Selling AI Chatbots to Small Businesses: What Agencies Charge, What the Tools Cost, and How to Work Out Your Margin

In brief:
Some agencies publish what they charge to set up an AI chatbot or phone line, and the tools underneath cost much less. Nobody publishes the margin agencies keep.
This guide gives you the published prices, a formula to work out your own margin, and 8 steps before you sell your first setup, with copy-ready prompts and a sample call.

Short answer: Some agencies publish their prices. One charges $150 per website page to set up a chatbot plus a $297–$997 monthly retainer. Another lists $2,500–$7,500 one-time for a starter automation such as an AI receptionist or lead follow-up, with larger packages carrying monthly fees of $500–$3,000. The tools underneath cost far less. For a business getting about 100 calls a month, the ready-made AI phone answering services listed below cost roughly $70 to $213 a month. Nobody publishes what margin agencies actually keep, so this article doesn’t give you one. It gives you a formula and a worked example so you can calculate your own, plus the legal points to check before you sell.

This article is general information, not legal, tax or financial advice. Prices change often. Check every vendor’s page before you quote a client.

01What you’ll have by the end

• A price comparison of AI phone answering tools, converted to your client’s call volume.
• A margin calculation per client, for setup and for the monthly fee.
• A written scope: what the agent answers, and when it hands over to a person.
• A demo agent you’ve called yourself, including with hard questions.

02What you need before you start

• One niche and one task in mind (for example, after-hours calls for one type of service business).
• A rough idea of the client’s call volume: calls a month and average call length.
• Access to an AI writing tool (ChatGPT, Claude, Gemini or another) for the prompts below.

03Quick glossary: 6 terms used in this guide

045 numbers worth knowing

05What agencies charge (published prices)

Most agencies don’t publish prices. The two below do, and they show two very different models. Treat them as examples, not as “the market rate”.

Per-page setup plus a retainer (AgentScott). A website chatbot costs $150 per page to set up, plus $0.10 per word for extra documents (minimum $50 per document). The monthly retainer is $297 for sites up to 10 pages, $497 for 11–30 pages and $997 for 31–75 pages. Usage is billed on top, at about $0.024 per response on the lighter model or about $0.10 on the standard one. The page’s own examples: an 8-page site costs $1,547 in the first month and $347 a month after that. A 22-page site with two documents costs $4,127 in the first month and $647 a month after that. Extra work costs $100 an hour ($75 for retainer clients). One caution: one of the page’s examples uses a $97 base (and bills the ‘Lightweight’ example at the $0.10 Standard rate). Its FAQ also says ‘There are no setup fees for retainer plans’, which conflicts with the $150-per-page setup fee. Ask the agency which applies.

Project build plus monthly support (SuperDupr). This agency’s pricing guide (updated Oct 1, 2026) lists a starter automation such as an AI receptionist or lead follow-up at $2,500–$7,500 one-time, a growth package at $7,500–$25,000 plus $500–$1,200/mo, and an ops platform at $25,000–$75,000+ plus $1,000–$3,000/mo. Ongoing support is $300–$1,500 a month. Its “market” ranges and ROI figures name no source, so we don’t repeat them.

Freelance rates for the same work. On Upwork, the median hourly rate for chatbot developers is $30–$61. Upwork’s AI engineer cost page says a basic chatbot built on third-party software costs “$500 to $1,000”. Upwork says these figures come from historical contracts worldwide and that the final price is negotiated (chatbot developers, AI engineers). The same job is published at anywhere from about $500 (Upwork’s basic-chatbot estimate) to $7,500 (SuperDupr’s starter package), plus monthly fees, depending on scope and support.

1. $500–$1,000 – Upwork, basic chatbot
2. about $1,500 first month – AgentScott, 8-page site
3. $2,500–$7,500 – SuperDupr, starter
4. Plus monthly fees. Published prices of single sellers, not a market average
5. Checked Oct 12, 2026

06What the tools cost: AI phone answering (virtual receptionist)

An AI answers the phone, takes messages, books appointments and sends a summary. Vendors price it in different units, so convert before comparing.

