Generative pricing: agent pricing evolution
Steven Forth is CEO of Ibbaka. Connect on LinkedIn
If chatbots are the dominant pattern for horizontal applications of generative AI agents are emerging as a theme for B2B and vertical AI.
Most people are aware of Salesforce’s Agentforce initiative and there goal to have one billion agents deployed (and monetized) by the end of 2025. See Will Salesforce’s stance on agent pricing frame AI pricing generally? Salesforce is not alone in the move to agents.
Microsoft is adding agents to Copilot.
ServiceNow has announced AI agents as part of its customer engagement suite.
Workday is introducing AI agents for HR and finance tasks.
HubSpot is positioning AI agents as virtual colleagues.
SAP has expanded its Joule AI copilot to handle more complex tasks, moving in the direction of autonomous agents.
Freshworks is introducing an AI agent to support IT teams and customer service.
And, of course, Intercom and Zendesk are innovating on pricing metrics for agents.
Intercom: AI applications edge toward outcome based pricing
Zendesk: Zendesk First in CX Industry to offer Outcome-Based Pricing for AI Agents
What is an AI Agent?
Agents tend to be more specialized than Chatbots or Copilots. Chatbots, like ChatGPT, need to be able to respond to many different prompts and to generate free form responses and follow ups. Agents are more task based (which will be a key to their pricing) and support a smaller number of business processes or SOPs (Standard Operating Procedures).
AI agents possess several defining features:
Autonomy: They can operate independently without constant human intervention (although, for the time being, they will mostly act with human supervision)
Perception: Agents use sensors to gather information about their environment
Reactivity: They can assess and respond to changes in their surroundings
Decision-making: AI agents analyze data and make decisions to achieve their objectives
Learning: Many agents can improve their performance over time through various learning techniques
Communication: They can interact with other agents or humans using different methods
What other companies are innovating on AI Agents?
A lot of the innovation in agents is happening with smaller companies. Here is a partial list …
Superagent: This open-source platform allows companies to build customized AI agents for tasks such as web research, sales, project management, and marketing. It is particularly favoured for its flexibility and ease of integration into existing workflows.
Aviso: Aviso's AI agents are designed to streamline sales operations by automating complex tasks like pipeline construction and account planning. These agents use Small Language Models (SLMs) and vector databases to enhance operational efficiency, providing features such as Virtual SDRs for personalized customer interactions.
Oracle: Oracle has integrated over 50 AI agents into its Fusion cloud business applications suite, covering areas like finance, HR, supply chain management, and customer service. These agents leverage large language models to perform multi-step processes and adapt to novel scenarios, enhancing productivity across various business functions.
B2B Rocket: Specializing in sales automation, B2B Rocket uses AI agents to automate lead generation, engagement, and meeting scheduling. Their AI-driven approach focuses on personalized interactions and efficient prospect qualification to maximize conversion rates.
Adept AI: Adept AI develops agents that respond to natural-language commands for controlling desktop applications and automating corporate workflows. Their focus is on integrating various software tools to streamline operations.
Moveworks: This company uses agentic AI to automate complex tasks with a focus on IT support and HR workflows. Moveworks' solutions include autonomous goal-setting and execution, supported by partnerships with major platforms like Microsoft.
Lancey: Lancey offers an AI-powered product manager that provides insights product feedback, and helps with operational efficiency by automating repetitive tasks
Fynt AI: Known for its AI agent called Paradigm, Fynt AI automates tasks like lead generation, data collection, and email outreach. It integrates easily with existing tools to streamline workflows.
Mr. Bean: This AI agent is designed for B2B sales and marketing processes, automating tasks such as prospecting and running campaigns across different platforms.
Aomni: Aomni builds AI agents for B2B sales to automate research tasks and sales development workflows, utilizing data from public web sources.
Moveworks: Moveworks leverages agentic AI to automate complex tasks in IT support and HR workflows, enhancing productivity through autonomous goal-setting and execution.
Beam.ai: Specializes in Agentic Process Automation, offering a platform for managing AI agents across business processes with a focus on operational efficiency and integration with existing tools.
