A Marketer's Guide to the Trusted Use of Einstein

Our Trusted and Transparent Approach

At Salesforce, we’re committed to developing ethical and inclusive technology. Given the tremendous opportunity and challenges emerging in the generative AI space, Salesforce is building on top of our Trusted AI Principles with a new set of guidelines focused on the responsible development and implementation of generative AI. The five guidelines Salesforce uses to guide the development of trusted generative AI are Accuracy, Safety, Honesty, Empowerment, and Sustainability. As part of our commitment to ethical technology use, our Research & Insights team collaborated with our Marketing Cloud Engagement product teams and the Office of Ethical and Humane Use to publish our Responsible Marketing Principles. The principles document outlines some tools that we developed so that our employees, customers, and partners can use AI more responsibly, accurately, and ethically.

Use Einstein Generative AI with Confidence

At Salesforce, trust is our #1 value. To keep your data secure, Salesforce has agreements in place with large language model (LLM) providers, such as OpenAI. These agreements allow organizations to use generative AI capabilities without their private data being retained by the LLM providers.

In Marketing Cloud Engagement, use Einstein generative AI to safely create subject lines and body copy in both Einstein Copy Insights and Content Builder.

Use Model Cards to Understand How Einstein Models Work

Consumers and marketers want to know how a model behaves and how it arrives at its prediction or recommendation in order to trust the model. Einstein model cards provide critical information about model inputs and outputs, the conditions when models work best, and the ethical considerations in their use. These cards answer questions about using the models as intended, and whether there are any biases to consider. Providing this transparency helps developers, customers, policymakers, and others better understand the impact of our AI on individuals, communities, and society. Check out these model cards for details.

• Einstein Send Time Optimization for Email determines the optimal time to send an email to a subscriber to maximize the probability that the subscriber opens the email.
• Einstein Send Time Optimization for Mobile determines the optimal time to send a mobile message to maximize the probability that a subscriber interacts with the message.
• Einstein Engagement Scores for Mobile evaluates each contact's engagement record to assign scores for the contact's likelihood to interact with mobile messaging.
• Einstein Messaging Insights monitors message engagement rates and notifies marketers when anomalies in their marketing performance occur.
• Einstein Engagement Frequency analyzes each contact's engagement record to determine the optimal frequency at which to send email messages to that contact.
• Einstein Subject Line Tester evaluates contact engagement based on email subject lines.

Check Data Quality Scores to Ensure Your Data is AI-Ready

Use data quality scores as indicators to help you determine how ready your data is for Einstein to make accurate predictions. With any AI model, the quality of the output depends on the quality, representativeness, and robustness of the input. Data quality scores show you when the data is in good or bad shape, why it's in that shape, and general tips for improvement.

• Gauge Your Einstein Engagement Scoring Data Quality
• Gauge Your Einstein Engagement Frequency Data Quality
• Gauge Your Einstein Send Time Optimization Data Quality

Review Bias Flags to Prevent Potential Disparate Impact

Bias flags indicate when selected inputs can introduce bias into the model. For example, Einstein Content Selection uses bias flags to indicate potentially sensitive attributes in profile data. And in the Einstein Subject Line Tester, bias flags show when words or phrases in a sample can be considered sensitive. Bias is contextual and based on industry and use cases. Einstein highlights potential biases so that you can evaluate how to proceed based on the context.

• Identify Attributes That Can Introduce Bias
• Test a Subject Line Before You Send

Call to Action

To optimize your marketing efforts, use these prompts toward deeper thought and discussion. The prompts aren't exhaustive and don’t apply to all organizations, teams, products, or features.

• Are your training and evaluation datasets representative of your customer population? How do you know?
• Have you assessed your datasets and models for bias?
• If you have launch criteria, does your model meet that criteria for various population subgroups?
• Does your model have similar performance for the various population subgroups or does it perform worse for one group versus another?
• Are you using AI to infer sensitive information about your users such as age, gender, health status, sexual orientation, political affiliation, financial status, or criminal convictions?
• Are you collecting and using sensitive data only when necessary and when it can benefit the customer?
• Are you using sensitive data proxies such as a first name or salutation that can relate to gender, or a ZIP code that can relate to race, unnecessarily?
• Are you giving customers control over the data that you collect and do you respect their preferences?
• Are you demonstrating the benefits that customers receive in exchange for data usage?

Summary

The ethical use of AI is an increasingly complex issue to address to ensure that it's safe and inclusive for all. Model cards, data quality scores, and bias flags are a few of the ways that we infuse ethics into our AI practices at Salesforce.

Resources

• Salesforce Responsible AI in Marketing
• Salesforce Responsible Creation of AI Trailhead
• Salesforce AI Ethics Blog
• Salesforce Artificial Intelligence Acceptable Use Policy

Join Us

If you're interested in participating in Salesforce User Research, sign up for our research program and influence product design and direction. If you need more incentive, we almost always provide a thank you gift!

Contact Us

If you have questions about the ethical use of Einstein, contact your Salesforce account executive, or email buildwithintention@salesforce.com. A team monitors this inbox and typically responds within 72 hours.