Chatbot ROI calculator
Hours saved, cost saved, and how long the build takes to pay for itself. It is built to return a no: on low volume, on thin margins, and on deflection rates nobody has measured. The arithmetic runs in your browser and the link in the address bar carries the result.
Result
Payback lands inside the horizon, with enough margin to absorb being wrong.
13.0 months to recover $18,000, then $1,380 a month after that. By month 18 the position is $6,840, build cost included.
How that is arrived at
- Tickets deflected per month
- 400
- Agent hours saved per month
- 60
- Agent hours saved per year
- 720
- Gross saving per monthHours saved multiplied by loaded cost. Before anything is spent.
- $2,280
- Net saving per monthGross saving less the running cost. This is the only line that pays back a build.
- $1,380
- Payback period
- 13.0 months
- Position at month 18Net saving across the horizon, less the build cost.
- $6,840
The two things that break this answer
The deflection rate you need
16.7 percent is the rate at which payback lands exactly on 18 months. You entered 20 percent. You are above the line, so the question is whether the rate is real. Deflection is the one input nobody can check for you, and it is the one every vendor is happiest to supply.
What happens if the rate is half of that
At 10.0 percent, payback moves to 75.0 months. That is outside the horizon, so the whole case rests on hitting the rate you entered. Run this line before the build, not after.
What we would say to this
This is worth scoping. The number that decides it is not on this page: it is how often the system is wrong, and who carries it when it is. A deflection that answers confidently and incorrectly costs more than the ticket it saved.
Retrieval builds start at $18,000. Before that, the readiness assessment tells you whether your data and sign-off are in a state that supports one, and it is willing to answer no.
Five lines of arithmetic, written out.
Nothing is hidden in the tool. If you would rather run it in a spreadsheet, or check what the calculator did, this is the whole model.
- 01
Deflected tickets per month
monthly volume × deflection rate
Deflection means fully resolved with no person involved. Tickets the system touches and then hands over are not deflected; they are the same ticket with an extra step in front of it.
- 02
Agent hours saved per month
deflected tickets × handling time in minutes ÷ 60
Handling time is the time a person spends on the ticket, not the time the customer waits. Queue time is a different problem, and a chatbot moves it for different reasons.
- 03
Gross saving per month
hours saved × loaded cost per hour
Loaded cost, not salary: employer taxes, benefits, tooling, management overhead. Using base salary sounds conservative, but it mostly produces a model your finance team will not recognize.
- 04
Net saving per month
gross saving less running cost
The line that actually pays back a build. Model calls, hosting, monitoring and maintenance all sit here, and a business case that omits them is not a business case.
- 05
Payback period
build cost ÷ net saving per month
If the net saving is zero or below there is no payback period at all, and the calculator says so rather than printing a very large number and leaving you to squint at it.
The calculator also solves the fourth line backwards, for the deflection rate at which payback lands exactly on eighteen months. That single figure is usually more useful than the payback period itself, because it converts an argument about a business case into an argument about one number that can be tested against your own ticket history.
Three things that will move the answer more than any input on this page.
A calculator is a way of being precisely wrong faster. These are the parts it cannot see, and they are the parts that decide most real projects.
Hours saved are not headcount saved
Ninety agent hours a month is not half a person you can remove. It becomes money when it either absorbs growth you would otherwise have hired for, or is deliberately redirected to work that was being dropped. If neither happens, the hours dissolve into the working day and the saving never appears in a budget.
A confident wrong answer costs more than the ticket it saved
Deflection has a quality gate hiding inside it. A system that resolves a billing question incorrectly has not saved a ticket, it has created a complaint, and in a regulated line it may have created a disclosure. That risk is not on this page because it is not something you can guess at; it has to be measured.
The build cost is the smaller number
Across three years the running cost usually exceeds the build cost. That is the normal shape for software, but it means a model that compares a one-time price against a monthly saving and stops there is answering an easier question than the one in front of you.
Then that is the useful output, and there is nothing to buy.
We publish this knowing it will talk some people out of an engagement, which is the point. A calculator that recommends buying on every input is recognized as a sales tool within about thirty seconds and is correctly ignored. If the numbers say no, the cheaper moves come first: reduce handling time on your three most common ticket reasons, publish the answers people are writing in to ask, and come back when volume has genuinely moved.
If someone is selling you one regardless, a vendor AI assessment at $2,500 over three days is the proportionate spend. We assess the supplier before you sign rather than after.
Chatbot ROI, answered directly.
These answers are the same ones in the structured data on this page, so an answer engine quoting us quotes this.
- How is chatbot ROI calculated?
- Multiply monthly ticket volume by the deflection rate to get deflected tickets. Multiply that by average handling time in minutes and divide by 60 to get agent hours saved per month. Multiply by the loaded cost per agent hour to get the gross monthly saving. Subtract the monthly running cost to get the net saving. Divide the one-time build cost by that net figure to get the payback period in months. If the net saving is zero or negative, no payback period exists.
- What is a reasonable payback period for an AI chatbot?
- This calculator holds itself to 18 months. Past 24 months the assumptions behind the model, meaning ticket mix, model pricing and the product itself, will usually have changed more than the forecast allows for, so a payback out there is a projection rather than a plan. Inside 9 months the case is strong enough that the deflection rate becomes the only serious argument.
- What deflection rate should I assume for a support chatbot?
- Assume nothing. Deflection means the ticket was fully resolved without a person, so a ticket answered by a bot and then re-opened has been deflected zero times. It is the single input that decides the answer and the one input nobody outside your business can measure for you. Run the calculation at half whatever rate you were quoted; if the case only works at the full rate, the case rests entirely on an unverified number.
- When is an AI chatbot not worth building?
- When the net monthly saving is negative, which happens whenever the running cost exceeds the value of the hours saved. Low ticket volume is the usual cause: a desk handling a few hundred tickets a month rarely saves enough agent time to carry model calls, hosting, monitoring and maintenance. It is also not worth building when payback only lands inside the horizon at a deflection rate that has never been measured on your own tickets.
- What should the running cost input include?
- Model calls, hosting, monitoring and the engineer time to maintain it. Leaving running cost at zero is the most common error in chatbot business cases, because a deployed system is a system somebody has to keep working: content goes stale, models are deprecated, and prompts that worked last quarter drift.
- Does this calculator send my figures anywhere?
- No. The arithmetic runs entirely in your browser and there is no server call and no database. The only record of what you entered is the query string in your own address bar, which is what makes a result shareable by sending the link.
- What does a support AI build cost at MetaMinds?
- Retrieval and document intelligence builds start at $18,000, AI agent and workflow automation at $22,000, and a guardrail and evaluation platform at $24,000. A three-day vendor AI assessment, where we assess a supplier you are about to buy from, is $2,500. Every price is published on the services page.
Two more tools alongside this one: the AI readiness assessment and the AI stack builder. Both work the same way: client-side, shareable by link, and willing to tell you to do nothing.
Send us the result, including a bad one.
If the calculator said no, that is a thirty minute conversation about what would have to change and roughly when, which is worth more to both of us than a proposal.
Typical reply within one business day · Vendor AI assessment $2,500 · Builds from $18,000