Salesforce Extends Einstein Machine Learning Features for Developers

By:  David Needle is kicking off its Trailhead developers’ conference June 28 with the release of Einstein Platform Services, a set of new tools designed to help users better understand and anticipate customer demand.

The company is also releasing a beta version of Einstein Object Detection, another developer tool that helps with the creation of apps that can recognize images and also object types and quantities within an image.

A retailer could use Einstein Object Detection to analyze what products have sold or are missing on store shelves. It could also help field service technicians with repairs by recognizing a component or part and pulling up a reference to it from a related database.

Salesforce released Einstein, an integrated set of AI technologies, with much fanfare at its Dreamforce conference last October and has been moving quickly to establish it as a core component of its CRM platform.

Jim Sinai, vice president of product marketing for Salesforce Einstein, said it’s part of a broader trend to leverage AI technologies such as machine learning to improve the effectiveness of enterprise apps, noting an IDC forecast that 80 percent of all apps will have an AI component by 2020.

Applications that use machine learning can acquire knowledge based on new data without being programmed. “If you’re a developer in Silicon Valley trying to raise money and machine learning isn’t a part of your stack you aren’t likely to get the time of today from investors,” Sinai told eWEEK. Machine learning “really is what makes a great app,” he said.

The other two key components Salesforce is making available in beta versions to developers are Einstein Sentiment and Einstein Intent.

Einstein Sentiment lets developers classify the “tone” of any text as positive, negative or neutral including emails from prospective customers, social media posts, customer reviews and message boards.

In a demonstration, Salesforce showed how developers can use Sentiment to recognize certain terms and phrases that customers have or are likely to use. For example, someone might say “I love this demo” in an online comment and that would be classified as positive.

But it also understands more nuanced comments recognizing, for example, the comment “This demo is not likely to work” as negative even though “likely” in isolation might be viewed positively. “Deep learning lets you train the system how to understand unstructured phrases. We didn’t train it to understand ‘not likely’, it learned what it means,” said Sinai.

With Einstein Intent developers can classify the underlying content of customer inquiries and have the system automatically route leads, escalate service cases and personalize marketing campaigns.

For example, a retail company could use Einstein Intent to create a custom app that automatically classifies inbound customer support queries, identify customers experiencing shipping problems and then proactively provide support messaging and tracking details.

An Einstein Intent app might help a mobile products vendor see there are shipping problems from comments such as “When will my phone arrive?”

“It doesn’t have to be word for word for Intent to understand something,” said Sinai. “We didn’t train it to understand ‘phone’ but it understands shipping.”

Salesforce’s second annual TrailheaDX developer conference in San Francisco runs June 28-29. Salesforce said that more than 4 million people have received training on how to use Salesforce using its free and self-directed Trailhead training system.