Vinay Mimani
Jan 28, 2025
At first glance, Agentforce pricing seems straightforward — $2 per conversation. But there's more to it, and understanding these nuances is crucial for planning your AI automation costs effectively.
You pay using 3 types of credits when you use Agentforce:
Conversations: Every time your agent talks to a customer
Einstein requests: Every time your agent needs to think or generate a response
Data Service Credits: Every time your agent needs to work with data
Think of it like running a call center. You're not just paying for the call (conversation), but also for the agent's ability to think and respond (Einstein requests) and access to knowledge bases and systems (Data Service Credits). Any Agentforce usage will generally consume credits from these 3 wallets (which is visible in SF’s Digital Wallet).
Let's break down each of these components and understand what actually impacts your costs.
First, Conversations
Any time someone in your org has a conversation with a specific Agent and an External user, that’s a conversation.
As of now, there are 3 types of Agents available for a conversation:
ASA (Agentforce Service Agent): Handles customer service interactions
SDR Agent: Manages lead engagement and email communications
Sales Coach Agent: Provides internal coaching and role-play capabilities
The list price is $2 per conversation. But don’t think of it as a single conversation. It’s more than that.
Here’s how a conversation is actually charged:
For example, if your SDR agent sends 40 emails to a single lead, it’s still ONE conversation. Or if your service agent engages with the customer over 20 messages in 2 hours? Still ONE conversation.
Join our webinar on February 19th at 8 PM IST | 6:30 AM PST as we break down Agentforce economics beyond the $2 per-chat price tag. Learn how to unlock the full economic potential of Agentforce in your business.
Register now
Next, Einstein requests
Anytime you are asking AI to help you with a decision, output or insight within Agentforce, you consume Einstein requests.
Summarising a case ticket?
Generating personalised email for a lead?
Analysing customer data to give a tailored real estate recommendation?
All of them need Einstein requests.
However, does each one cost the same? No.
You pay based on 2 variables:
Length (in words) of prompt request and generated response
LLM model being used
Imagine your AE wants to generate a customer proposal.
Your input prompt is 500 words
The generated response is 14,500 words
So, total words: 14,500 + 500 = 15,000
Now, Salesforce measures in units of 1500 words. For 15,000 words, you're using 10 units. This is called your 'call size factor.'
Important: Even one word over these brackets counts as a full unit. For example:
15,000 words = 10 units
15,001 words = 11 units
Additionally, your choice of AI model affects the final Einstein request count:
Using your own model (BYOM): 7x multiplier. Example: 10 units × 7 = 70 Einstein requests
Using Salesforce's model: 10x multiplier. Example: 10 units × 10 = 100 Einstein requests
This means the same proposal could cost you 30% more Einstein requests depending on your model choice.
Finally, Data Service Credits
As you know, Generative AI is most valuable where traditional coding methods fail — “unstructured text” like Product Documentation, Legal Contracts, Candidate Resumes, Product Descriptions, Customer Feedback, etc. — exactly the type of material humans will need to refer to for decision making.
Most impactful agents will almost always refer to such data, and often in large quantities. Such data processing consumes Data Service Credits. You can broadly group them into 3 cost drivers:
Depending on the use case, industry, and exact workflow, these factors will change and need to be optimised.
Why three types of credits?
Agentforce mirrors how a human agent works, which is why the pricing splits into three distinct components. When your human agents work, they spend time talking to customers, thinking through responses, and accessing information from your systems. Agentforce does the same, and each action has its own cost structure.
Conversations are your agent's direct interactions. They measure actual engagement with external users, much like tracking customer touchpoints.
Einstein Requests represent your agent's thinking power. Every time they need to understand context, analyze information, or generate content, they're using computational resources. The more complex the task, the more resources needed – just like how a human agent needs more time for complex problems.
Data Service Credits cover how your agent works with your business information. They need to access knowledge bases, process documents, and analyze customer histories. This is your agent's ability to learn from and use your business data effectively.
This three-part system directly ties to how businesses create value.
Your conversation credits translate to tangible business outcomes – leads qualified, support cases resolved, or coaching sessions completed.
Einstein Requests align with the complexity of your needs. Simple responses cost less than detailed analyses, and you can choose between cost efficiency with your own model or convenience with Salesforce's. You're paying for exactly the processing power you need.
Data Service Credits turn your raw data into actionable insights. This is where your agent transforms documents, emails, and customer histories into valuable business intelligence. The value comes from making better-informed decisions based on your entire data ecosystem.
Understanding these components helps you optimize for what matters in your specific use case. A customer service deployment might prioritize efficient conversations, while a document processing workflow might focus on optimizing Einstein requests and data handling.
Now let’s bring everything together using a real use case.
Let’s say you want a chatbot on your website that can have a 2-way conversation with leads, gather their information, and after lead qualification, send it to Salesforce.
While the exact details will depend on multiple factors, here’s a high-level approach to think about it:
Q: Will my agent be talking to an External user?
Yes. So I will be paying per conversation for my SA (Service Agent).
Q: Will I need help from AI?
Yes, my Service Agent will need to understand and reply in the context of what the lead is saying. This means every time your agent generates a response, you will consume Einstein Requests.
Q: What behind-the-scenes data activities will be needed?
Storing different process documents that capture talk tracks for the agent
Running a query on lead data to qualify based on a ruleset
Sending the details to Salesforce with the required metadata
Depending on the volume and complexity, these three operations will consume Data Service Credits.
Optimizing your Agentforce costs
Enterprise licenses come with substantial starting credits — 1.5M Einstein requests, 250k data credits, and 1TB storage. While these are generous allocations, good optimization ensures you're getting maximum value from each credit.
Optimizing Conversations
Understanding the 24-hour window is crucial. Since all interactions within 24 hours count as one conversation, you can structure your workflows accordingly. For Service Agents, this means maximizing resolution within that window. For SDR Agents, you can batch your lead engagement activities to maximize each conversation credit.
Managing Einstein Requests
Your Einstein requests are calculated based on word count, making prompt design critical. Start by analyzing your current patterns — are your prompts unnecessarily verbose? Could you achieve the same results with more concise inputs? Remember: every 1500 words counts as a unit, so even small reductions add up.
The model choice matters significantly here. While Salesforce's model comes with a 10x multiplier, using your own model (BYOM) reduces this to 7x. For a 15,000-word interaction, this difference means 30 Einstein requests saved. Consider testing both options to find your optimal balance of cost versus convenience.
Data Credit Efficiency
Data credits power your agent's ability to work with information, but they need careful management. Start by organizing your knowledge base efficiently. Every time your agent needs to search through unstructured data, you're consuming credits. Structure your data thoughtfully, and you'll reduce the processing overhead.
Consider these practical steps:
Organize reference materials efficiently to minimize search operations
Batch process similar data operations instead of handling them individually
Structure your data input processes to reduce transformation needs
If you’d like to discuss how Agentforce can help move a business metric and want clarity on costs and expected RoI, please reach out for a free pricing consultation with our team.