Before you pay for an AI website assistant, it makes sense to ask whether it will pay for itself. The honest answer is that it depends on your numbers: how many visitors and enquiries you get, how many you currently lose, what a booking is worth and what the assistant costs. This guide gives you a straightforward method and formulas to work it out with your own figures, worked examples for three types of business and a way to check the result after a month, so you are judging real outcomes and not promises.
The figures in this article are illustrations to show how the method works, not forecasts or benchmarks. Use your own numbers, and be cautious in your assumptions.
The simple version: one formula
Monthly return = (extra bookings or sales recovered × average value per booking) + (staff time saved × hourly cost) − monthly cost of the assistant.
If the result is positive, it pays for itself. To turn it into a ratio, divide the benefit by the cost. The work is in estimating the two benefit terms honestly.
Step 1: Find your baseline
- Website visitors per month, from your analytics.
- Enquiries per month by every route: forms, calls, messages, emails.
- Bookings per month that come from the website.
- Enquiries you cannot answer in time: out-of-hours messages, missed calls, forms answered a day later.
- Average value of a booking (revenue or gross profit; use the number you trust).
- Time your team spends on repetitive questions each week.
If you do not know some of these, estimate cautiously and mark them as estimates. You can improve them by tracking for a few weeks.
Step 2: Estimate the recovered enquiries
The assistant helps most where enquiries are currently lost. Think in three buckets.
| Bucket | How it helps | How to estimate |
|---|---|---|
| Out-of-hours enquiries | Answers and books at once | Number arriving outside working hours × share that would book if answered |
| Busy-hour enquiries | Answers while your team is occupied | Number missed or delayed × share that would book |
| Visitors who leave with a question | Answers the question that stopped them | Visitors who reach your pricing or contact page and leave × a small share |
Use conservative conversion rates. If you are not sure, assume that only 10 percent of extra conversations become bookings. You can adjust upward once you have real data.
Step 3: Estimate the staff time saved
If your team answers the same twenty questions repeatedly, count the minutes. Ten questions a day at three minutes each is 30 minutes a day, or around 10 hours a month. If the assistant handles most, multiply the hours saved by the hourly cost of the person who would have answered. Be realistic: some questions still need a person, and you will spend time reviewing the assistant.
Step 4: Add up the cost
| Hinta | Muistiinpanot |
|---|---|
| Platform subscription | For example $49, $149 or $399 a month for Crescelo; see pricing |
| AI usage | With bring-your-own-key, paid to your provider; see BYOK vs bundled pricing |
| Setup time | Your time to gather content and test; a few hours |
| Ongoing review | Ten to fifteen minutes a week |
Worked example 1: a two-chair salon
| Line | Number |
|---|---|
| Enquiries per month outside working hours | 40 |
| Share that would book if answered at once (cautious) | 15% |
| Extra bookings | 6 |
| Average booking value | $70 |
| Extra revenue per month | $420 |
| Staff time saved (4 hours at $15) | $60 |
| Benefit | $480 |
| Cost: Growth plan plus AI usage | about $165 |
| Net monthly return | about $315 |
Even with cautious assumptions, the assistant covers its cost about three times over in this illustration. If the salon books only two extra appointments, the result is roughly break-even.
Worked example 2: a dental practice
| Line | Number |
|---|---|
| New-patient enquiries missed or delayed per month | 25 |
| Share that would book if answered at once | 20% |
| Extra new patients | 5 |
| Average first-year value of a new patient | $600 |
| Extra revenue per month | $3,000 |
| Cost: Growth plan plus AI usage | about $165 |
| Return ratio | roughly 18 times the cost |
High-value patients make the arithmetic easy, which is why practices often see the strongest returns. The caveat is quality: the assistant must answer accurately and hand emergencies to a person.
Worked example 3: a home services contractor
| Line | Number |
|---|---|
| Missed calls and forms unanswered within an hour | 30 per month |
| Share that would become estimates | 20% |
| Estimates | 6 |
| Estimate to job conversion | 40% |
| Extra jobs | about 2.4 |
| Average job value | $450 |
| Extra revenue per month | about $1,080 |
| Cost: Starter or Growth plus AI usage | about $60 to $165 |
| Net | clearly positive at these numbers |
A small number of extra jobs can carry the cost, and instant lead alerts add to the effect by shortening response time.
What can make the return smaller
- Very low traffic, so few enquiries are lost to begin with.
- A low-value booking with thin margins.
- An assistant with poor content that answers badly or too rarely.
- No one acting on leads that the assistant captures.
- Setup that is never tested, so bookings fail or confirmations do not arrive.
The strongest lever is your own follow-through: read the conversations, fix gaps and respond to leads quickly.
Step 5: Check the result after a month
- Record bookings and leads from the chat.
- Tag those that came out of hours or during busy periods; these are the clearest gains.
