Why most ROI claims are nonsense

Vendor ROI calculators overstate the return in three ways: they assume you capture none of your after-hours enquiries today, apply best-in-industry conversion rates, and leave out setup time, content work and the cost of a wrong answer.

Open any AI chatbot vendor's website and you'll find an ROI calculator. Plug in some numbers, watch it return "12× ROI in 30 days." The math behind these calculators is almost always inflated in three predictable ways.

Inflated baseline. The calculator assumes you currently capture 0% of after-hours inquiries. In reality, most businesses capture some - through email, voicemail, contact forms - so the "incremental" lift from a chatbot is smaller than the calculator suggests.

Inflated conversion rates. Vendor calculators assume the chatbot converts inquiries to customers at industry-best rates. In practice, your conversion rate is whatever your conversion rate already is. A chatbot capturing more leads at the top doesn't fix a leaky funnel below.

Hidden costs ignored. Setup time, content preparation, monthly review, occasional content updates, and the cost of a wrong answer that costs you a customer - none of this shows up in vendor calculators.

The result: a calculator that returns "$50,000 in additional revenue" when the honest number is closer to "$3,000-8,000 in additional revenue, plus $1,500 saved in staff hours, minus $2,400 in chatbot cost and roughly 8 hours of your time."

That's still a positive ROI for most businesses. It just isn't the number on the marketing page.

The honest formula

Three inputs, one division.

ROI = (incremental conversions × value per conversion + staff hours saved × hourly cost) ÷ monthly chatbot cost

A result above 1 means the assistant pays for itself. Below 1, it does not - yet.

Incremental conversions

Count only the customers you would not have won without the assistant: questions answered after hours, at weekends or while your team was busy, that turned into a booking, a sale or a qualified lead. Your current response time is the baseline. If you already answer email within the hour, an enquiry that would have come by email anyway is not incremental.

To measure it, tag every lead the assistant captures for one month and count how many arrived outside working hours or went unanswered by a person.

Value per conversion

Use the margin on a typical first purchase or booking, not the lifetime value of your best customer. If a new client is worth $50 in margin on the first visit, use $50.

Staff hours saved

Take the repeat questions the assistant answers in full - opening hours, prices, directions, what to bring - and multiply by the minutes a person spends on each. Five minutes a question and 200 questions a month is about 16 hours. Value an hour at what it costs you, including overheads, not at the wage alone.

Monthly chatbot cost

The subscription plus your own time: setup in the first month, then an hour or two a month to read the questions it could not answer and add what was missing.

A worked example

In the example below, six extra conversions and ten saved staff hours a month against $199 of total cost give an ROI of about 2.5, with break-even at about four conversions a month.

An illustrative small business, numbers rounded:

  • 6 extra conversions a month from after-hours questions, at $50 margin each: $300
  • 10 staff hours saved at $20 an hour: $200
  • Monthly cost: a $99 subscription plus 2 hours of your time at $50: $199

ROI = ($300 + $200) ÷ $199 ≈ 2.5

Without the staff time, the break-even point is the monthly cost divided by the value per conversion: $199 ÷ $50 ≈ 4 conversions a month. That is where the "3-5 captured leads" in the summary comes from. With a higher margin per customer, break-even comes sooner; with a lower one, later.

When the number comes out below 1

An ROI below 1 usually has one of three causes: few enquiries arrive outside working hours, the assistant cannot answer too many questions, or each conversion is worth little. Each has a different fix.

  • Too few enquiries outside working hours. If almost every question already reaches a person quickly, there is little to capture. The staff-time side may still justify it; check that before anything else.
  • Too many questions the assistant cannot answer. Every hand-off to a person is time not saved. Read the unanswered questions and add the missing pages or answers to its knowledge.
  • A low value per conversion. Then the case rests on staff time, and the assistant should be judged on the repeat questions it takes off your team.

Measure it, do not estimate it

Run the numbers before you start, with your own baseline, and again after 30 days with real counts: leads captured outside working hours, questions answered without a person, minutes those questions used to take. Real inputs replace guesses, and the formula stays the same.