Table of Contents
- Key takeaways
- What is the business case for a customer support chatbot?
- What are the benefits of a customer support chatbot?
- What are the limits of a customer support chatbot?
- How much does a customer support chatbot cost compared with a person?
- When does a customer support chatbot break even?
- Worked example: an online store with 2,000 routine questions a month
- When should you not use a customer support chatbot?
- How to choose the right customer support chatbot setup
- Where SiteGPT fits, and where it does not
- Frequently asked questions
- Benefits
- What is the main benefit of a customer service chatbot?
- Do customers like customer service chatbots?
- Cost and ROI
- How do you calculate chatbot ROI for customer service?
- Is a chatbot cheaper than hiring an agent?
- Limits and risks
- What are the disadvantages of chatbots in customer service?
- Can a chatbot handle refunds or order changes?
- Getting started
- How many questions do you need before a chatbot is worth it?
- How do you test the business case before paying for a year?
- Sources
Created by
Do not index
Created time
Sep 11, 2023 06:44 PM
A customer support chatbot pays for itself when it answers at least a few dozen routine questions a month that a person would otherwise handle, and when it hands everything else to a person with the conversation attached.
TL;DR
- The main benefits are instant answers at any hour, the same policy wording every time, and fewer routine questions in the team queue.
- The main limits are that the bot only knows what your content says, cannot act in systems it is not connected to, and loses customers after one bad answer. Only 27% of customers would try a chatbot again after a negative experience (Gartner, 2026).
- A person answering a 5-minute question costs about $1.79 in wages at the US median. SiteGPT Growth costs about 0.8 cents per answered question billed yearly, or 1.3 cents billed monthly.
- On Growth billed yearly ($79 a month), the bot breaks even once it takes about 45 routine questions a month off your team.
- Skip the chatbot if you get fewer questions than that, if the answers are not written down yet, or if most questions need an account action.
Key takeaways
Question | Short answer | Number to use |
What does a chatbot save? | Wage time on routine questions the bot answers correctly | $1.79 per 5-minute question at the BLS 2025 median wage |
What does it cost? | A plan fee plus review time | $79 a month for SiteGPT Growth billed yearly, plus about 2 hours a week of transcript review |
When does it break even? | When it removes enough routine questions to cover the fee | About 45 questions a month on Growth billed yearly |
What is the biggest risk? | Wrong or dead-end answers that push customers away | 27% would try a chatbot again after a bad experience |
When should you not use one? | Low volume, no written answers, or action-heavy questions | Under about 45 routine questions a month |
What is the business case for a customer support chatbot?
The business case for a customer support chatbot is the money and time it saves on routine questions, minus what it costs to run and what a wrong answer costs you. It is a narrow case. The bot earns its fee on repeat questions with written answers, such as shipping times, opening hours, plan limits, and how-to steps.
This guide covers that math and its limits. For what a chatbot is and how it works, read what a customer service chatbot is. For a ranked list of tools, read the best customer service chatbots.
What are the benefits of a customer support chatbot?
The benefits of a customer support chatbot are speed, round-the-clock cover, consistent answers, and fewer routine questions for the team. Each benefit holds only when the answer is correct.
Benefit | What the business gets | When it fails |
Instant answers | No queue for questions the content already answers | A fast wrong answer creates a second contact |
Cover at any hour | Night and weekend questions get an answer without a night shift | Urgent cases still wait for a person |
Consistent answers | Every customer gets the same policy wording | Outdated content gives every customer the same outdated answer |
Fewer routine questions | Agents spend time on disputes, sales, and complex cases | Nobody reviews transcripts, so content gaps stay open |
Many languages | One set of content answers customers in their own language | Policy terms that do not translate well need a check |
A record of what people ask | Unanswered questions show what the help center is missing | The data sits unused if nobody owns it |
Round-the-clock cover is the benefit with the largest dollar value. Staffing one chat seat every hour of the week takes about 4.2 people. The full math for that case is in the 24/7 customer service chatbot guide.
What are the limits of a customer support chatbot?
The limits of a customer support chatbot come from three places: what it knows, what it can do, and how much patience customers give it. Plan for each one before you count savings.