1. Per call or call bundle.
– Smith.ai AI Receptionist: a free plan with 25 calls a month, then $3.00 per call. The Pro plan is $150 a month for 75 calls ($2.00 per call), with extra calls at $2.50. Enterprise is $500 for 300 calls. Calls can be handed to live agents. Check the vendor’s terms for how spam calls are counted.
– Upfirst: Starter $24.95 a month for 30 calls (extra calls $1.50), Premium $59.95 for 90 calls (extra calls $1), Pro $159.95 for 300 calls. Annual billing is 20% cheaper. Check the page for how short or spam calls are billed and whether a trial is offered.
– Goodcall: bills per unique caller per month, not per call. Starter is $79 a month for 100 unique customers, then 79¢ each.

2. Minute bundles.
– Dialzara: Business Lite $29 a month for 60 minutes (extra minutes $0.48), Business Pro $99 for 220 minutes (extra minutes $0.45). A US/Canada local number is $3 a month (toll-free $5).
– Rosie: Professional $49 a month for 250 minutes, Scale $149 for 1,000 minutes.
– Frontdesk: $99 a month ($79 if billed annually). That includes 200 voice minutes, 100 chatbot conversations and 1,000 overage credits. Voice uses 25 credits per minute, and an auto-reload costs $10 per 1,000 credits.

3. Flat rate.
– NextPhone: Pro $199 a month with unlimited calls, and you set it up yourself. Growth is $299, and the vendor builds it for you.

4. Per minute, build it yourself (platform).
– Retell AI: voice costs $0.07–$0.31 a minute depending on the model and voice, and phone numbers are $2 a month. Ask how billing is rounded and whether silence on the line counts as billed time. There’s no ready-made receptionist. You (or the agency) build the agent.

Worked example: 100 calls a month

Our assumptions, not real data: 100 calls a month, averaging 3 minutes each (300 minutes).

The ready-made services range from $69.95 to $212.50 for this volume. Retell is cheaper per minute, but someone (usually the agency) has to build and maintain the agent. That is what a setup fee pays for.

For context: the median US receptionist earns $18.27 an hour (BLS, May 2025), wages only. A receptionist does things an AI line can’t, so this is context for a sales conversation, not proof the AI is “worth it”.

• Upfirst $69.95
• Goodcall $79
• Frontdesk ~$119
• Dialzara $138
• Rosie $149
• NextPhone $199

07What the tools cost: website chatbots

• Dialzara website chatbot: $39 a month standalone. It’s included with Dialzara voice plans (100 sessions a month, then $0.15 per session).
• Frontdesk: 100 chatbot conversations are included in the $99 plan.
• Zapier Chatbots: the Zapier pricing page lists 2 chatbots on the free tier and a paid tier at about $13.33 a month.
• Retell chat agents: billed per AI message, from about $0.0005 to $0.10 depending on the model.

08How to work out your margin

Monthly margin per client = what the client pays you − tool costs you pay − (your hours × the value of your hour)

Setup margin = setup fee − (build hours × the value of your hour) − any one-time tool costs

Worked example (our assumptions, not real data)

A dental office gets about 100 calls a month. You set up Upfirst Premium and keep the account under your billing.
– You charge a $297 monthly fee. That’s AgentScott’s published small retainer, used here only as a reference point. Your own market may pay more, less or nothing.
– Tool cost: $69.95 (from the table above).
– Your time: 3 hours a month for call reviews, script fixes and reports. AgentScott’s page estimates about 2–3 hours a month for its smallest retainer.

Before overhead and taxes, that leaves $297 − $69.95 = $227.05. Divided by 3 hours, that’s about $75.68 an hour. If the work takes 6 hours, it drops to about $37.84 an hour. Now compare that with the $30–$61 median Upwork shows for chatbot developers. Two more things to count: the hours you spend finding the client (which most people forget), and self-employment tax. In the US that’s 15.3% on 92.35% of net self-employment earnings; it applies once net earnings are $400 or more (IRS).