- Ask new customers how they found you and whether the chat helped.
- Compare with the same month before, and with seasonality in mind.
- Read a sample of conversations to see missed opportunities.
- Recalculate with real numbers.
If the result is below your estimate, look first at content and follow-up before blaming the tool. See how to review your chatbot's conversations each week and cost per booked appointment.
Beyond the direct numbers
- Better data: you learn what customers ask and where your website fails to answer.
- Consistency: the same accurate answers at every hour.
- Team morale: fewer interruptions.
- Speed to lead: faster first response, which affects conversion; see speed to lead.
- Resilience: coverage during holidays and peaks.
These are harder to price, so do not count them in your first calculation, but they are real.
A blank calculator to copy
Copy this into a spreadsheet and fill in your own numbers. Each row has a formula in the third column.
| Line | Your number | How to work it out |
|---|---|---|
| A. Enquiries lost or delayed per month | Out-of-hours plus busy-hour plus abandoned visitors | |
| B. Share that would book if answered at once | Start with 10 to 15 percent and adjust with data | |
| C. Extra bookings per month | A × B | |
| D. Average value per booking | Revenue or gross profit, whichever you trust | |
| E. Extra revenue per month | C × D | |
| F. Hours saved per month | Repetitive questions per week × minutes each × 4 ÷ 60 | |
| G. Hourly cost of your time or staff | Wage plus overhead | |
| H. Staff time value | F × G | |
| I. Monthly cost | Platform plus AI usage plus review time | |
| J. Net monthly return | E + H − I | |
| K. Return ratio | (E + H) ÷ I |
If K is above 1, the assistant covers its cost on your assumptions. Test how sensitive the answer is by halving B. If the result is still positive, you have a comfortable margin; if it turns negative, you need more evidence before buying.
Break-even: how many bookings do you need?
Break-even is the number of extra bookings per month that covers the cost. Divide your monthly cost by the value of a booking. If the total cost is $165 and a booking is worth $70, you need about 2.4 extra bookings, so three. If a booking is worth $600, you need a fraction of one. This is the most useful number to write down before buying, because it turns a fuzzy question into a specific target that you can check after a month.
| Monthly cost | Value per booking | Bookings needed to break even |
|---|---|---|
| $60 | $50 | 2 |
| $60 | $300 | 1 |
| $165 | $70 | 3 |
| $165 | $450 | 1 |
| $430 | $120 | 4 |
| $430 | $800 | 1 |
Traps in ROI calculations
- Counting all chats as new business. Many chats come from people who would have booked anyway. Count only the extra ones.
- Using revenue rather than margin when your margin is thin.
- Ignoring the cost of your time to set up and review.
- Forgetting seasonality, which can make one month look better or worse than it is.
- Attributing everything to the assistant when other changes happened at the same time.
- Over-precision. A range is more honest than a single number.
A sensible way to present the result to a partner or team is a range: pessimistic, expected and optimistic, each with its assumptions.
Deciding whether to try it
Because the assistant has a low setup cost and monthly plans, the risk of trying it is small compared with the potential return in many businesses. A reasonable approach is to run it for one to three months, measure against the baseline you wrote down, and keep it if it beats the break-even. You can test how it answers on your own content first, at no cost, in the playground.
Frequently asked questions
How do I calculate the ROI of an AI chatbot?
Estimate the extra bookings or sales it recovers multiplied by their value, add staff time saved, subtract the monthly cost, and compare the benefit with the cost.
What conversion rate should I assume?
Be conservative. Assume a small share of recovered enquiries become bookings, then replace the assumption with real data after a month.
How quickly does an AI assistant pay back?
It depends on your traffic and booking value. Businesses with high-value bookings and many out-of-hours enquiries can see payback quickly; low-traffic sites take longer or may not benefit.
Should I count staff time saved?
Yes, but carefully. Count only the time that is genuinely freed, and remember you will spend some time reviewing the assistant.
What costs should I include?
The platform subscription, AI usage, the time to set up and a few minutes a week for review.
What if the numbers do not work?
Do not buy it yet. Track your enquiries and lost opportunities for a month, then recalculate.
Try the assistant on your own documents in the playground to judge answer quality first, then start a free trial to measure real results.
Further reading
- Google Search Central: creating helpful, people-first content: What Google looks for in useful pages.
VASTAA JOKAISELLE KÄVIJÄLLE. VARAA AIKA NIILLE, JOTKA OVAT VALMIITA.
Ota Crescelo käyttöön verkkosivustollasi 15 minuutissa
Opeta avustajasi hinnastoillasi, asiakirjoillasi ja palveluehdoillasi. Ota käyttöön lähteisiin perustuvat vastaukset ympäri vuorokauden ja päällekkäisyydettömät ajanvaraukset suoraan chatissa.