- It only knows your content. A knowledge-trained chatbot answers from the pages, docs, and files you give it. If the answer is not written down, the bot should say it does not know.
- It cannot act without a connection. Refunds, order changes, and account edits need a bot connected to the system that holds the data. A bot trained on content alone cannot do them.
- Customers give it few chances. Gartner surveyed 3,566 customers in February and March 2026. Only 27% would try a chatbot again after a negative experience.
- Stated interest is not usage. In the same survey, 49% said they would have used a company chatbot, but only 7% used one in their most recent service interaction.
- People still want a person. 87% of customers say access to a human agent is essential when a company uses GenAI in customer service.
The Gartner advice is to prioritize reliability over reach. Start with the questions the bot answers well, and widen the scope only after the results hold. A company also answers for what its chatbot says, as the cases in support chatbot examples show.
How much does a customer support chatbot cost compared with a person?
A customer support chatbot costs cents per answered question, while a person costs dollars. The comparison below uses one routine question as the unit.
- A person. The US Bureau of Labor Statistics puts the 2025 median pay for customer service representatives at $21.53 an hour. That is about 36 cents a minute in wages, before benefits and management. If a routine reply takes 5 minutes, it costs about $1.79.
- Fin. Fin charges $0.99 per outcome, at most once per conversation. One outcome can cover several questions in the same conversation.
- SiteGPT. A SiteGPT message is one visitor question plus the bot's reply. Growth includes 10,000 GPT-4.1-mini messages a month, or 1,000 GPT-4.1 messages.
Plan | Price | GPT-4.1-mini messages a month | Cost per answered question (all GPT-4.1-mini) |
$39 a month billed yearly ($468 a year), or $59 billed monthly | 4,000 | 0.98 cents yearly, 1.48 cents monthly | |
$79 a month billed yearly ($948 a year), or $129 billed monthly | 10,000 | 0.79 cents yearly, 1.29 cents monthly | |
$259 a month billed yearly ($3,108 a year), or $429 billed monthly | 40,000 | 0.65 cents yearly, 1.07 cents monthly |
These per-question figures assume the plan's full quota is used. A plan that answers 1,000 questions a month costs more per question than one that answers 10,000. The 5-minute handling time is an example, so replace it with your own team's number. For prices across more vendors, read the chatbot cost guide.
When does a customer support chatbot break even?
A customer support chatbot breaks even when the wage time it saves covers its monthly fee. Divide the fee by the wage cost of one routine question.
Plan and billing | Monthly fee | Questions a month to break even at $1.79 each |
Starter, billed yearly | $39 | 22 |
Growth, billed yearly | $79 | 45 |
Growth, billed monthly | $129 | 73 |
Scale, billed yearly | $259 | 145 |
The fee is the small part. Someone also has to read transcripts and fix the content behind weak answers. Two hours a week at the median wage is about $187 a month, and that cost applies on every plan.
Worked example: an online store with 2,000 routine questions a month
An online store's team answers 2,000 routine questions a month in email and chat. The store adds a chatbot on SiteGPT Growth, billed yearly.
- Visitors ask the bot 1,500 questions in the first month. That uses 1,500 of the 10,000 GPT-4.1-mini messages.
- The bot resolves 600 of them with no person involved. The other 900 go to the team, as before.
- Saved wages: 600 × $1.79 = $1,074 a month.
- Costs: $79 for the plan plus $187 for 2 hours a week of transcript review, or $266 a month.
- Net saving: $1,074 minus $266 is $808 a month, or $9,696 a year.
The 600 resolved questions are an assumption, not a benchmark. Measure your own resolution rate during the trial before you count on any saving. If only 150 questions resolve, the saving falls to $268.50 against $266 in costs, which is close to break-even.
When should you not use a customer support chatbot?
You should not use a customer support chatbot when the volume is too low, the answers are not written down, or the questions need actions the bot cannot take. Each case has a cheaper fix.
