Companies are starting to pay for what the agent achieves (closed sales, resolved inquiries, recovered opportunities) and not for licenses, messages, or hours. Here is how the market is changing and how to calculate the return before you sign.
Outcome-driven AI agents change the question companies ask when they buy conversational technology. What matters is no longer how many licenses, messages, or integrations the contract includes, but how many sales the agent closes, how many cases it resolves, and how many customers it wins back, and at what cost. The market is already moving in that direction: Sierra, the customer service AI agent company founded by Bret Taylor and Clay Bavor, reported in October that it surpassed $400 million in annual recurring revenue, with more than $150 million added in the third quarter alone, and frames its pricing as paying for work done well, not for access to software.
The context explains the shift. A 2025 preliminary report from MIT NANDA concluded that 95% of the organizations surveyed saw no measurable return on their generative AI investments, and that only 5% of custom enterprise tools made it to production. The sample is limited and the results are self-reported, but the signal for anyone managing a budget is clear: buying technology without tying it to an outcome is the most expensive way to experiment.
What is an outcome-driven AI agent
An outcome-driven AI agent is a conversational system that completes a process end to end, such as qualifying a lead, quoting, or resolving a complaint, connected to the company’s systems and evaluated by the outcome it delivers, not by the licenses or messages it consumes.
From chatbot to AI agent: from answering questions to resolving processes
The traditional chatbot works with menus and preset answers. It handles a frequently asked question, but when the customer needs to get a quote, change an order, or complete a purchase, it hands off to a person or a form, and that is where the sale is lost. The AI agent changes the scope: it understands the request in natural language, checks the catalog, the inventory, or the insurer’s system, and carries out the next step in the same channel. In the auto insurance sales flow that ChatCenter designs, a text or voice agent qualifies the lead, collects the vehicle and driver data, generates the quote, answers objections about coverage, sends the payment link, and stays with the customer until the policy is issued. In eCommerce, the same principle prepares the cart and completes the purchase inside WhatsApp. Unlike a form that captures data and waits for a call, the conversation moves forward at the moment of highest interest.
Outcome-based pricing versus licenses, usage, and hours
The pricing model defines who bears the risk. A license is paid even if the agent does not sell. Usage-based pricing, whether by messages, conversations, or tokens, is paid even if the conversation resolves nothing: a long interaction that does not solve the problem consumes more and is worth less. Classic BPO hours are paid for capacity, not for outcomes.
In outcome-based pricing, the provider bills when something measurable happens, such as a completed sale or an inquiry resolved without human intervention, and so it has an incentive to also take care of integration, training, and continuous improvement. That logic is the one behind the AINS model explained in What Is an AINS and How It Differs from BPO.
| Model | What you pay for | Who bears the risk | When it makes sense |
|---|---|---|---|
| License | Access per user or platform | The customer | Internal tools with predictable use |
| Usage | Messages, conversations, or tokens | The customer | Variable volumes with no defined outcome |
| Hours (classic BPO) | Human agent capacity | The customer | Processes that are not well standardized |
| Outcomes | Completed sale, resolved inquiry, or recovered opportunity | Shared, with more weight on the provider | Processes with a clear, measurable outcome |
The model has limits worth examining. In a July 2026 analysis on services, Gartner notes that adoption of outcome-based pricing is uneven because of problems with measurement, value, and accountability. That is why the contract must define precisely what counts as an outcome and who audits it. ChatCenter’s commercial model is built on that logic: it is based on completed sales or chats initiated.
How to measure the ROI of AI agents
Return is calculated with four families of metrics, all tied to a number the decision-maker already tracks. In conversion, the percentage of conversations that end in a sale, compared with the previous channel. In sales, the revenue generated per conversation and the revenue recovered from carts and leads that stopped responding. In operating costs, the cost per case resolved without human intervention, which comes from dividing the total cost of the service by the cases the agent closed, along with first response time. In resolution, the percentage of inquiries closed without a handoff and the reopen rate, because a case that comes back was not resolved.
These are ChatCenter’s figures by type of process: the average end-to-end sales conversion is 18%, click-to-WhatsApp campaigns convert at 7%, and abandonment recovery reaches 25%. Measuring the baseline before launch makes it possible to know how much of that difference belongs to the agent.
AI, human support, and operations: the three pieces of the outcome
In outcome-driven AI agents, the outcome rarely comes from AI alone. It comes from three coordinated pieces. AI responds in seconds, around the clock, and sustains campaign and peak volumes. People step in where the customer needs judgment, negotiation, or trust. Operations connects everything: integrations with the catalog, payments, and CRM, a knowledge base, tags to analyze each chat, and daily reports to optimize.
This design answers the usual objections. “AI doesn’t sell” is answered with a complete sales flow and a handoff to a person defined by rules. “I lose control” is answered with metrics by conversation and by tag, visible every day. “It’s expensive” is answered with pricing tied to what the agent achieves. “It doesn’t scale” is answered with human agents sized on demand for peaks.
Use cases: sales, customer service, and recovery
In sales, Santander optimized its conversational channel for its Autocompara insurance line with a hybrid model of AI and human support and, with inbound and outbound channels, doubled its sales capacity and sustained its conversion rate.
In recovery, ROUGE sent 29,819 automated abandoned-cart messages in 2024, recovered 11% with 3,261 sales, and reached 15% during Hot Sale, with a 100% AI setup. In customer service, the deciding indicator is the percentage of cases the agent closes without a handoff, and it is best to agree on it from the start, along with the reopen rate.
ChatCenter: a partner that answers for the outcome
ChatCenter is a Meta Business Partner and designs, implements, and manages outcome-driven AI agents by chat and voice, in a 100% AI or hybrid model, integrated with eCommerce and the client’s systems. With more than 10 million chats managed, its implementation timeline runs from connecting with Meta and the systems through training, automations, and measurement, with launch between weeks 9 and 12. Because its commercial model is built on completed sales or chats initiated, its revenue is tied to what the channel produces. That is how you evaluate a strategic partner, not a technology vendor.
What to do differently starting tomorrow
To evaluate outcome-driven AI agents, three decisions organize the purchase:
- First, choose a process with a measurable outcome and an existing budget, such as quotes, cart recovery, or the most repetitive complaints, and measure its current baseline: conversion, cost per case, and response time.
- Second, ask any provider to define in the contract what counts as an outcome, who audits it, and what happens when the agent hands off to a person.
- Third, validate with a limited pilot, with a numeric target set in advance, before expanding the scope.
Discover how to turn WhatsApp into your main sales channel with AI-powered end-to-end automation.