Understanding Consumption Based Billing Model in Salesforce Einstein

Thoughts on Salesforce Data Cloud

I had written about credit-based pricing in the context of Data cloud earlier and I could not find any clarity on what this meant. I now understand how this translates for Einstein related services at least.

In simple terms AI-based services will bill you based on the complexity of the task AI needs to perform. To be more specific, Salesforce charges you for AI tokens required to provide a response to your request. You can use Tokenizer service provided by OpenAI to understand how many tokens would be consumed by an average request.

So how does this pan out in real life scenarios? Let’s check with a few examples

Request	Tokens
Write an introductory sales email	6
Write an introductory sales email to a new customer	10
Write an introductory sales email to a new customer based out of United States	14
Write an introductory sales email to a new customer based out of United States who uses a competing product	19
Write an introductory sales email to a new customer based out of United States who has never used a similar product before	24
Write an introductory sales email to a new customer based out of United States who has never used a similar product before. Include details about Salesforce Einstein and how it can help a financial services company reduce service times and save money	44

My Thoughts On Salesforce Einstein Generative AI Features

As you can see in the table above, the token usage scales up with every bit of complexity and detailing that is added to the request. Suggesting generic requests will get you a very generic mail, but for better results you will need to use more detailing and this will mean a higher token usage. This in turn would mean higher costs.

From a business point of view this can complicate budgeting and a lot of education to prevent costs from shooting up. I’m not sure what tools will be available for business to manage cost, but an ideal scenario will be an option to set a per license ceiling. This can prevent an overzealous employee from blowing up the all budgeted tokens.

What we don’t know

1. Cost of a token.
2. What tools are available to keep things in check.
3. Do these data tokens in Data Cloud work the same way?