Client pays	$297
Tool cost (Upfirst Premium)	$69.95
Left before overhead and taxes	$227.05
Your hours	3 → $75.68/hr
If 6 hours → $37.84/hr

Run your own numbers. You can ask an AI tool to do the arithmetic, as long as you give it every price:

You are helping me calculate the monthly margin on one AI phone answering client.
Don't add any prices of your own. Use only the numbers I give you, and show each step in a table.

Formula:
Monthly margin = what the client pays − tool costs − (my hours × the value of my hour)

My numbers:
- Client pays me per month: [amount]
- Tool plan and monthly price: [plan name, price]
- Calls or minutes included in the plan: [number]
- Overage price: [price per extra call or minute]
- Client's volume: [calls per month] calls, [average minutes] minutes each
- My hours per month on this client: [hours]
- Value of my hour: [amount]

Show:
1. The tool's monthly cost at this volume, including overage.
2. The margin in dollars.
3. The margin per hour of my time.
4. The same calculation if my hours double.
End with one line: which assumption changes the result most.

Sample input: client pays $297; Upfirst Premium $59.95 for 90 calls, extra calls $1; 100 calls a month; 3 hours a month; value of an hour: your own figure.

What to check in the output:
– That the tool didn’t add a price you didn’t give it.
– That overage was counted only on calls above the plan’s limit (here, 10 calls).
– That “margin per hour” divides by your hours, not by the number of calls.
– Check one line by hand or on a calculator. AI tools sometimes get arithmetic wrong.

09Step by step: before you sell your first setup

Step 1: Pick one niche and one task

Pick one niche and one task, such as after-hours calls for one type of service business.

How to do it:
1. Write down 2–3 types of local businesses you understand.
2. For each one, write the most common reason people call after hours.
3. Choose the one where the answer is simple and repeatable (opening hours, booking, a message for the owner).

How you know it worked: you can say in one sentence who you serve and which calls the agent handles.

Step 2: Build a demo on a free plan or trial

Smith.ai has a free plan with 25 calls. Dialzara and Rosie each have 7-day trials. Call the demo yourself many times, including with hard questions.

How to do it:
1. Open a free plan or trial with one vendor.
2. Load your invented “business facts”: hours, services, booking link, and how to reach a person.
3. Call it at least 10 times: simple questions, a question it shouldn’t answer, an angry caller, a caller who asks for a person.

How you know it worked: you have notes on what the demo answered correctly, where it got stuck, and whether it handed over to a person when asked.

Invented example, for illustration: an after-hours call with a demo agent

Invented example for illustration. The business, hours and names are not real.

Agent: Thanks for calling Maple Street Dental. Our office is closed right now. I’m an automated assistant. How can I help?

Caller: Hi, can I book a cleaning for next week?

Agent: I can take your request. What’s your name, and which days work for you?

Caller: Sam. Tuesday or Wednesday morning.

Agent: Thanks, Sam. I’ve noted Tuesday or Wednesday morning. The office will confirm a time by text when it opens.

Caller: How much does a cleaning cost?

Agent: I don’t have approved prices. I’ll add your question to the message so the office can answer it.

Caller: And I have bad tooth pain right now.

Agent: I can’t give medical advice. If this is an emergency, please call your local emergency number. I’m marking your message as urgent for the office.

What the example shows: the agent said it was automated, took a booking request without inventing a time, didn’t make up a price, refused medical advice and flagged the message as urgent. That’s what your scope (step 5) should make it do.

Illustration: the example conversation above, shown as a chat screen.

Call summary – Maple Street Dental (demo)

AgentThanks for calling Maple Street Dental. Our office is closed right now. I’m an automated assistant. How can I help?

CallerHi, can I book a cleaning for next week?

AgentI can take your request. What’s your name, and which days work for you?

CallerSam. Tuesday or Wednesday morning.

AgentThanks, Sam. I’ve noted Tuesday or Wednesday morning. The office will confirm a time by text when it opens.

CallerHow much does a cleaning cost?

AgentI don’t have approved prices. I’ll add your question to the message so the office can answer it.

