Buying contact center technology used to be relatively simple.
Enterprises paid per agent, per month, added a few premium capabilities, and had a reasonable idea of what their bill would look like at the end of the quarter.
AI has disrupted that model.
Today, a customer service AI agent may be priced by the minute, message, conversation, action, token, credit, or successful resolution.
In some cases, it is bundled into a human-agent license. In others, it sits on top of that license as a separate consumption charge.
As well as deciding which AI is the right fit for their company, contact center leaders must now also work out how to pay for it without losing control of the budget.
“Transparent consumption pricing is an absolute must,” said Zeus Kerravala, Founder and Principal Analyst at ZK Research.
“You need to be able to provide customers clarity that if I run these types of transactions, here’s the cost. Perhaps if you do it this way, it’s going to be cheaper.”
That concern is helping reshape the AI CX market. The old agent seat is not disappearing, but it is becoming less useful as the single unit of value.
A human agent may handle dozens of interactions a day. An AI agent could handle thousands, perform back-office tasks, and trigger several separate billable events before the customer journey is complete.
So, what should buyers look for?
The Old Unit of Purchase Is Breaking Down
Salesforce’s recent Agentforce pricing changes offer the clearest example of how quickly the market is shifting.
The CRM giant initially introduced Agentforce at $2 per conversation. That made sense for a customer-facing AI agent handling a support query. But it was less useful for an agent completing internal tasks, updating a CRM record, generating a quote, or supporting an employee workflow.
Salesforce then added Flex Credits, priced at $500 per 100,000 credits. A standard Agentforce action consumes 20 Flex Credits, which Salesforce values at around $0.10.
The vendor still offers conversation-based pricing, but now also gives customers pay-as-you-go, pre-commit, and pre-purchase payment options.
“We’ve been listening to our customers and developing a new framework called action-oriented pricing,” said Bill Patterson, EVP of Corporate Strategy at Salesforce.
“This model doesn’t charge for small talk or general fluff like: ‘How was your day?’ It charges based on the actual work performed by the agent, what we call ‘taking action’.”
That shift shows that even vendors are struggling to settle on one fair and easily understood way to price AI.
The result is a market where CX buyers are often comparing products with entirely different units of measurement.
The Six AI Pricing Models CX Leaders Will Encounter
1. Per-Seat Pricing and Bundled AI Entitlement
Per-seat pricing remains familiar territory for most contact centers.
Genesys Cloud CX, for example, sells named-user packages ranging from $75 to $240 per user, per month on an annual commitment.
Its CX 4 plan includes 30 AI Experience Tokens per named agent each month, while all plans include a smaller organization-wide token allowance.
Microsoft is taking a similar hybrid route. Premium Dynamics 365 SKUs include 1,000 Copilot Credits per user, per month, pooled at tenant level.
This model works best for contact centers with stable agent populations and mature AI use cases. It gives finance teams a clear baseline cost and avoids sudden spikes in spend.
However, it can become inefficient when AI takes on a larger share of the workload. An enterprise may end up paying for human-seat capacity that it no longer needs, while still facing extra consumption charges for AI activity.
2. Channel-Based Consumption Pricing
Amazon Connect is the strongest example of a consumption-first contact center platform.
It does not require seat-based licenses or long-term minimum commitments. Instead, customers pay according to the channel they use.
Amazon Connect currently charges $0.038 per voice minute, $0.010 per chat message, $0.080 per email, and $0.014 per SMS or third-party messaging interaction.
AWS includes a broad set of AI capabilities within those platform rates, including agentic self-service, agent assistance, conversational analytics, quality management, forecasting, and post-contact summaries.
Kerravala believes Amazon has been the most disruptive vendor in redefining the economics of contact center technology:
“Historically, this has been a market that has lived by per seat per month pricing. For any kind of seasonal business that’s not optimal.”
For a retailer, Kerravala’s point is obvious. Interaction volumes between Thanksgiving and Christmas may outstrip the rest of the year. Paying for peak agent capacity across every month is difficult to justify. Consumption pricing gives these businesses a more flexible option.
Yet the bill can still become complicated. Voice, chat, email, and messaging all have separate meters. Telephony and third-party charges may also apply.
3. Component-Level Usage Pricing
Google Cloud takes usage-based pricing even further.
Its Conversational Agents product charges for chat requests and voice audio duration. Agent Assist chat is priced per message. Conversational Insights offers message- and minute-based pricing, alongside named and concurrent agent options.
For organizations already building on Google Cloud, this can provide considerable flexibility. Teams can choose only the components they need and scale usage gradually.
But it also creates a comparison problem.
Google is not offering one simple ‘AI contact center’ price. It is offering a portfolio of separately metered tools. That may suit technically mature enterprises, but it can make procurement far more difficult for buyers trying to forecast total cost of ownership.
4. Credits and Tokens
Credits and tokens are now the preferred middle ground for several major vendors.
