AI Receptionist Setup: Your First Week, Day by Day (2026)

AI Receptionist Setup: Your First Week, Day by Day (2026)

Setup is not the hard part. Signing up for KeyBot Lite and getting a working AI receptionist answering your line takes about ten minutes, and the honest truth is that the ten minutes are boring: your business details, your service list, your hours, your escalation number, and a call-forwarding rule. Anyone can do it between jobs.

The part that separates shops who love this from shops who quietly stop looking at the messages is what happens in the seven days after those ten minutes. As of August 2026, every AI receptionist ships with sensible defaults, and defaults are exactly that — sensible, and not yours. Your area has a neighborhood nobody spells correctly. Your price sheet has a line that means something different than the words suggest. You have one type of call that must reach you inside five minutes and another type that can absolutely wait until morning, and no default knows which is which.

This is the operator's version of that first week. It is written for a locksmith shop, because that is the trade where the stakes are sharpest — a lockout caller who gets a vague answer at 11 PM is gone in ninety seconds — but the structure applies to any service business. Each day has a specific job, and none of them take more than twenty minutes.

Before day one: the prep list

Everything below is information you already have. The only work is putting it in one place before you sit down to configure. Do this on paper or in a notes app first; the setup form goes fast when you are not inventing answers in the field.

1. Your service list, in customer language. Not your internal SKUs. What a caller would actually say: car lockout, house lockout, key made from scratch, key fob programming, ignition repair, rekey, lock change, safe work, commercial door hardware. Include the things you do not do, because a receptionist that confidently declines the wrong work is worth as much as one that books the right work.

2. Your price sheet. This is the single highest-leverage prep item and the one most owners skip on day one and regret by day three. During setup you upload your price sheet, the system parses it, and you confirm every parsed price on screen before anything goes live. After that the receptionist states those exact prices when asked — word for word, never negotiated, never discounted, never invented. Anything not on the confirmed sheet becomes a message for you rather than a guess for the caller. Get the sheet right up front and you skip the most common week-one correction. How to structure a price book so it survives being read aloud on a phone call is its own topic, and we covered it in the locksmith price book guide.

3. Your service area. Cities, ZIP codes, and the honest edges. Where do you go for a lockout at 2 AM versus where do you go for a scheduled rekey? If there is a distance past which the answer is "we can refer you," write that boundary down now. Out-of-area calls handled badly are the most annoying kind of wasted message.

4. Your hours, including the real ones. Posted hours, emergency hours, and whether after-hours pricing differs. If you charge an after-hours premium, decide now whether the receptionist states it — it should, and it should be in the first number the caller hears rather than sprung later. We argued that case in after-hours emergency pricing.

5. Your escalation number and the rule attached to it. Which phone gets the urgent alert, and what counts as urgent. Do not overthink the rule on day one — "car lockout" and "person locked out of a house at night" is enough to start, and you will refine it on day four.

6. Who actually watches the alerts. This is an ops decision, not a software setting, and it is the one that determines whether the whole thing works. Messages arrive in Telegram within seconds of every call, with the recording attached. Somebody has to be looking. Name that person before go-live — you, a dispatcher, whoever is on call. An alert nobody reads is a voicemail with extra steps.

7. A dry run. Before touching your real line, go to the instant demo: enter your business info and the AI calls your phone in about 30 seconds answering as your own receptionist. No signup, first demo free. Do this with a coffee and a critical ear. You can also dial the demo line directly at +1 (716) 350-6391.

Day 1: go live on forwarding

Setup, then forwarding, then two test calls. Twenty minutes total.

Configure. Business name and how you want to be greeted, service list, service area, hours, price sheet upload with on-screen confirmation, escalation number. On Lite this is about ten minutes.

Forward your line. Your number does not change and nothing is ported — you add a forwarding rule at your carrier and calls arrive at the AI. Start with forward-on-no-answer if you are nervous: your phone rings first, and the AI catches only what you would have missed. That configuration changes nothing about how you work today and immediately stops the bleed. Switch to full forwarding when you are ready, which for most shops is somewhere around day three once the messages start proving themselves.

Make two test calls yourself. One easy — a scheduled rekey with a clean address. One hard — a mumbled street name, background noise, a caller who interrupts, a price question about something not on your sheet. The hard call is the informative one.

Confirm the message actually lands. Check that the Telegram message arrived, that the recording link plays, and that the person you named in prep can see it too. Do not skip this. A working receptionist whose alerts land in a chat nobody has opened is the most expensive failure mode in this entire process.

Set the expectation with your crew. Tell your techs the line is answered now and that jobs will arrive as messages, not as your voice relaying them. Ten seconds of communication prevents a week of "did you get my call?"

Days 2–3: listen, and correct the intake

This is the real work of the week, and it is nothing more than listening to recordings with a pen.