Step 3: Convert every vendor’s price to the client’s volume

Convert every vendor’s price to the client’s volume: calls a month times average length, as in the table above.

How you know it worked: you have one table, like the worked example, where every vendor’s price is shown as a monthly cost for this client.

Step 4: Decide who owns the account

Either the client pays the vendor directly and you charge only for your work, or you resell (for example through a white-label plan such as Frontdesk’s, from $500 a month). Write it down.

How you know it worked: your proposal has one sentence that says who pays the vendor, and what happens to the account if the client leaves.

• Client pays the vendor directly – you charge for your work
• You resell (e.g. white-label, from $500/mo) – you bill the client for everything
• Who pays the vendor?
• What happens if the client leaves?

Step 5: Write the scope

Write the scope: what the agent answers, when it transfers to a human, how many revision rounds are included, response times and the monthly hours included.

Prompt: draft the agent’s instructions and scope

Help me write instructions for an AI phone agent for one small business.
The agent should handle only this business's calls, not act as a general AI assistant.

Business details:
- Type and location: [for example, dental office in a small town]
- Calls the agent handles: [2–4 call types from step 1]
- Approved information: [hours, booking method, services, how to reach a person]
- When to transfer or take a message: [for example, prices, complaints, anything urgent]

Write plain-English instructions, under 200 words, that include:
1. Say it's an automated assistant at the start of the call.
2. Handle only the listed call types. For anything else, take a message or transfer.
3. Never invent prices, appointment times, promotions or promises.
4. Never give medical, legal or financial advice. For emergencies, tell the caller to contact emergency services.
5. Don't ask for full card numbers or government ID numbers.
6. Offer a person whenever the caller asks, or when unsure.
Then list 5 scope lines for my proposal: what's included, what's not, revision rounds, response time and monthly hours.

Sample input: “Dental office. Calls: after-hours booking requests, opening hours, messages for the office. Hours: Mon–Thu 8am–6pm. Booking: the office confirms by text. Transfer: anything urgent or about prices.”

What to check in the output:
– That there’s an explicit rule against inventing prices and times.
– That a person (transfer or message) is offered more than once: when asked, and when unsure.
– That the agent says it’s automated.
– That the scope lines include revision rounds and monthly hours, so the job has an edge.

How you know it worked: you have instructions for the agent and a written scope you can attach to the proposal.

Step 6: Price setup and monthly support separately

Run the margin formula with your real hours. Don’t start from someone else’s number. Our guide to AI automation agency pricing shows how published build-plus-monthly packages are structured.

How you know it worked: you have two numbers, a setup fee and a monthly fee, and a margin calculation behind each.

Step 7: Check the language

Test the agent in the language the client’s callers actually use. Pricing pages give counts (Upfirst 35+ languages, Frontdesk 20+), not a guarantee that your callers’ language works well.

How you know it worked: you’ve made several test calls in that language, including names, numbers and addresses, and noted what the agent got wrong.

Step 8: Review real calls in the first weeks

Review real calls in the first weeks and fix the script.

How to do it:
1. Once a week, read the call summaries.
2. Note three things: calls handed to a person, questions left without an answer, and the vendor’s usage or overage on the bill.
3. Turn every repeated unanswered question into an update to the approved information.

How you know it worked: after a few weeks you know how many calls the agent handles on its own and what the tool actually costs per month.

• Handed to a person9
• Unanswered4
• Usage this week27 calls
• Repeated questions
• Parking
• Insurance accepted?
• Example

