caution about anyone who says a lock must be drilled right away. It also tells customers to expect the locksmith to ask for their identification, too. That check happens in person, by a person. The AI can tell callers what to have ready, such as ID, and that the locksmith will confirm it on arrival. It doesn’t decide who gets in.

Giving estimates. Prices depend on the job. The AI can explain that a locksmith will give an estimate before any work starts, and collect the details needed for that call. It doesn’t quote prices it doesn’t have in writing, and Marisol decides whether any prices go into its knowledge sources at all.

Deciding what’s urgent. A locked car with a child inside is different from a locked shed. Marisol writes the guidance the AI follows when someone describes a safety emergency, including telling them to call 911 first. She reviews those instructions herself, and any call that sounds urgent is transferred to a person.

Handling complaints and hard conversations. If someone is upset about a bill, a broken key, or a late arrival, a person calls them back. The AI’s job is to take a clear message, not to resolve it.

Saying thank you. The note to a longtime customer, the call to the property manager who sends rekey work every month, and the end-of-year card to the office building that trusts them with master keys all come from Marisol and Dev, in their own words.

The quiet risk: trusting the tool too much

There’s a reason “keep people in charge” is on the list, and it’s not just about what the AI can’t do. It’s about what people stop doing once the AI is running.

Researchers have studied this for years. A systematic review of automation bias by Goddard, Roudsari, and Wyatt (2012), focused on clinical decision support in health care, describes automation bias as the tendency to over-rely on automation. The authors found that factors like workload, task complexity, and time pressure made over-reliance more likely. They also identified things that helped reduce it, including training, emphasizing the user’s accountability, and design choices such as giving information rather than a recommendation.

A locksmith’s office isn’t a hospital, and the study doesn’t predict anything about small business tools. But the pattern is worth remembering. The busiest, most tired moments are exactly when people are most likely to accept whatever the tool says without checking.

For Marisol, that means a few simple habits:

• Read the summaries, not just the count. A number that says “twelve conversations handled” tells her nothing about whether they were handled well.
• Name an owner. Every message the AI takes has a person responsible for the follow-up, by name.
• Check the hard ones first. Complaints, urgent calls, and anything the AI couldn’t answer get reviewed before the routine ones.
• Don’t approve when exhausted. If she’s reviewing AI-drafted replies at 11 p.m., she can let them wait until morning.

The SBA guide makes a similar point in plain terms: have another person review what AI produces, and make sure a person assesses the messages it generates.

The information is the job

An AI receptionist is only as good as what it’s given. For the shop, that means the most important work of the whole project isn’t turning anything on. It’s writing down, clearly and accurately, what the business does and doesn’t do.

Marisol’s starting list might include:

• Hours, including after-hours. When the shop answers live, when it takes messages, and what happens on holidays.
• Service area. The towns and neighborhoods they serve, and what happens if someone is just outside it.
• Services. Rekeying, deadbolt installation, key cutting, car lockouts, small commercial work.
• What they don’t do. No safes, no work on a door unless the person can show they have the right to it.
• How estimates work. A locksmith gives an estimate before work begins, by phone or on arrival.
• What to have ready. ID and, for cars, proof of ownership.
• What the AI should say when it doesn’t know. Something honest and short, followed by a promise that a person will follow up.

The platform’s guidance is to keep these sources current. When the service area changes or the holiday hours are set, the knowledge sources change too. It’s a small habit that prevents a lot of wrong answers.

When the AI doesn’t know

One of the most important lines in the whole setup is the one the AI says when it doesn’t have an answer.

Marisol wants it to sound like her: plain, honest, and quick. Something like this illustrative wording: “I don’t have that information, but I can take your name and number and have Marisol or Dev call you back.” No guessing, no made-up prices, and no promises about arrival times the shop can’t keep.

It’s tempting to make the AI sound more capable than it is. That’s a mistake, especially for a business built on trust. A customer who hears “I don’t know, but a person will call you” is more likely to trust the shop than one who gets a confident answer that turns out to be wrong.

