How to calculate the ROI of an AI support chatbot (with a worksheet)

A simple, honest way to estimate whether an AI support chatbot pays for itself: the formula, a worksheet you can copy, and the numbers you must measure yourself instead of borrowing from a vendor.

DDivyaSoft Team 8 min read
How to calculate the ROI of an AI support chatbot (with a worksheet) cover image

An AI support chatbot pays for itself when the support work it fully resolves is worth more than the chatbot costs to set up and run. The calculation takes four numbers from your own business: how many conversations you get, what each costs you today, what share the chatbot can resolve on its own, and what the chatbot costs per month. This guide gives you the formula, a worksheet to copy and the traps to avoid. The example figures are made up to show the method. Replace them with yours.

What does "ROI" mean for a support chatbot?

For support, return on investment is the monthly value of the work the chatbot removes or speeds up, minus its monthly cost, compared with what you spent to set it up. It has two parts:

  • Hard savings. Support hours you no longer need to buy, or new hires you can avoid as volume grows.
  • Soft gains. Faster first replies, answers outside office hours, fewer abandoned carts, happier customers. These are real but harder to price, so track them separately and do not let them inflate the maths.

Start with hard savings. If the case works on those alone, extra benefits are a bonus.

What numbers do you need before you start?

Collect these from your helpdesk, inbox or chat history for the last three months:

InputWhere to find it
Support conversations per monthHelpdesk, shared inbox, chat and WhatsApp logs
Share that are repetitive questionsRead a sample of 100 to 200 conversations and tag them
Fully loaded cost of support per monthSalaries, benefits, tools, management time
Hours (or minutes) of agent time per conversationHelpdesk reports, or time a sample
Current first-response timeHelpdesk reports
Chatbot subscription or build costVendor quote or your build estimate
One-off setup costYour time plus any paid setup

Do not guess the share of repetitive questions. Tag a real sample. Typical categories are order status, shipping and returns, opening hours, pricing, how-to questions and account changes.

What is the formula?

  1. Cost per conversation today = monthly support cost ÷ conversations handled per month.
  2. Resolution rate = conversations the chatbot fully resolves without a person ÷ conversations it receives. Measure this in a pilot. Do not assume it.
  3. Conversations resolved by the chatbot = monthly conversations × share routed to the chatbot × resolution rate.
  4. Monthly savings = conversations resolved by the chatbot × cost per conversation.
  5. Net monthly benefit = monthly savings − monthly chatbot cost.
  6. Payback period (months) = one-off setup cost ÷ net monthly benefit.
  7. First-year ROI = (net monthly benefit × 12 − setup cost) ÷ setup cost.

A conversation counts as resolved only when the customer got a correct answer and did not need to contact you again about the same issue. Counting every chat the bot touched overstates savings.

Worked example (illustrative numbers only)

A small online store. These figures are invented to demonstrate the calculation.

LineValue
Support conversations per month1,200
Monthly support cost (fully loaded)$2,400
Cost per conversation$2,400 ÷ 1,200 = $2.00
Share that are repetitive and in scope for the bot60% (720 conversations)
Resolution rate measured in the pilot50%
Conversations resolved by the bot720 × 50% = 360
Monthly savings360 × $2.00 = $720
Chatbot monthly cost$150
Net monthly benefit$720 − $150 = $570
One-off setup cost$600
Payback period$600 ÷ $570 ≈ 1.1 months
First-year ROI($570 × 12 − $600) ÷ $600 ≈ 9.4, or 940%

Now stress-test it. If the resolution rate is only 25%, monthly savings drop to $360, net benefit to $210, and payback stretches to about 2.9 months. The case still works, but the margin is thinner. Always run a low case.

Copy-ready worksheet

Paste this into a spreadsheet and fill column B.

RowA: ItemB: Your number
1Conversations per month
2Monthly support cost
3Cost per conversation (row 2 ÷ row 1)
4Share in scope for the chatbot (%)
5Resolution rate from pilot (%)
6Conversations resolved (row 1 × row 4 × row 5)
7Monthly savings (row 6 × row 3)
8Chatbot monthly cost
9Net monthly benefit (row 7 − row 8)
10One-off setup cost
11Payback in months (row 10 ÷ row 9)
12Low case: repeat rows 5 to 11 with half your resolution rate

What costs do people forget?

  • Content work. Someone must write and update the answers and policies the bot uses. Budget a few hours a week at first.
  • Review time. A person should read a sample of conversations every week.
  • Channel fees. Messaging platforms such as WhatsApp charge for some conversation types. Check the platform's current price list.
  • Integrations. Order lookups and bookings may need setup work.
  • Hand-off cost. Conversations the bot cannot solve still reach a person, sometimes with more context needed. A good hand-off saves time. A bad one costs more than no bot.

What benefits are worth tracking even if you do not price them?

  • First response time, which drops to seconds for covered questions.
  • Out-of-hours coverage for customers in other time zones.
  • Conversion help: pre-sale questions answered while the shopper is still on the page.
  • Agent focus: staff spend time on complex and high-value conversations.
  • Customer satisfaction on bot-handled chats, if you ask for a rating.

Record these before and after launch. Convert them to money only if you can defend the conversion, for example by comparing conversion rates of visitors who chatted with those who did not, over a long enough period.

How do you run a pilot that gives you a real number?

  1. Pick one channel and the five most common question types.
  2. Load accurate answers and your real policies.
  3. Launch to a share of traffic for two to four weeks.
  4. Read every unresolved conversation and ask: wrong answer, missing knowledge or out-of-scope?
  5. Fix gaps, then measure the resolution rate again.
  6. Use the pilot's rate, not a vendor's headline figure, in the worksheet.

Common mistakes

  • Using a vendor's claimed deflection rate instead of your own measured rate.
  • Counting conversations the bot touched rather than resolved.
  • Leaving out the cost of maintaining the knowledge base.
  • Assuming staff cost disappears. Unless you reduce hours or avoid a hire, savings are time freed, not cash saved.
  • Forgetting the risk side: a wrong answer about refunds or safety can cost more than many correct ones save.

Frequently asked questions

What is a good resolution rate for an AI support chatbot?

It depends on your questions, your data and how well the bot is set up. There is no honest universal number. Measure it in a pilot on your own conversations and improve it from there.

How long does it take for a chatbot to pay back?

If your volume is high and many questions are repetitive, payback can be quick. If your volume is low or your questions are complex, it may not pay back at all in cash terms. The worksheet shows which case you are in.

Should we count customer satisfaction in the ROI?

Track it, but keep it out of the core calculation unless you can show how it changes revenue or retention. Savings you can defend are more convincing than gains you cannot.

Does an AI chatbot replace support staff?

For most businesses it handles the repetitive layer and frees people for harder work. Whether that reduces headcount depends on your volume and growth. Plan for hand-off to people for anything sensitive.

Can DivyaSoft help us estimate this?

Yes. Send us your monthly volume and a sample of questions and we will help you size a pilot. See our website chatbots, AI chat services and pricing, or get in touch. If you sell on Shopify or WooCommerce, read how an order tracking chatbot reduces where-is-my-order tickets.

#AI Chatbots#Customer Support#ROI

Need this done for you?

Talk to the DivyaSoft team

Software, AI, video ads and digital marketing from our team in Kathmandu. Tell us what you need and we reply by email with questions or a quote.

Get new guides by email

Ready to grow with AI, marketing & technology?

Book a consultation and tell us what you're working on. We'll get back within one business day with a concrete next step — not a sales pitch.