Last week during the Salesforce World Tour in London we sat down with the head of the Salesforce Einstein product team to discuss how they take cutting edge data science techniques and make them consumable for customers of its SaaS platform.
Since Salesforce acquired MetaMind and gave its founder Richard Socher the title of chief scientist, the goal has been to increasingly embed intelligent features into its range of customer relationship management (CRM) software - branded as ‘clouds’ - giving customers in sales, marketing or customer service access to more actionable insights and a system that learns the more they use it.
This is the founding idea behind the Einstein brand, which was announced at Dreamforce in 2016 and has since hogged the limelight when it comes to new features in Salesforce’s three-yearly update releases (some will cost you extra).
The person in charge with taking artificial intelligence and machine learning technology and embedding it within the Salesforce platform is Marco Casalaina, VP of product for Einstein and a Salesforce employee since 2005.
Casalaina’s team works with the product teams on each ‘cloud’ within Salesforce - be it Marketing, Sales, Service or Commerce - to find areas where AI might be put to good use. There are now nearly 30 examples of Einstein features baked directly into the various Salesforce clouds, as well as the MyEinstein platform, which exposes the underlying technology to customers looking to do some more customised deployments.
“It doesn’t really require much setup and they kind of just flow into the application people are using,” is his one-line pitch for Einstein.
Richard Socher is still very much at the forefront of what Einstein does, but his role differs from Caslaina’s in some key ways.
“If you look at some of the research [Socher] and his team are doing as part of the Einstein team, basically his job is to come up with the visionary stuff and my crew’s job is to productise it and bring it to scale,” Caslaina explained. “In the coming year you will see some things that have been research projects come to fruition as products.”
So how do you build trust in the AI?
“A lot of it is comporting with user’s intuition,” he said. “People’s intuition is more often than not correct, especially if they have been doing something for a while so if they see the why, most of it should comport with their intuition.”
For example, the popular Einstein lead scoring feature will assign a ‘score’ to a lead within Sales Cloud according to the algorithms expectation that a sale will close. The key for Casalaina however is to explain why that lead score has been assigned right there within the application.
For example, most salespeople will know a lead from an industry event is better than a web lead, generally. The system may also throw up some more unique insights, but once the system is seen to line up with intuition, the trust starts to build.
One of the latest additions to the Einstein portfolio was MyEinstein, announced at Dreamforce last year as part of a wider company push towards more personalised CRM solutions for customers.
“MyEinstein still represents what Einstein is, which is AI for CRM. So it’s not a general purpose AI toolkit, you’re not going to build a self driving car with it,” Casalaina said.
The idea is to allow customers to build custom AI models on their Salesforce data to help predict a business outcome, like customer churn.
However: “One of the most common things people want to predict is attrition, but there is no standard concept of attrition within Salesforce or in the wider world of business. Your concept of attrition may be wildly different from mine, so to get there we had no choice but to expose the Einstein platform for our users.”
Which customers are using Einstein?
Casalaina is particularly fond of two customers when it comes to explaining the ease at which customers can seemingly adopt Einstein capabilities: US Bank and furniture retailer Room and Board.
Room and Board is a long-time Marketing Cloud customer. It has recently adopted Einstein for more intelligent email segmentation. Say the company wants to target emails at men in California over the age of 35; now within Marketing Cloud customers can add parameters for the likelihood that someone would open the email. Naturally this has helped boost open rates, by as much as 15 percent, according to Casalaina.
He added: “We have a distinct advantage being Salesforce as customers entrust their data to us. So Room and Board have been using Marketing Cloud all along so when they turned Einstein on it did its thing on the data already there and its outputs were fairly seamless.”
However he was keen to reiterate that their model will always stay completely separate from anybody else using Marketing Cloud.
Naturally customers get quite a lot of say when it comes to requesting Salesforce features, be it through upvoting requests on the Ideas page at Salesforce.com, events like World Tour or customer advisory boards.
“In my experience the number of votes on the Idea exchange tends to correlate to what customers anecdotally ask me for,” Casalaina said.
More specifically he said that customers should be able to get excited about new features pretty early on as they are already in pilot.
“These AI things take longer to marinate than your usual feature because, in part, of predictions vs actuals. So if you’re looking to predict attrition three months out the only way you can know if that was a true prediction is to wait three months. So we have a tendency to telegraph our route by putting these things into pilot quite far in advance.”
One feature that is currently in pilot is Einstein prediction builder, “which is built on the MyEinstein platform as well and is aimed at putting predictions right where you are in Salesforce”, he said.