Zero One: Salesforce Einstein AI Getting Smarter

Salesforce’s new customizable myEinstein brings artificial intelligence closer to the action and completes the AI loop for reinforced learning.

Tom Kaneshige, Writer

December 5, 2017

3 Min Read

When is artificial intelligence not artificial intelligence? Answer: When it doesn’t have an AI loop – that is, when it can’t learn from successes and failures because it doesn’t know the outcomes of its decisions.

For business people, one of the most important artificial intelligence technologies is Salesforce’s Einstein, which, among other things, predicts business outcomes and scores leads. Does Einstein have an AI loop? If a lead is scored a certain way, Einstein needs to know what happens to the lead in order to evolve and become better.

Salesforce has set in motion plans to close the AI loop thanks to myEinstein. Unveiled last month at Dreamforce, myEinstein lets admins create a custom AI model inside Salesforce with access to custom data, which includes the outcomes.

By doing so, Einstein is about to get a whole lot smarter.

Not bad for a piece of technology that debuted a little more than a year ago. Given the huge impact artificial intelligence will have on businesses, Salesforce has been focused on Einstein development. Today, there are some 20 Einstein products on the market with at least 10 more in the hopper.

Related: Zero One: Playing the AI Game

Einstein performs all sorts of tasks, from classifying images to structuring data to scoring leads. Sales Cloud Einstein also delivers opportunity insights whereby deep learning technology extracts signals from a salesperson’s email to inform next steps in the sales process, says Marco Casalaina, vice president of products for Salesforce Einstein.

Casalaina says Einstein already has access to historical outcomes and learns from them.

“We do have a notion of predicted versus actual, so we are measuring the performance of our predictions historically,” Casalaina says. “Then these models, they generally retrain themselves, usually monthly. It depends on the specific product, some of them retrain at a slightly higher frequency. But regardless, all of the Einstein products automatically relearn on a periodic basis.”

Not everyone agrees.

Manny Medina, CEO of Outreach, a Salesforce competitor, says Salesforce’s relational database overwrites fields whenever a salesperson updates a sales opportunity. This makes it difficult to tie Einstein actions with outcomes, he says.

“AI runs on closed loops, and you need the loop of the action and the outcome,” Medina says. “If you don’t, then you don’t have anything to train on.”

To Medina’s point, most software products touted as artificial intelligence today don’t actually have an AI loop. They’re basically hard-programmed predictive analytics engines, not learning machines. Marketers label them as artificial intelligence to sound smart and innovative.

With customization, though, myEinstein gets closer to the sales rep action and has direct access to outcome data. It goes beyond guesswork, Medina says. With reinforced learning, myEinstein will become more accurate with predictions and insights over time. Medina says it will take at least a year for the data collection and learning to kick in.

Casalaina also sees myEinstein having a more complete AI loop.

“Salesforce ships with a certain schema, but our customers customize that very heavily and now about 80 percent of the data is custom,” Casaliana says. “So that implies that we couldn’t have been complete without having a customizable artificial intelligence capability, which we now have.”

For myEinstein customers, the idea of Einstein making more accurate predictions about sales leads and providing better actionable insights into the sales process should sound exciting. Einstein can be customized and tuned to your industry, your company, your sales reps.

But keep in mind artificial intelligence is a learning system, and the AI loop is one of trial and error. Both success and failure outcomes are needed. The learning curve can be long and arduous. Adding randomness, for instance, can help the system learn but discourage users.

Can you trust it? That’ll be the next AI challenge.

Tom Kaneshige writes the Zero One blog covering digital transformation, AI, marketing tech and the Internet of Things for line-of-business executives. He is based in Silicon Valley. You can reach him at [email protected].

Read more about:


About the Author(s)

Tom Kaneshige

Writer, Channel Futures

Tom Kaneshige writes the Zero One blog covering digital transformation, AI, marketing tech and the Internet of Things for line-of-business executives. He is based in Silicon Valley. You can reach him at [email protected]


Free Newsletters for the Channel
Register for Your Free Newsletter Now

You May Also Like