Website chatbot vs phone AI: which first for a small service business?

Tuesday at 2:14 p.m. a homeowner typed “same-day water heater” into your contact form while you were under a sink. The form sat. At 2:19 they opened Maps and called the next business. That same night at 10:40 p.m. a different caller hit your Google listing, got voicemail, and booked the company that answered. Two different leaks. One dashboard will not fix both.

Fix the channel that is already losing work. If most paid leads arrive on the website (forms, chat bubbles, after-hours FAQ) and die in an unread inbox, buy a website chatbot first. If most jobs still walk because the phone dumps to voicemail while you are on a job or after hours, buy phone coverage first: missed-call text-back, then an AI receptionist when the ring itself must get answered. Buy both only after one path is proving booked jobs for a month. Buy neither if you still cannot say which channel lost last week’s leads.

This is not “get every bot.” It is a Tuesday filter for a one-truck or small US service business that cannot afford two half-wired stacks.

The trap is buying the aisle you saw in an ad

Owners often paste a chat widget the week Google Ads spend goes up, then wonder why voicemail still eats the Maps calls. Or they buy a night voice bot because a competitor did, while the brochure site still has no hours, no service area, and no way to capture a form lead after 6 p.m.

The phone aisle and the website aisle answer different questions. Site chat answers people who already found your URL. Phone AI and missed-call text answer people who dialed the number on Maps, the truck wrap, or the ad. Mix them up and you pay for coverage on the quiet channel while the loud one keeps dumping.

If you want the phone fork in more depth, read AI receptionist vs live answering, recover missed calls without hiring a receptionist, and after-hours call coverage. This piece is the cross-aisle decision: which channel first.

Count last month before you buy anything

Pull thirty days of proof. You do not need a BI tool.

How many website form or chat threads sat longer than five minutes before a human reply? (See speed to lead if that number hurts.)
How many inbound phone calls went to voicemail or rang out while you were open?
How many after-hours rings hit voicemail and never became a booked morning job?

If the website pile is bigger, start in chatbots. If the phone pile is bigger, start with text-back or voice. If both are ugly, still pick one. Wire it, measure booked recoveries for four weeks, then add the second. Parallel installs on a one-person business usually mean neither gets tested from a real customer phone.

What a website chatbot actually does

A website chatbot is a widget (or a trained FAQ bot) on your site. It answers hours, service area, and pricing ranges. It captures name, phone, ZIP, and what they need when you cannot sit on the bubble. It is not a receptionist for the number on the truck.

Public floors from live Counter & Crew cards (check the review before you buy; dollars move):

Tidio: Free plan with 50 human conversations a month. Starter is $29 a month ($24.17 billed annually) for 100 human conversations. Growth starts at $59. Lyro (the AI agent) as a paid add-on starts at $39 a month for 50 AI conversations. Lyro does not answer the phone.
tawk.to: Core widget is Free forever. AI Assist Hobby is 100 free credits a month. Growth is $29 a month per property when you need more AI. You still have to staff live chat if you want a person in the bubble.
Chatbase: Hobby is $40 a month (700 credits). Free is 50 credits and tiny. You train on PDFs and your site. Good when you will babysit answers. Not a live help desk.
LiveChat: Starter is $25 a user a month ($19 annual). We rate it skip for most trades. Per-seat math gets expensive when two people answer.
Crisp: Mini is $45 a month. Essentials is $95 when you want WhatsApp and the chatbot. We rate it skip for plumbers, HVAC, salons, dentists, and law firms that needed a simple widget.

Smith.ai also sells chat, but the product people buy for missed jobs is still the phone path. Treat chat on Smith.ai as a second channel, not the reason to start there.

What phone AI and missed-call coverage actually do

Phone coverage is for the ring. Missed-call text-back fires an SMS when the line dumps. An AI receptionist greets the caller, takes intake, and (on the right tier) books or transfers. Live and hybrid floors put a person behind the bot when judgment matters.

Public floors from the same cards:

OpenPhone: Starter is $19 a user a month ($15 on annual). Missed-call SMS (auto-replies) sits on every paid plan. Sona is the named AI receptionist on credit packs (about 10 included calls, then paid packs from $25 to $199).
Rosie: Professional is $49 a month for 250 minutes. That plan texts a booking link. Calendar booking and live transfer start at Scale, $149 for 1,000 minutes. Growth is $299 for 2,000 minutes.
Smith.ai: AI Pro is $150 for 75 calls. Human Starter is $300 for 30. Hybrid is custom. Overage and live-agent transfers are the catch.
Podium: Custom quote, default 12-month auto-renew. Shared inbox covers SMS, website chat, and reviews. Missed-call text-back is real. There is no honest public monthly list price on the live card. Do not invent one.