Situation | Why a chatbot does not pay | Do this instead |
Fewer than about 45 routine questions a month | The fee and review time cost more than the wage time saved | Publish a clear FAQ page and answer by email |
No help center or written policies | A knowledge-trained bot has nothing to answer from | Write the top 20 answers first, then add a bot |
Most questions need a refund, an order change, or an account edit | The bot can explain the policy but cannot do the task | Use an AI agent connected to your helpdesk or order system |
Every conversation is a high-value sale or a dispute | Customers expect a person and a wrong answer is expensive | Start chats with a person and let the bot cover after hours |
Patients will share health data covered by HIPAA | You need a signed Business Associate Agreement (BAA) | Use a vendor that signs one. On SiteGPT, the BAA is available on the Enterprise plan only |
How to choose the right customer support chatbot setup
The right customer support chatbot setup depends on where your answers live and whether the bot must act. Match your main need to the row below.
If you need | Pick | Why |
Answers to documented questions from your website, docs, and files | 1. SiteGPT, a knowledge-trained chatbot | Trains on your content, answers in 95+ languages, and escalates to a person on every plan. From $39 a month billed yearly |
A bot that resolves cases inside a helpdesk and takes actions | 2. An AI agent in your helpdesk, such as Fin | Fin charges $0.99 per outcome and works with Intercom and other helpdesks |
Answers at night without hiring a night shift | 3. A chatbot plus a day team | The 24/7 chatbot guide compares the cost with human cover |
Fewer than about 45 routine questions a month | No chatbot yet | A FAQ page and email cost less than any plan |
Where SiteGPT fits, and where it does not
SiteGPT is the first pick when the questions already have written answers. It trains on website URLs, sitemaps, files, and connected sources such as Google Drive, Dropbox, OneDrive, Box, Notion, GitHub, and YouTube. Growth and above refresh that content on a schedule, so answers follow your site.
Human support works on any plan. By default the AI escalates when the visitor asks for a person, when the situation is urgent, or when the visitor is frustrated after an answer that did not help. You can also start every chat with a person, or follow your working hours.
SiteGPT has real limits. It has no free plan, and the 7-day trial needs a credit card. Chat integrations such as Slack, Crisp, and Messenger start on Growth, and the Zendesk AI Agent needs Scale. There is no Intercom or HubSpot integration today, and no live page describes a built-in order lookup.
If the numbers above work for your volume, the 7-day free trial shows your real resolution rate on your own content. The build tutorial walks through the setup in 9 steps.
Frequently asked questions
Benefits
What is the main benefit of a customer service chatbot?
The main benefit is fewer routine questions reaching your team, answered instantly at any hour. That saving only holds when the bot answers correctly and hands the rest to a person.
Do customers like customer service chatbots?
Customers are open to them but quick to give up. In a 2026 Gartner survey, 49% said they would use a company chatbot, 7% used one in their latest service interaction, and 27% would try again after a bad experience.
Cost and ROI
How do you calculate chatbot ROI for customer service?
Multiply the routine questions the bot resolves each month by the wage cost of one question. Subtract the plan fee and the time spent reviewing transcripts. At $1.79 a question, 600 resolved questions save $1,074 a month before costs.
Is a chatbot cheaper than hiring an agent?
Per question, yes. A 5-minute reply costs about $1.79 in median wages, and SiteGPT Growth costs about 0.8 cents per answered question billed yearly. A chatbot does not replace an agent one for one, because escalated and complex cases still need a person.
Limits and risks
What are the disadvantages of chatbots in customer service?
The main disadvantages are wrong answers when content is missing or outdated, no ability to act without a system connection, and fast loss of customer trust after one bad answer. Customers also expect a route to a person.
Can a chatbot handle refunds or order changes?
Only if it connects to the system that holds the order. A chatbot trained on content alone can explain the refund policy but cannot issue the refund.
Getting started
How many questions do you need before a chatbot is worth it?
On SiteGPT Growth billed yearly, the bot breaks even at about 45 routine questions a month at $1.79 each. Starter billed yearly breaks even at about 22.
How do you test the business case before paying for a year?
Run the trial on your own content with real customer questions. Count how many the bot resolves without a person, then put that number into the break-even table above.
Sources
Checked on 3 November 2026.
Written by
