Your CFO wants one number for contact center AI pricing. Google, Amazon and Microsoft each bill by a different unit, and specialist vendors use a fourth. Below, a pricing-model table and three total-cost examples put every quote on one monthly basis.
The pricing models side by side
| Vendor | Billing unit | Usually extra | Rates | Best fit |
|---|---|---|---|---|
| Google (Conversational Agents, Dialogflow CX) | Per request on chat, per second of audio on voice | Telephony, Agent Assist, Conversational Insights | Google pricing page | Bot-led volume that swings |
| Amazon Connect | Pay as you go, per voice minute and per chat message | Telephony, number rental, analytics, knowledge assistant, bots | Amazon pricing page | Spiky volume, no seat fee |
| Microsoft Dynamics 365 Contact Center | Per user per month | Copilot and other AI add-ons, speech usage | Microsoft pricing page | Steady volume, busy agents |
| Specialist voice-agent vendors | Per minute of call time | Set per vendor, so ask for the all-in rate | Each vendor's own page | Pilots and fast launches |
| Specialist analysis vendors (such as Insight7) | Per call or transcript analysed, in plans | Overage past the plan cap | Plan table below | Scoring every call |
Rates change often. Copy each number from the vendor's own page on the day you build your model. If a vendor gives no all-in rate, mark that cell "not published" until a written quote arrives.
Compare the billing unit first
The unit decides how your invoice behaves when volume moves.
Google's conversational agent product bills by use. Conversational Agents, which includes Dialogflow CX, charges for counted requests on chat agents and seconds of audio on voice agents. Voice is priced per second, so multiply the listed rate by 60 to get a per-minute figure.
Amazon Connect is pay as you go. You pay per minute of voice and per message on chat, with no seat fee. Microsoft Dynamics 365 Contact Center is priced per user per month. Take the exact figure from Microsoft's page.
- Usage billing: monthly cost is minutes times rate. If average handle time grows 10%, your bill grows about 10%. A seasonal spike raises it too.
- Seat billing: monthly cost is agents times price. A busy month costs the same as a quiet one, and idle agents still cost you.
Pick the model that matches how steady your volume is.
Add-ons and meters change the total
The headline rate is rarely the full bill. Budget your carrier or SIP trunk separately from Google's agent. Agent Assist and Conversational Insights have their own Google pricing pages, so add their unit charges and any seat licence before you compare.
On Amazon, telephony is a separate line and number rental can add a daily charge. Analytics, the knowledge assistant and bots may have their own meters. Read every meter on Amazon's Connect pricing page and put each one in your model.
On Microsoft, ask whether Copilot and other AI features are add-ons and how speech is metered. Ask what licence prerequisites apply, such as Teams or Dynamics.
Google also bills voice on the audio the agent hears and speaks, so a call can bill fewer minutes than its clock time. Check the page for how each turn is rounded. Data store storage has its own line, with a free allowance before a per-GiB monthly charge.
Specialists charge per minute or per analysis
Specialists bill in two units, and each one tracks something different.
Per-minute voice-agent vendors charge for call time. Ask each one for the all-in rate at your handle time and your number of simultaneous calls, including model, voice and telephony fees. Buyers ask for this early and want it in writing, so they can compare vendors and brief their own teams. One buyer wanted pricing before any pilot because per-minute cost scaled in ways they could not predict, and they did not want to commit and then switch tools. Our advice is to model each quote at double your current volume, and put per-minute and subscription quotes on the same monthly basis.
Per-analysis plans charge for each call or transcript scored. Insight7 sells this way, and it scores 100% of calls against a weighted scorecard your team owns. Cost inside a plan stays flat until you reach its analysis cap, and longer calls cost the same as short ones.
| Plan | Price, billed annually | Users | Call analyses |
|---|---|---|---|
| Pro | $99/month | 1 | 50 |
| Business | $299/month | 3 | 200 |
| Plus | $1,499/month | 20 | 2,500 |
| Enterprise | Custom | Unlimited | Unlimited |
Three total-cost examples on one monthly basis
Run one volume through each model, and the quotes become comparable. The seat price and per-minute rate below are placeholders. Cloud rates change often, so swap in the figures from the linked pages above. The Insight7 plan prices come from its published plans.