For the phone-only explainer and the after-hours split, use the sibling posts linked above. Do not treat a chat widget as night coverage for Maps callers.

Buy, skip, catch

Buy website chatbot first when

Paid traffic and organic land on your site, and form or chat threads go cold for hours.
Callers already find you online and ask brochure questions (hours, ZIP coverage, “do you do X”) that a trained widget can answer.
You can reply to hot chat leads from a phone in under five minutes, or you will pay for AI replies that capture phone and ZIP when you cannot.
Floor: Tidio Free or Starter at $29, or tawk.to Free if someone will actually sit on the bubble. Chatbase at $40 only if you will train and re-test answers monthly.

Buy phone AI / missed-call stack first when

Google, ads, and the truck still send most jobs by phone, and voicemail is where money dies.
After-hours and mid-job rings outnumber unread website forms.
Floor: OpenPhone at $19 a user for shared SMS on the miss, Rosie at $49 when the ring must get a voice, Rosie Scale at $149 when you need calendar booking and live transfer, Smith.ai when a human safety net is worth the meter. Read missed-call text-back before you jump to a full voice plan.

Buy both when

One channel already recovers leads for a full month with a clear booked-job count.
Website visitors and phone callers are both real volume, not vanity.
You can name who owns chat replies and who owns night intake. Two unowned inboxes is worse than one.

Skip (for now) when

You cannot say which channel lost last week’s jobs. Measure first.
You want LiveChat or Crisp as a “pro” widget for a one-truck trade business. Tidio or free tawk.to is the closer widget buy; we skip LiveChat and Crisp as the default for that job.
You treat Podium as a cheap chat experiment. It is a year-long inbox decision with custom pricing.
You buy Rosie Professional thinking it books the calendar. The $49 plan sends a link. Booking and transfer start at $149.
You paste Lyro or any unsupervised AI on medical or legal advice. Tidio’s own terms push that away for a reason.

The catch

The quiet failure is wiring neither path all the way. A chat widget with no notification to the phone in your pocket is decoration. A voice bot with FAQs you never wrote loses the scared caller on the second question. A missed-call text that says “we will call you back” with no intake trades a ring for a dead thread. Seconds and fields beat a prettier logo. If you would rather not become the software person, done-for-you installs are flat rate and nationwide, or book a fit call.

A usable protocol for this week

Count last month’s website delays vs phone dumps vs after-hours voicemails. Write the three numbers on one note.
Pick the larger leak. Only one.
If website: install Tidio Free or Starter, or tawk.to Free. Write five answers (hours, service area, what you do not do, emergency vs estimate, callback window). Route hot leads to the phone you already carry. Test from a second phone after dinner.
If phone: turn on missed-call SMS on the line you already pay for (OpenPhone auto-replies are the cheap floor). If nights still die, add Rosie at the tier that matches booking vs pickup-and-text. If a miss is a real emergency job, read Smith.ai before you guess.
Run fourteen days. Count recovered threads that became booked work. Do not buy the second aisle until that number is real.
Only then add the other channel, or stop. Stopping is allowed.

FAQ

Can a website chatbot answer my business phone?

Not as the product you think you bought. Tidio Lyro does not answer the phone. Chatbase voice on higher tiers is a Twilio experiment, not a trades receptionist SKU. tawk.to Voice Agents is a separate add-on with custom usage. If the job is the ring on Maps, stay in AI receptionists and missed-call text-back.

Is free chat enough for a small service business?

Sometimes. tawk.to Free and Tidio Free are real. Free fails when nobody watches the bubble, when AI credits run out mid-month, or when the leak was never the website. Paid Tidio Starter at $29 is the common next step when Free is too thin and you still refuse a per-seat LiveChat bill.

Should I buy phone and chat in one suite?

Only if you will live in that suite. Podium bundles inbox, chat, SMS, and reviews on a custom 12-month path. That can be right for a multi-tech business that already wanted one inbox. It is the wrong first experiment for a one-truck owner who only needed a miss text or a $29 widget. Prefer separate, cancelable floors until the leak is proven.

What if both channels look fine on paper?

Then you may not have a bot problem. Check speed to lead, review ask after the job, and whether Maps is dialing the number you think it is. Tools are not the only leak.

Next step

Start in the aisle that matches the count: website chatbots for site leakage, missed-call text-back and AI receptionists for phone leakage. Read the live cards for Tidio, tawk.to, Chatbase, OpenPhone, Rosie, and Smith.ai before you sign anything annual. Or skip the shopping: installs are nationwide, or start here. We set up the channel that is actually losing jobs. Once. Quietly.