Every call generates a message with the recording attached. Pull the ones from your first 48 hours and listen to them at full speed — most are under two minutes. You are listening for four specific failures:

Wrong words. Street names, neighborhood names, local landmarks, vehicle models, and brand names the receptionist did not catch cleanly. Every trade has them. Note the ones that recur and add them to your service and area configuration so they are recognized rather than guessed.

Missing fields. A message that says "customer locked out, call back" is a failed intake. A good one has name, confirmed callback number, exact address, vehicle year/make/model where relevant, whether they have proof of ownership, what kind of lock or key, and how they heard about you. If a field you need is missing on more than one call, that is a configuration gap, not bad luck. The benchmark for what a strong first two minutes should collect is laid out in the first-90-seconds call handling checklist — hold your recordings against that list and the gaps are obvious.

Price answers you did not like. Listen specifically for how the receptionist handled pricing. It should state exactly what is on your confirmed sheet and take a message for anything else. If it stated a number you would not have quoted, that is a price-sheet correction, not a behavior problem — fix the sheet and it is permanently fixed.

Calls that should not have reached you at all. Robocalls, vendor pitches, wrong numbers. Screening handles the bulk of it automatically, and the residue is worth eyeballing once so you know what your junk baseline looks like. There is more on that pattern in spam and robocall screening.

Make your corrections in one sitting at the end of day 3 rather than reactively after every call. Batching keeps you from over-tuning on a single weird caller.

Days 4–5: tune escalation versus message

By day four you have thirty or forty real calls and you can see the shape of your traffic. Now you make the decision that determines how the receptionist affects your life: what interrupts you, and what waits.

Sort your captured calls into three buckets:

• Interrupt now. Someone is standing outside a car or a house, at night, in weather, or unsafe. Emergency commercial lockouts where a business cannot open. These get the urgent alert, immediately, with the address and callback number at the top of the message.
• Same-day, not urgent. Rekeys, lock changes, key duplication, a fob that works intermittently. These land as messages and get worked from the queue when you are between jobs.
• Next business day. Quote requests, general questions, commercial hardware inquiries, anything with a scheduling preference attached. These wait, and nothing is lost, because they are sitting in a searchable queue rather than a voicemail box.

Then tune two edges specifically, because these are where week-one setups usually get it wrong.

The over-escalation edge. If everything is urgent, nothing is — you will start ignoring alerts by week three, and that is the death of the system. If your phone buzzed more than a handful of times overnight in your first four days, tighten the urgent rule.

The under-escalation edge. The opposite failure is quieter and more expensive: a genuine emergency sitting in the general queue until morning. Re-read your urgent criteria against your actual overnight calls and ask whether any of them should have woken someone.

Also settle the safety boundary now if you have not: if a caller describes an emergency involving injury, a child or pet locked in a hot vehicle, or anything medical, the receptionist tells them to call 911 first. Intake never comes before an emergency, and that should be true in your configuration on day one, not day thirty.

This is also the right moment to think about where the captured information should end up beyond Telegram. If you want jobs flowing into your existing systems rather than living in a chat thread, CRM, calendar and Stripe integrations covers what connects to what and which parts are platform features rather than Lite features.

Days 6–7: review the week and decide what changes

Set aside twenty minutes at the end of the week. You are answering five questions, and you now have the data to answer all of them.

1. How many calls did you actually get? Almost every owner's first-week reaction is that the number is higher than they thought. That gap is the calls you were previously losing without knowing it.

2. How many were real? Total minus spam, wrong numbers, and vendor pitches. That ratio is your junk baseline, and it is worth knowing for its own sake — it is also what makes per-call billing models expensive elsewhere.

3. How many were captured completely? Full intake, nothing missing, no ambiguity you had to resolve by listening to the recording. If this is below 80 percent, you have one or two configuration gaps left, and days 2–3 told you what they are.

4. How many turned into work? Booked, quoted, or dispatched. Compare it honestly against what you think your previous conversion looked like on the same traffic.

5. How fast did you respond to the urgent ones? Not "did the AI answer" — it did, every time. The question is how long it took a human to act on the alert. This is the number nobody wants to look at and the only one that is entirely on your side of the line. Answering instantly and calling back in four hours is not coverage.

If question 5 has a bad answer, the fix is operational, not technical. Someone owns the alerts, or the alerts do not work.

The week at a glance

The three mistakes that cost week one

Nobody owns the alerts. Named above because it is first by a distance. The receptionist answers perfectly, the messages arrive in seconds, and they sit unread until Tuesday. If your shop has no one whose job it is to watch the feed during a given block of hours, fix that before you fix anything else.

Skipping the price sheet. Owners defer it — "we'll add pricing later" — and then spend three days annoyed that the receptionist takes a message on every pricing question. That is correct behavior with no sheet loaded; it is also the most common avoidable friction of week one. Upload it, confirm every parsed line on screen, and the receptionist starts quoting your real numbers exactly.