Salesforce Flex Credits can be used across a range of Agentforce actions. Genesys AI Experience Tokens can support capabilities such as virtual agents, predictive routing, summaries, and agent copilot tools. Microsoft uses Copilot Credits as a common currency across its AI services.
Indeed, in discussing this model, Arpita Maity, Director of Product Marketing AI at Genesys, said:
“By paying only for the AI functionalities you actually use, tokenization offers a scalable, cost-efficient way to integrate AI into your operations.”
The upside is flexibility. A contact center can start small, activate new features, and shift usage across different AI capabilities without buying a separate license for every tool.
The downside is that a token is not a customer outcome; it is a vendor-defined measure of activity. Before committing, buyers need to understand how many tokens a voicebot session, AI summary, translation, routing decision, or agent-assist interaction will consume.
Without that detail, credits can quickly become a translation problem for finance teams.
5. Action-Based Pricing
Action-based pricing is an emerging option for enterprises moving beyond basic self-service.
Rather than charging for a conversation, Salesforce’s Flex Credit model aims to charge for the actual work an AI agent completes. That could include updating a customer record, generating a quote, retrieving information, or progressing a workflow.
This approach is particularly relevant for organizations using AI agents across service, sales, and internal operations.
A conversation may be brief but commercially valuable. Alternatively, a long customer interaction may involve plenty of discussion but no meaningful action.
The risk is that complex workflows can contain multiple actions. Buyers should ask vendors to model end-to-end journeys, not just quote the price of one action in isolation.
6. Outcome-Based or Per-Resolution Pricing
Zendesk offers the clearest example of outcome-based AI pricing.
Its AI agents are included in Support and Suite plans, with a limited number of automated resolutions included per agent each month. Additional automated resolutions cost $1.50 on a committed basis or $2.00 through pay-as-you-go.
Zendesk defines an automated resolution as a customer request successfully resolved by AI without escalation to a human agent.
For high-volume, repeatable customer service journeys, that is an attractive proposition. Buyers pay when the technology produces the result they actually want.
Kerravala has already seen one customer push this logic further:
“I talked to one company, an NBA team, that they’ve negotiated: I’m only going to pay the vendor for agents that complete the tasks.”
However, resolution is not always as straightforward as it sounds.
A customer may appear satisfied in the moment, only to recontact later because the issue was not properly fixed. The AI may avoid escalation when a human agent would have been the better option.
That is why resolution pricing must be judged alongside recontact rates, customer satisfaction, complaints, refund leakage, and escalation volumes.
Which AI Pricing Model Fits Your Contact Center?
Unsurprisingly, there is no universal winner.
Organizations with stable agent populations and predictable demand may prefer a per-seat or hybrid model, particularly if they want firm budget control.
Businesses with sharp seasonal peaks may benefit from Amazon Connect-style channel consumption pricing.
Teams running an early AI pilot should usually avoid large upfront commitments. Pay-as-you-go pricing can be useful here, provided spending alerts and hard caps are in place.
Enterprises rolling out multiple AI tools across self-service, agent assistance, analytics, quality management, and workflow automation should assess pooled credits or tokens. But they should insist on a usage simulator based on their own historical data.
Outcome-based pricing can be compelling for repeatable, low-risk service requests. It is less suited to complex cases where a successful resolution is subjective, delayed, or dependent on a human employee completing part of the journey.
The Costs Hidden Behind the Headline Rate
The published AI price is rarely the full cost.
Contact center leaders also need to account for telephony, messaging, model usage, knowledge-base preparation, integrations, workflow orchestration, security, compliance, testing, observability, human quality assurance, and the cost of handling failed AI interactions.
These costs are particularly important for agentic AI deployments. An agent may complete more work, but it also requires stronger governance, monitoring, and controls.
Kerravala compared the risk to the early days of cloud adoption:
“People love the flexibility of cloud. So they loaded up all the data in the cloud and they got these massive bills in the back end. We don’t want to do that with AI.”
He recommends “building the observability in the front end to allow companies to regulate what their users are doing and understand the cost of it.”
A cheap per-message rate is not cheap if the agent creates repeat contacts. A low per-resolution price is not valuable if the AI incorrectly processes a refund or damages customer trust.
Five Questions Every CX Buyer Should Ask
Before signing an AI CX contract, procurement teams should ask:
- What exactly triggers a billable event?
- Which capabilities are included, and which create additional usage charges?
- Can the vendor model project costs using our historical interaction data?
- What happens when we exceed our committed usage?
- How will we prove that an AI interaction was genuinely resolved, rather than merely deflected?
The final point is arguably the most important.
CX leaders should not buy solely against the vendor’s preferred meter, whether that is a seat, minute, message, credit, token, action, or resolution. They should buy against the cost per verified resolution.
That means measuring whether the customer’s issue was actually solved, whether they came back, whether the outcome was compliant, and whether the AI improved the experience rather than simply moving work elsewhere.
The vendors may continue changing the unit of purchase. CX buyers should keep their focus on the unit that matters: a customer problem resolved properly.