A weekly look

Turning things on is the beginning, not the end. The platform’s guidance suggests reviewing live conversations regularly, with weekly as a good starting point, and changing one thing at a time. Conversation logs can show which questions the AI couldn’t answer well.

For the shop, a weekly look might mean:

• Reading a handful of conversations from start to finish.
• Checking any calls that were transferred or flagged.
• Noting questions the AI couldn’t answer, and adding answers to the knowledge sources if they’re the kind of thing it should know.
• Removing anything that’s out of date.
• Changing one instruction at a time, then watching what happens.

Configurations can also be tested with simulated conversations before they go live. Marisol can pretend to be a customer locked out at night, a landlord asking about rekeying, or someone asking for a price, and see what the AI says before anyone real hears it.

Who owns what

A simple way to keep people in charge is to write down who owns each part of the work. For the shop, it might look like this:

• Marisol owns the information. Hours, service area, services, and the “I don’t know” wording.
• Marisol owns the review. The weekly look at conversations and any changes to instructions.
• Whoever is on call owns transferred calls. If the AI hands off a call, that person answers it.
• Dev owns estimates and on-site checks. Prices and verification happen with a person, at the job.
• Both own the thank-yous. No AI writes the year-end notes to longtime customers.

Writing it down makes it real. It also makes it easier to notice when something falls through the cracks.

Being honest about the tool

Should customers know they’re talking to AI? The SBA guide notes that while there are currently no federal or state laws requiring businesses to disclose AI use, disclosure is becoming an expected best practice, and it suggests considering a public statement about how the business uses AI. That’s general information, not legal advice, and rules can change.

The same fact sheet tells consumers that if a company answers the phone with a generic phrase rather than a business name, they should ask for the legal name of the business, and call someone else if they don’t get it. That’s a useful reminder for any AI greeting. It should use the shop’s real name, every time.

For a locksmith, being upfront fits the business. Marisol might have the AI introduce itself as the shop’s virtual assistant and explain that it can answer questions and take messages, while a locksmith handles the job itself. It’s honest, and it sets the right expectation.

The same exercise, different work: a senior center

To see how Start, Stop, Keep changes with the work, picture a different invented example: a neighborhood senior center run by a small staff and a group of volunteers.

The center’s Start list might be similar on the surface. An AI chat receptionist could answer questions about hours, the activity calendar, and how to sign up for a class. A voice receptionist could answer after hours and take messages.

But its Keep list would be different. Many callers are older adults, and some calls are about more than a schedule. Someone calling to ask about lunch might really be lonely. A family member might be calling with a worry about a parent. The center would keep those conversations with people and make sure calls that need a person get transferred or returned quickly. It would also be careful not to collect more personal information than it needs.

Its Stop list might include retyping the monthly activity schedule into five places, or leaving voicemails unreturned during busy program days.

The exercise is the same. The lists are different because the people and the work are different.

Your Start, Stop, Keep for 2027

If you’d like to do this for your own business, here’s a short version you can use in an hour:

1. Gather the year. Pull up your missed calls, your messages, and your calendar for a few busy weeks.
2. List the repeated questions. Write down the questions people ask most often and the answers you give.
3. Mark what’s simple. Use the test: written down, same for everyone, easy to fix, and visible afterward.
4. Pick two things to start. Not ten. Two.
5. Name one or two things to stop. Habits that cost time without adding care.
6. Write your Keep list. The judgment calls, conversations, and thank-yous that stay with people.
7. Name owners. Who owns the information, the review, and each kind of follow-up.
8. Set a weekly look. Put it on the calendar now.

Holding the keys

The woman on the porch in December didn’t need software. She needed a locksmith. What she didn’t get was a sign that anyone heard her call.

That’s the honest promise of AI for a small business like this one. It won’t open the door. It can make sure the call is heard, the message is clear, and the right person knows to call back. The judgment, the care, and the keys stay with people.

If you’d like help sorting your own year into Start, Stop, and Keep, start a conversation with WeKinnect. Bring a list of the questions you answer most often, and we’ll look at which ones a tool can carry and which ones should stay with you.

Photo by Filip Szalbot on Unsplash