The marketing and advertising industries are being influenced by consumer demands for greater privacy and data control, as well as impending artificial intelligence (AI) regulation. Kathy Baxter shares how Salesforce builds trusted AI into marketing automation software. More from Kathy here.
We are in a crisis of trust. According to Salesforce’s State of the Connected Customer, 92% of consumers believe there is a moderate to major need for companies to improve their trustworthiness, and 61% reported that it’s difficult for a company to earn their trust.
The growing use of artificial intelligence (AI) in marketing makes things even more complicated. Fewer than half (48%) of customers trust companies to use AI ethically, and 65% are concerned about unethical use of AI. This may be because we see more negative headlines about biased AI (as in facial recognition, healthcare recommendations, or hiring) than positive ones about AI (for example, identifying cancer tumors that doctors miss).
Optics aside, the burden is on marketers to respect privacy when personalising marketing messages with AI. There’s a fine line between useful, targeted ads and ones that are downright invasive. It’s also easy to slip into stereotype bias – attribution of particular qualities to a member of a particular social group – or exclude segments of your customer base by focusing on demographic variables.
So, how do you ensure your marketing is accurate, inclusive, valuable, and privacy-preserving? AI can only be trusted when it is built transparently and has protection for human rights, explainability, accountability, and inclusion at its core. (Salesforce infuses ethical practices and processes into our AI development.)
Next, it is up to marketers to make sure the technology benefits customers. Last month, Emily Witt, Sarah Flamion, and Annie Chin of Salesforce shared four principles for responsible marketing with data:
Let’s apply these principles in the context of AI in marketing.
Einstein Content Selection scans the fields in your models and highlights any sensitive fields (such as age, race, gender) or their proxies (ZIP code, name). Using these variables can add stereotype bias into your models. By highlighting these fields and proxies, customers can make an informed decision about whether or not they should include these fields in their decision making.
Our Lookalikes model excludes age, race, gender, and income level to mitigate that type of bias. Relying more on behavior or interest-based variables rather than solely on demographics is less biased and more inclusive. You’re likely to discover new prospects you didn’t even know might be interested in your products or services!
This refers to being open and clear about how you developed your AI models. What training data did you use? Was it representative of all your customer base or only part of it? What are the variables in your model, and did you weigh or exclude any? We publish model cards for our global models so all of our customers can understand how we created them.
The State of the Connected Customer report found that 61% say they feel like they’ve lost control over how their personal information is used. No wonder there is a crisis of trust. Now is the time to give your customers control over their data.
Third-party cookies are on their way out and digital marketers need to start preparing now. One way to do this is by collecting first-party data directly from customers. In addition to being more accurate than third-party data brokers, it’s also cheaper because you cut out the middleman. Nederlandse Publieke Omroep (the Netherlands version of the BBC) has moved from cookie-based ad sales to context-based ad sales and seen their profits grow by 62%.
Using features like Einstein Content Selection and Copy Insights helps personalise message and tone. Einstein Send Time Optimisation and Einstein Engagement Frequency ensure messages arrive at the right time and not so often that they overwhelm your customers. Together, these features ensure you send the right message to the right customer at the right time.
Putting these four principles into practice will help customers trust your company with their data and increase loyalty and sales, because your customers see the value you provide.
Want to learn about removing bias from your data and algorithms? Take the Responsible Creation of Artificial Intelligence trail on Trailhead.
This post originally appeared on the U.S.-version of the Salesforce blog.