"Will an AI chatbot actually pay for itself?" is the first question most business owners ask us, and it's the right one. The honest answer is: it depends on your numbers, and you can estimate them in about ten minutes before spending anything.
Here's the simple model we use.
The five numbers you need
- Conversations per month across chat, email, WhatsApp and calls.
- The share that are repetitive: order status, prices, opening hours, booking questions, the same FAQs.
- The share of those a bot can fully resolve without a person. Be conservative here.
- Minutes a person spends per conversation, including looking things up.
- Fully loaded staff cost per hour: salary plus overheads, not just the wage.
Then add the costs: a one-time build cost and a monthly running cost (hosting, AI model usage, channel fees such as the WhatsApp Business API, and maintenance).
The calculation
- AI-handled conversations = conversations × repetitive share × resolution share
- Hours saved per month = AI-handled conversations × minutes per conversation ÷ 60
- Gross monthly savings = hours saved × staff cost per hour
- Net monthly savings = gross savings − monthly running cost
- Payback period = build cost ÷ net monthly savings
That's it. No magic, and every input is something you can measure or estimate yourself.
Where people go wrong
Overestimating the resolution rate
A chatbot that "handles" a conversation but leaves the customer confused hasn't saved anything. It's created a second conversation. Use a conservative resolution share for planning, then measure the real number in the pilot.
Forgetting running costs
Model usage scales with conversations. Channels like WhatsApp can add per-conversation fees. Someone also needs to update the knowledge base when prices or policies change. Put all of that in the monthly cost.
Ignoring the value that isn't time
Faster replies can mean more leads converted and fewer abandoned carts. Out-of-hours answers can capture enquiries you'd otherwise lose. These are real but harder to estimate, so treat them as upside rather than baking them into the payback.
Assuming headcount cuts
In most small businesses the time saved goes into faster responses, sales follow-up and better service, not redundancies. That's usually where the real return is.
How to get real numbers before you build
Export last month's conversations and tag the top 30 questions. This one exercise tells you:
- your actual repetitive share,
- which questions the bot must answer well,
- and gives the bot its first approved knowledge base.
If the repetitive share is low, a chatbot probably isn't your best first automation. Look at lead follow-up or data entry instead.
Run a small pilot
Scope the bot narrowly (one channel, the top questions), give it a clear handover to a person, and run it for two to four weeks. Compare hours spent and response times against your baseline. If the numbers hold, expand. If they don't, you've learned cheaply.
Try the free calculator
We built a free calculator that does this maths for you in USD, PKR, GBP, EUR, AUD or CAD. Change any input and the hours saved, net savings and payback update instantly. It runs in your browser and stores nothing:
AI Chatbot ROI Calculator on Hugging Face
If you're not sure your business is ready for a chatbot at all, start with the two-minute AI Readiness Checker.
I'm Ghulam Mustafa, founder of 313 Automations, an AI automation and software development agency in Lahore, Pakistan. What's stopping you from trying a chatbot? Let me know in the comments.
Top comments (0)