Example A: a 50-seat team on seat licences. The formula is seats times price. With a placeholder of $100 per user a month, 50 seats cost $5,000. That figure is the same in a quiet month and a busy one. Add-ons for AI features and speech come on top.
Example B: 100,000 minutes on a usage platform. The formula is minutes times rate. With a placeholder all-in $0.05 a minute, 100,000 minutes cost $5,000. Each extra cent per minute from your carrier adds $1,000. If handle time grows 10%, the bill rises to $5,500. A 200-agent team on the seat placeholder above would pay $20,000. Check what each price includes before you call one cheaper.
Example C: scoring every call on a per-analysis plan. The formula is plan price divided by analyses. The Plus plan is $1,499 a month for 2,500 analyses, or about $0.60 a call. Manual QA reviews only a 2 to 5% sample. Break-even works like this: extra review hours equal (all calls minus sampled calls) times minutes per review, divided by 60. The plan costs less than manual review once your loaded QA hourly rate is above $1,499 divided by those extra hours. Use your own review time. At 15 minutes a review and a 5% sample, that is 594 extra hours and about $2.52 an hour. Treat those two inputs as an example.
Hidden costs to add to any quote:
- Telephony and number rental
- Speech-to-text and LLM usage
- CRM and dialer integrations
- Implementation and training
- QA headcount
- Overage past a plan cap
What moves the number: handle time, call volume, simultaneous calls, the share of calls the AI contains, and contract term.
Measure before you commit. Only a quarter of 432 AI customer-service use cases produce a return, and 42% have unclear ROI. So set a baseline cost per contact first, or you cannot tell whether your bill buys anything. Customer-service AI spending rose 38% while overall service budgets rose 2%. That means your AI line will be questioned, so keep a one-page model ready.
When each option fits
- Google: your bots handle most contacts and volume swings. Pay for what the bot uses.
- Amazon Connect: volume is spiky and you want no seat fee. Watch the many meters.
- Microsoft: volume is steady and agents are busy most of the day. Seats stay predictable.
- Per-minute voice-agent vendors: you want a fast pilot. Get the all-in rate in writing.
- Per-analysis vendors: your goal is scoring every call, not handling it. Cost is flat inside the plan.
Related reading: AI Call Center Companies Buyer's Guide With Evaluation Criteria and a Vendor Shortlist Template, How to Scale Customer Support to 10x More Customers Without Hiring More Agents and AI Voice Agent Implementation Timeline With a Week-by-Week Plan From Demo to Launch.
Questions people also ask
How much does contact center AI cost per agent per month?
Seat-priced platforms such as Microsoft's list a fixed fee per user, before AI add-ons. Check it on Microsoft's pricing page. Usage platforms have no seat fee, so a per-agent figure depends on minutes handled.
How much does Amazon Connect cost per minute?
Voice is billed per minute, and the current rate sits on Amazon's pricing page. Telephony, analytics, the knowledge assistant and bots may be separate meters, so add each one before you compare vendors.
Is Google Contact Center AI priced per seat or per usage?
Usage. Dialogflow CX bills per request on chat and per second of audio on voice. Cost tracks conversations the bot handles, not headcount. Agent Assist and Conversational Insights have separate pricing pages.
Is Microsoft Dynamics 365 Contact Center cheaper than Amazon Connect?
It depends on talk time per agent. Divide the seat price by the per-minute rate to find the break-even minutes per seat, before telephony. Agents who talk more than that number favour the seat licence.
What hidden costs come with AI contact center solutions?
The biggest are telephony, speech-to-text, LLM usage, integrations and implementation. Also ask about minimum commitments, concurrency caps, overage rates and how each vendor rounds billed time. Google, for example, bills voice by turn.
Should I choose per-minute or per-seat pricing for call center AI?
Choose per-minute if volume swings or you are piloting. Choose per-seat if volume is steady and agents are busy most of the day. Model each quote at your current volume and at double.
How do I estimate the ROI of AI contact center solutions before a pilot?
Set a baseline cost per contact, assume a containment rate, and subtract the full vendor cost. Then ask for a short pilot on a small sample, paid on measured lift.
Pick the unit before the price
The billing unit decides your cost more than any list price. This week, pull last month's call minutes and average handle time, then run them through each vendor's unit price to get one monthly number. Insight7 scores every call on your own scorecard.
