"What does it cost?" is often the first question and the hardest to get a straight answer to. The price of an AI agent is made up of several parts that are rarely shown side by side. This guide goes through what drives the cost, how to calculate the total and what to ask so you can compare offers on equal terms. It contains no price lists, because those change. Current prices are on the pricing page.
The seven parts of the cost
| Part | What it consists of | What is often missed |
|---|---|---|
| Subscription | A fixed fee for the platform, often per plan | What the plan includes and what requires an upgrade |
| Usage | Conversations, bookings or leads that are billed separately | How the price behaves when volume grows |
| Channels | Web chat, email, messaging channels and phone | That each channel can have its own pricing |
| Telephony | Numbers, call minutes and any forwarding | That telephony is often billed by another party |
| Connections | Calendar, customer records, shop and ticketing systems | Whether a connection is included or costs extra |
| Setup | Time to write knowledge, set boundaries and test | Your own working time, even if no invoice is sent |
| Ongoing upkeep | Updating knowledge and following up handovers | That it is not a one-off job |
The pricing models and when they fit
Three models are common. A flat monthly fee is easy to budget and fits when volume is high and steady. Price per conversation can suit low or uneven volume, but becomes harder to predict when campaigns or seasons lift traffic. Price per booking or lead ties cost to outcome, but requires that you agree on what counts as a booking or a lead. Read the comparison of pricing models for a closer look.
Always calculate at least three scenarios: a quiet month, a normal month and a busy month. A price that looks good in the vendor's example can look different at your actual peak volume.
How to calculate total cost
Use the same basis for both sides of the comparison. That way you avoid comparing a salary cost with a subscription fee and calling it a saving.
- Define the need. Which hours, which channels and which case types should be covered?
- Calculate the staffing side. Salary and employer costs for the time required, cover for absence, telephony, training and management.
- Calculate the AI side. Subscription, usage, telephony, connections, your own setup time and the time your team spends on handed-over cases.
- Include what does not go away. An AI agent does not remove every case. Those that need a person remain, so staffing rarely drops to zero.
- Compare the monthly total. And look at the range between a quiet and a busy month, not just the average.
The support cost calculator helps you structure the calculation, and when automating customer service pays off shows when the comparison tends to tip one way or the other. For phone specifically, read AI phone agent vs a hired phone operator.
The cost of what never happens
Part of the benefit does not show on an invoice. Missed calls and unanswered enquiries have a value, but that value is an estimate built on your own assumptions. Missed phone calls and what they can cost shows a simple calculation. Treat the result as a calculation, not as a promise of what you will earn.
Questions to ask the vendor
- What does the plan include, and what requires an upgrade or is paid separately?
- How is usage billed, and what happens when volume rises sharply?
- Which channels and connections are included for our platform, and which do you set up during onboarding?
- What does telephony cost, and who invoices it?
- Are there setup fees or minimum terms?
- What terms apply to cancellation, and can we export our knowledge base?
More questions are in the vendor checklist.
Learn the real cost with a pilot
The safest way to know what it costs is to measure. Run a limited pilot with one type of question, track volume, handovers and follow-up time, and recalculate the cost on that outcome. Then the decision rests on your numbers and not on an example. See how a pilot works.