Tuning off a single bad call. One caller mumbles through a street name and the owner rewrites half the configuration. Batch your corrections at the end of day 3 and again at day 5, look for repeats rather than incidents, and you will make three good changes instead of twelve noisy ones.

What week two looks like — and when to upgrade

If week one went normally, week two is quiet. You stop listening to every recording and start spot-checking. Messages arrive, get worked, and the phone stops dictating your schedule. Most shops make one or two more configuration tweaks in the second week and then leave it alone.

The decision that comes up around week three or four is scope. Lite is a message-taking receptionist: it answers, screens, runs a real intake, states your confirmed prices, and delivers structured messages with recordings to your Telegram. It does not book onto a calendar, take payment, or dispatch. That limit is deliberate and we say it plainly rather than letting you discover it — the honest breakdown of when message-taking stops being enough is in Lite versus Core.

The signals that you have outgrown message-taking are consistent: you are calling every captured lead back to book them and that callback is now a job in itself; you are losing bookings in the gap between message and callback; or you want deposits collected on the call. That is the full platform — Core at $500/month with 500 AI minutes, Pro at $750 with 1,000, Elite at $1,200 with 2,500, each with live quoting, booking, dispatch, and payments, and each with a 14-day free trial. Everything, including overage rates, is on the pricing page.

There is no penalty for starting on Lite and staying there for a year. A large number of shops never need more than a receptionist that never misses a call and writes everything down. The locksmith-specific version of what that receptionist does on each call is on the locksmith AI receptionist page.

The bottom line

The ten-minute setup is real, and it is not the work. The work is one focused week: gather your services, price sheet, service area, hours and escalation number before you start; go live on forward-on-no-answer so nothing changes about how you operate today; spend fifteen minutes on each of days two and three listening to real recordings and correcting mis-heard local names, missing intake fields and price-sheet errors; spend day four and five deciding what interrupts you versus what waits, tightening the urgent rule until every overnight alert is deserved; and close the week by reviewing five numbers — call volume, junk ratio, complete-capture rate, jobs won, and human response time on urgent alerts. The single biggest predictor of whether this works is not configuration, it is whether a named person is actually watching the Telegram feed. KeyBot Lite is $149 a month with 100 calls included and 50¢ per minute after, no contract, live in about ten minutes, and your number stays. Before you commit a single day to it, spend thirty seconds at the demo and let it call you back as your own receptionist.

Frequently asked questions

How long does it take to set up an AI receptionist?

Initial setup on KeyBot Lite takes about ten minutes — business details, service list, service area, hours, price sheet upload, escalation number, and a call-forwarding rule at your carrier. Getting it genuinely tuned to your shop takes about a week of light work: roughly fifteen minutes a day on days two and three listening to real call recordings and correcting mis-heard local names or missing intake fields, then twenty minutes on days four and five setting escalation rules.

What do I need to have ready before setting up an AI receptionist?

Have seven things written down first: your service list in customer language, your price sheet, your service area with its honest edges, your hours including emergency hours, your escalation phone number, the rule for what counts as urgent, and — most importantly — the name of the person who will actually watch the alerts. Every one of these is information you already have; gathering it before you open the setup form is what turns a fumbling session into ten clean minutes.

How does the AI know my prices?

You upload your price sheet during setup, the system parses it, and you confirm every parsed price on screen before anything goes live. From then on the receptionist states those exact prices when a caller asks — word for word, never negotiated, never discounted, and never invented. Anything not on your confirmed sheet becomes a message for you to answer rather than a guess given to the caller on a recorded line.

Do I have to change my phone number or forward all my calls immediately?

No to both — your number stays exactly as it is, and you control how much traffic the AI sees. The recommended week-one setting is forward-on-no-answer, where your phone rings first and the AI catches only the calls you would otherwise have missed, which changes nothing about how you work today. Most shops move to full forwarding around day three once the captured messages have proven themselves, and the forwarding rule can be switched off at any time.

What should I measure at the end of the first week?

Five numbers: total calls received, how many were real versus spam and wrong numbers, how many were captured completely with no missing intake fields, how many turned into booked or quoted work, and how fast a human responded to the urgent alerts. The last one is the one owners avoid and the only one entirely on your side — the AI answered every call instantly, so a slow callback is an operations problem, not a product problem. If complete-capture is below about 80 percent, you have one or two configuration gaps left to fix.

What does it cost to run an AI receptionist for the first month?

KeyBot Lite is $149 per month with 100 calls included and 50¢ per minute after that, with no contract and no per-seat charges, and the first 5 answered calls are free on a 7-day trial. The full platform — which adds live quoting, booking, dispatch, and payments on the call — is $500 for Core, $750 for Pro, and $1,200 for Elite, each with a 14-day free trial. All included minutes and overage rates are published at https://www.thekeybot.com/pricing.