NinjaTech AI: Develops an AI assistant platform with agentic capabilities for tasks like coding and advising, integrating multiple external AI models for enhanced performance
Rapid Innovation: Known for its expertise in developing intelligent AI agent systems to optimize and streamline business operations
Databricks: Offers a unified analytics platform that integrates AI and machine learning to analyze large-scale data, build ML applications, and more.
Dialpad: Provides a customer intelligence platform with AI tools for customer engagement, sales intelligence, and team collaboration.
And that is just the beginning of the list!
How will AI Agents be priced?
We can see a clear progression in how AI agents are getting priced. There is a movement from inputs to agents to actions to outcomes.
Generative AI applications companies are justifiably concerned with costs. Costs are based on three things:
the model used
inputs used to generate outputs
outputs
Inputs and outputs are generally measured in tokens, the main way that the size of prompts are measure.
More recently, there are also hidden intermediate inputs/outputs, like the reasoning tokens in OpenAI o1.
Input tokens
Input tokens are the tokens that make up the prompt or message sent to the model. These are the words, characters, and symbols that you provide as input for the model to process and respond
Output tokens
Output tokens, also called completion tokens, are the tokens generated by the model in response to the input. These are the visible tokens that form the model's response or completion
Reasoning tokens
Thinking tokens, also referred to as reasoning tokens, are a new concept introduced with OpenAI's o1 series models. These tokens are used internally by the model to "think" or reason about the prompt, breaking down its understanding and considering multiple approaches before generating a response
Per Agent
Another way to price AI agents is per agent. Large companies may deploy fleets of agents and one pricing metric could be per agent, which is in some ways analogous to paying per server. Just as one can quickly spin servers up and down depending on demand (think Amazon Elastic Cloud), one will be able to add and delete agents depending on demand and this will shape pricing. On demand, reserved and spot agents will all be part of the agent pricing pattern.
Task based pricing of AI agents
Another way to think about AI agent pricing is in terms of the task being performed. This is a natural approach for agents which tend to be built for specific purposes, like customer support, lead prospecting, manufacturing control, supply chain management and optimization, and so on. Task-based pricing of AI agents will have three sub stages.
In some pricing models we have seen the price per agent is based on the complexity of the task addressed. Ee believe this is a pricing stage that will fade away as people come to get a better understanding of task value (which will require task level value models). The goal will be to get to a price for task completion, where there is only payment for and on task completion. This is already happening for agents where there are clear completion criteria and the value of task completion is clear.
Outcome based pricing for AI Agents
The goal will be to get to outcome or task completion based pricing wherever possible. This will lead to higher prices. Every price, especially every value based price, has an implicit risk discount. Buyers who are unsure that they will receive the promised value need a discount to reduce this performance risk. Removing that risk leads to higher willingness to pay.
We are already seeing outcome based pricing with AI agents, especially in areas like customer support (pay per issue resolution) and AI SDRs (pay per lead generated, or per meeting that takes place).
There are two criteria for moving to outcome based pricing: clarity on the value of the outcome and the ability to attribute the results.
Clarity: Outcome based pricing requires clarity on what the outcome is and if it has been achived.
Attribution: Many outcomes, especially the most significant ones, are the combination of multiple causes. This can make attribution based pricing more difficult. There are emerging ways to address this, leveraging approaches from causal machine learning and evidence based medicine, but these techniques are not yet widely accepted in the business world.
Conclusions
AI agents are emerging as a significant trend in B2B and vertical AI applications, with major companies like Salesforce, Microsoft, and SAP incorporating them into their offerings.•
Unlike general-purpose chatbots, AI agents are more specialized and task-oriented, designed to support specific business processes and standard operating procedures.•
The pricing models for AI agents are evolving rapidly, moving from simple cost-based approaches to more sophisticated task-based and outcome-based pricing strategies.•
Outcome-based pricing for AI agents is becoming increasingly popular, especially in areas like customer support and sales development, where clear completion criteria and value metrics exist.
Challenges in implementing outcome-based pricing include the need for clarity on the value of outcomes and the ability to accurately attribute results, particularly for complex business processes with multiple contributing factors.
More on Generative Pricing
Book a demo today and discover how our unique B2B SaaS solution can increase sales, improve retention, and drive growth.