Search for the ROI of an AI support chatbot and you will drown in numbers: save 30%, deflect 80%, cut costs in half. None of them are yours. They are marketing averages pulled from someone else’s business, and they tell you nothing about whether a chatbot will pay off for your support operation.
This post gives you something more useful: a simple model you can fill in with your own numbers in about ten minutes. No spreadsheet gymnastics, no vendor spin — just the three inputs that decide whether automation is worth it.
The only formula that matters
Strip away the noise and chatbot ROI comes down to one line:
Monthly savings = (tickets deflected × cost per ticket) − platform cost
If that number is comfortably positive, automation pays for itself. If it is marginal or negative, you either fix your inputs or wait. Everything else is a footnote. Let’s fill in the three variables.
Input 1: Your fully-loaded cost per ticket
Most teams underestimate this because they only count salary. The real cost of handling a support ticket includes agent pay and benefits, the tools they use, management overhead, and the time lost to context-switching. A quick way to get close:
Cost per ticket = (monthly support cost, fully loaded) ÷ (tickets handled per month)
Add up everything you spend to run support in a month — people, software, a slice of overhead — and divide by ticket volume. For many small teams this lands somewhere between a few dollars and well over ten dollars per ticket. Use your real figure, not an average from a blog.
Input 2: A realistic deflection rate
Deflection is the share of tickets your chatbot resolves without a human. This is where honesty matters most, because it is the number vendors inflate hardest.
Not every ticket is deflectable. The ones that are tend to be repetitive and answerable from documentation: how-to questions, account and billing basics, status checks, policy questions. Bespoke problems, angry escalations, and account-specific bugs are not — and pretending otherwise is how chatbot projects earn a bad reputation.
Rather than guess, estimate deflection from your own ticket mix: what share of last month’s tickets were variations of your top 20 questions? That share is your realistic ceiling. Then assume you will capture a portion of it, not all of it, at least at first.
Input 3: The true platform cost
This is the easy one, with one addition people forget: setup time. Take the subscription cost of your chatbot platform and add the value of the hours you will spend building and maintaining it. A tool that is cheap but takes weeks of engineering to wire up is not cheap.
A worked example (illustrative)
Suppose a team handles 2,000 tickets a month at a fully-loaded cost of 6 dollars each. They review their ticket log and find that roughly 40% are repetitive tier-1 questions. They conservatively assume the chatbot will contain three-quarters of those in the first quarter — about 30% of all tickets, or 600 tickets a month.
- Tickets deflected: 600
- Cost per ticket: 6 dollars
- Gross monthly value: 3,600 dollars
- Platform plus amortised setup: say 300 dollars per month
- Net monthly savings: about 3,300 dollars
These numbers are illustrative — the entire point is to replace them with yours. But the structure holds: the deal lives or dies on ticket volume and deflection rate, not on the platform’s sticker price.
What the formula leaves out (mostly upside)
The model above is deliberately conservative because it only counts deflected tickets. In practice, a good assistant also delivers value the formula ignores: instant responses at 2am, leads captured after hours that would otherwise vanish, and, maybe most importantly, your best agents freed to work the complex, high-value conversations instead of resetting passwords.
There is downside it ignores too. A chatbot built on a thin or outdated knowledge base will deflect far fewer tickets and frustrate customers into more escalations, not fewer. Accuracy is not a nice-to-have here; it is the input that makes the whole model work.
How to de-risk the decision
You do not have to trust the estimate. Test it. Take your top 20 questions, ground a chatbot in the documentation that answers them, and put it live on one page or one segment for 30 days. Measure real containment — the share of conversations that end without a human — and feed that number back into the formula. Now your ROI is based on your data, not a guess.
The honest conclusion
If your ticket volume is tiny, or nearly every question is genuinely bespoke, a support chatbot may not pay off yet — and it is better to know that before you buy than after. But for most growing B2B and SaaS teams drowning in repetitive tier-1 questions, the math is not close. Run your own numbers and find out.
Monology is built to make this measurable: chatbots answer only from your verified knowledge base, and built-in analytics show your real deflection and containment rates so your ROI model runs on facts, not marketing.
