Calculator
AI Chatbot ROI Calculator
Most ROI cases for a website assistant count deflected tickets and stop there, which understates it for a marketing site and overstates it for a support desk. This counts both sides separately -- hours you stop paying for, and buying questions that turn into leads -- then subtracts what the thing actually costs, including the setup nobody budgets.
- Two returns, counted apartDeflection and leads are not the same money
- Setup cost includedThe line most cases leave out
- Runs in your browserNothing is sent anywhere
Work it out
Your numbers
Result
- Handling time saved, per year —
- Hours returned to the team, per year —
- Lead value, per year —
- Total cost, first twelve months —
- Return on cost —
- Break-even month —
- Larger contributor —
Every figure above is yours; nothing is assumed on your behalf and nothing is sent anywhere. Two honest cautions. Deflection is the number vendors inflate most, so if you are unsure, run it at 25 percent and see whether the case still holds. And lead value only counts if someone actually follows up -- a captured lead nobody calls is worth zero, not the number above. For the month you break even rather than the annual figure, use the <a class="tlink" href="/tools/assistant-payback-calculator/">Assistant Payback Calculator</a>.
What the numbers mean
Count the two returns apart, always
Deflected handling time and lead value come out of different budgets and convince different people. Adding them into one number before showing the split is how ROI cases get picked apart in the meeting where they matter.
Deflection is not resolution
A deflected conversation is one the visitor never had to escalate. If the assistant answers and they open a ticket anyway, that is not deflection, and counting it as one is how these figures end up wrong by half.
A lead nobody follows up is worth zero
Lead value assumes somebody acts on the signal. If buying questions land in a transcript nobody reads, set the capture rate to zero and the case has to stand on deflection alone.
Setup belongs in year one
Source mapping, configuration and QA are real cost whether you pay money or spend your own hours. Leaving them out is the most common reason a first-year ROI number does not survive contact with finance.
Fully loaded rate, not the salary rate
Salary plus employment cost plus tooling plus context switching, divided by actual working hours. Teams that use the headline hourly rate understate the saving by roughly half, which makes the case weaker than it really is.
Questions
Yes. Payback answers when the assistant covers its own cost, which is a timing question. This answers what the twelve-month number is and which line it comes from, which is a budgeting question. Run both; they use the same inputs.
Twenty-five to forty percent for a first deployment on ordinary documentation. Vendors publish sixty to eighty because those are their best customers on mature help centres. If the case only works above fifty, treat that as the finding.
Set the capture rate to zero and read the deflection half on its own. The reverse is true for a marketing site with no support desk: set deflection low and let lead value carry it.
Your own estimate, or ours. On a done-for-you pilot the source mapping, configuration and QA are on us, so the honest thing to model is your internal review time rather than a services fee.
No. It computes in the page as you type. There is no request, no storage, and nothing is retained when you close the tab.
That is what the free teardown produces. We run your questions against your current assistant and Creobot, and the deflection and refusal rows come back measured rather than assumed.
Want this run against your actual site?
Tell us the page your assistant lives on and the questions you care about. A person runs them and sends back the raw answers.