10Where to be careful

• Health data. HHS lists a “third-party vendor Artificial Intelligence (AI) chatbot on a provider’s patient portal” that handles patient information as an example of a business associate. That situation requires a business associate agreement (BAA) (HHS). Retell lists HIPAA/BAA on Enterprise. If you serve clinics, confirm this with the vendor in writing and get legal advice.
• Outbound calls. The FCC ruled in February 2024 (FCC 24-17) that AI-generated voices count as an “artificial or prerecorded voice” under the TCPA, so these calls need the called party’s prior express consent (FCC). Inbound answering is one thing. AI calling out to lead lists is another.
• WhatsApp instead of phone or web chat? Meta’s per-message pricing works differently. See our Hebrew guide WhatsApp bot for business: costs and setup (in Hebrew).
• Your own marketing. The FTC’s 2024 rule bans fake reviews and testimonials, including AI-generated ones (FTC). Don’t invent case studies or “client results”.
• Hidden usage costs. Per-minute platforms may bill for silence on the line, overage rates differ between plans, and some vendors count unique callers rather than calls. Put a usage cap or a pass-through clause in your contract.
• Vendor terms. Check each vendor’s cancellation notice period before you sign a client. Prices and plans can change after you’ve quoted a client.
• Don’t promise more revenue. No source shows how many extra bookings an AI line produces. Sell answering calls you’d otherwise miss, and measure. To show a client the break-even math, use our formula for whether an AI tool pays for itself.
• Client data in AI tools. When you use the prompts in this guide, don’t paste real callers’ names, phone numbers or call recordings. Use invented examples.

11Bottom line

1. Published agency prices for the same chatbot job range from about $1,500 for a small site (AgentScott) to $2,500–$7,500 for a starter build (SuperDupr), with monthly fees on top.
2. The tools cost much less. For about 100 calls a month, ready-made AI receptionists run roughly $70–$213.
3. Your margin is your fee minus tool costs minus your real hours. Calculate it per client, and check health-data and outbound-call rules before you sell.

Facts were checked against official sources in an automated (AI) check by Mamashi’s fact checker. Updated Oct 12, 2026.

12Sources (all checked Oct 12, 2026)

• AgentScott – Pricing: https://aiagentscott.com/pricing/
• SuperDupr – AI automation pricing (updated Oct 1, 2026): https://superdupr.com/blog/ai-automation-pricing
• Smith.ai – AI Receptionist pricing: https://smith.ai/pricing/ai-receptionist
• Upfirst – Pricing: https://upfirst.ai/pricing
• Goodcall – Pricing: https://www.goodcall.com/pricing
• Dialzara – Pricing: https://dialzara.com/pricing
• Rosie – Pricing: https://heyrosie.com/pricing
• Frontdesk – Pricing: https://www.myaifrontdesk.com/pricing
• NextPhone – Pricing: https://www.getnextphone.com/pricing
• Retell AI – Pricing: https://www.retellai.com/pricing
• Zapier – Pricing: https://zapier.com/pricing
• Upwork – Chatbot developer cost: https://www.upwork.com/hire/chatbot-developers/cost/
• Upwork – AI engineer cost: https://www.upwork.com/hire/artificial-intelligence-engineers/cost/
• BLS – Receptionists, Occupational Outlook Handbook: https://www.bls.gov/ooh/office-and-administrative-support/receptionists.htm
• IRS – Self-employment tax: https://www.irs.gov/businesses/small-businesses-self-employed/self-employment-tax-social-security-and-medicare-taxes
• HHS – Business associates: https://www.hhs.gov/hipaa/for-professionals/privacy/guidance/business-associates/index.html
• FCC – Declaratory Ruling FCC 24-17: https://www.fcc.gov/document/fcc-confirms-tcpa-applies-ai-technologies-generate-human-voices
• FTC – Final rule banning fake reviews and testimonials: https://www.ftc.gov/news-events/news/press-releases/2024/08/federal-trade-commission-announces-final-rule-banning-fake-reviews-testimonials

---

QUESTION

Have a question about this guide?

Leave a phone number or an email and your question. We read every message.

Name (optional)

How should we reply? (phone or email, at least one)

Phone

Email

What would you like to ask?About "Selling AI Chatbots to Small Businesses: What Agencies Charge, What the Tools Cost, and How to Work Out Your Margin":

I agree that Mamashi may contact me about this inquiry (required). Details in our privacy policy.

I'd also like to receive marketing content from Mamashi (optional).

Leave this field empty

Your details go to Mamashi (getmamashi@gmail.com) and are kept in the site's system for up to 12 months.