How the savings and ROI are calculated
A support chatbot saves agent time on the queries it resolves on its own. In exchange, you pay for AI usage, the chatbot platform and a one-time implementation. The calculator compares both scenarios:
Current cost (month) = queries × agent minutes ÷ 60 × agent cost per hour
AI cost per conversation = input tokens ÷ 1,000,000 × input price
+ output tokens ÷ 1,000,000 × output price
Cost with chatbot (month) = queries × AI cost per conversation
+ platform
+ unresolved queries × agent minutes ÷ 60 × agent cost per hour
Monthly savings = current cost − cost with chatbot
Annual savings = monthly savings × 12
First-year cost = implementation + (AI + platform) × 12
First-year ROI = (agent savings per year − first-year cost) ÷ first-year cost
Payback (months) = implementation ÷ monthly savingsThe chatbot answers every query first, so the AI cost applies to all of them. Queries it can’t resolve go to an agent, who still spends the usual time on them. If the chatbot costs more each month than the agent time it saves, the investment never pays back.
AI prices come from the list that is updated automatically every day, in US dollars per million tokens. If a conversation goes over a model’s long-context threshold, the calculator applies the higher price, as in the LLM API cost calculator.
Worked example
An online store’s support team handles 5,000 queries a month. Each one takes an agent 6 minutes, and an agent costs $28 per hour with benefits. The team expects the chatbot to resolve 40% of the queries on its own. A typical conversation uses 2,500 input tokens (instructions, help-center articles and history) and 400 output tokens. The platform costs $1,200 a month and the implementation $20,000.
With a model that costs $1.00 per million input tokens and $5.00 per million output tokens (sample prices; the calculator uses the current ones):
- AI cost per conversation: 2,500 ÷ 1,000,000 × $1.00 + 400 ÷ 1,000,000 × $5.00 = $0.0045
- Current cost: 5,000 × 6 ÷ 60 × $28 = $14,000 a month
- Cost with the chatbot: $22.50 of AI + $1,200 of platform + $8,400 for the 3,000 queries agents still handle = $9,623 a month
- Savings: $4,378 a month, or $52,530 a year
- First-year cost: $20,000 + ($22.50 + $1,200) × 12 = $34,670
- First-year ROI: 94%
- Payback period: 4.6 months
Amounts are rounded the same way as in the calculator; each step uses the unrounded values.
The AI itself is the smallest part of the bill: $22.50 a month against $1,200 for the platform and $8,400 for the agents. Even a model ten times more expensive would barely change the result. What matters most is how many queries the chatbot really resolves.
How the resolution rate changes the result
With the same volume and costs, this is how the result changes depending on the share of queries the chatbot resolves:
| Resolved by the chatbot | Monthly savings | Annual savings | First-year ROI | Payback |
|---|---|---|---|---|
| 20% | $1,578 | $18,930 | -3% | 12.7 months |
| 40% | $4,378 | $52,530 | 94% | 4.6 months |
| 60% | $7,178 | $86,130 | 191% | 2.8 months |
| 80% | $9,978 | $119,730 | 288% | 2 months |
At 20%, the chatbot saves money every month but takes more than a year to cover the implementation. Every additional 20 points adds about $33,600 a year in savings.
Tips for a better return
- Start with your most frequent questions. Order status, returns, hours and account access usually have clear answers the chatbot can give on its own.
- Keep your help content up to date. The chatbot can only resolve what is documented; outdated articles lead to wrong answers and handovers.
- Make handover to an agent easy. Customers who get stuck with the bot cost more than the time you save. Pass the conversation history so the agent doesn’t start from scratch.
- Measure real resolution, not deflection. A query only counts as resolved if the customer doesn’t come back through another channel.
- Control tokens. Send only the relevant articles and summarize long histories; the AI cost grows with every token you send.
- Run a pilot. Launch on one channel or topic, measure the resolution rate and adjust the numbers before scaling up.
What this estimate doesn’t include
It doesn’t include benefits that are hard to quantify, such as 24/7 availability, faster responses or agents who can focus on complex cases. It also leaves out some costs: maintaining the content, reviewing conversations, extra fees from platforms that charge per resolution, and the risk of incorrect answers. Add the recurring costs you can estimate to the platform cost, and use the result to compare scenarios, not as a guaranteed figure.
Frequently asked questions
How is the chatbot's ROI calculated?
The first-year ROI is the agent cost the chatbot avoids in a year, minus the first-year cost, divided by the first-year cost. The first-year cost is the implementation plus twelve months of AI usage and platform fees. The payback period is the implementation cost divided by the monthly savings.
Why does the AI cost apply to every query, not just the resolved ones?
Because the chatbot answers every customer first, including the ones it later hands over to an agent. Those conversations still use tokens. This makes the estimate slightly more conservative.
Where do the AI model prices come from?
From the same price list as our LLM API cost calculator. It is checked automatically every day against public sources that track the official rates, and the page shows the date of the last update and links to each provider's pricing page. Prices are in US dollars; if you pick another currency, the calculator suggests a reference exchange rate that you should adjust to today's rate.
What resolution rate should I use?
Start with the share of your queries that are repetitive and have a documented answer, such as order status, opening hours or password resets. If you already run a pilot, use the rate you measured. When in doubt, compare a cautious and an optimistic scenario.
Is the data I enter stored?
No. All calculations run in your browser. Values only leave your device if you use the share button, which puts them in the link you copy.