So What Makes Einstein Different?

Data in Salesforce

It all starts with the data you collect. The data you’ve been putting into Salesforce since the first time you logged in. But, Einstein also brings in email, calendar, social, IoT, and external data. This data becomes the fuel needed to train our AI models. Sounds complicated? It’s not. Because the data already lives in Salesforce, you don’t have to do anything to it. It’s already structured so that Einstein can start learning.

Tailored Predictions

But every business’s data in Salesforce is different. In fact, 80% of all records in Salesforce are custom objects. Salesforce customers each have their unique data conventions, meaning there is a trove of custom objects housed on the platform. For us to deliver AI to each customer with different use cases and data, we’d need an army of data scientists. And if we didn’t and you wanted to build AI into your business’s customer data, you’d need your own data scientists. Instead, we at Salesforce have created something pretty special under the hood of Einstein that can scale to all of our customers, across all of your use cases, so neither of us needs armies of data scientists. We call it automated machine learning (AutoML).

Let’s dive into an example on how AutoML works.

Say your commerce team wants to predict how likely it is a customer will buy an item. In order to predict anything, your team needs a list of customers who bought that item, customers who didn’t, and all of the attributes surrounding those customers - like age, location, other items purchased, and so on. This data can help your commerce team understand which factors were the most significant in leading to purchasing that item. But, let’s be honest. There might be duplicate entries with a few fields not completed in one of them, field usage may change over time, and the fields may not be completely standardized in general across all entries. AutoML’s data cleansing sifts through your data and detects these errors and either automatically fixes them or flags them to be fixed.

Once the data has been cleaned, your data needs to be trained so that a predictive model can be created. Before you train your data, you must identify attributes, or features, that are significant to predicting propensity to buy that item. Examples include “length of time being a customer,” “customer address,” or “last purchased item.” But we know you have a massive amount of data, and we don’t want you to sift through it all. AutoML also includes feature engineering, which automatically combs through your data and begins identifying the most significant features for buying the item, so you don’t have to. As you feed the system more cleaned data, your data gets trained, and the more accurate the identification of the features will be.

OK, so your data is trained and the features of the dataset have been engineered to know which ones might influence a purchase. Lastly, AutoML uses automated model selection to build a unique predictive model that weighs the significance of each feature. The higher the weight—relative to the other weights—the more significant the feature is for predicting propensity to buy.

Even better, Einstein tells you the most significant features and determines the percentages of the impact they have on the purchase. So now you have enough information to decide how best to engage your customer to influence a purchase.

With AutoML, the data cleansing, feature engineering, and automated model selection are automated, so no need to hire a data scientist to get those same business predictions.

It Lives on the Salesforce Platform

And finally, because Einstein is part of Salesforce’s trusted platform, all Einstein insights, predictions, recommendations, and actions are served up inside Salesforce—meaning you can take advantage of the same model management and monitoring tools you’ve come to know and love.