Einstein will allow Salesforce administrators and business users, as well as developers, to bring features such as smart image identification or policy automation to applications. In building these capabilities directly into its applications, Salesforce has gone further than other vendors' AI services, including those from Microsoft, IBM, Google and Amazon, which target developers and data scientists.
By packaging up AI so that it can be harnessed using "clicks not code," Salesforce aims to "democratize AI just as we democratized CRM," says John Ball, Salesforce senior vice president and general manager of Einstein:
There aren't enough data scientists in the world to build predictive models for every company. So [Einstein] is built right into the [Salesforce] platform. We're democratizing AI so that customers can get the benefit from AI without having to hire those data scientists.
Einstein in 3 levels
Unveiled tomorrow in the run-up to an official announcement at the cloud giant's annual Dreamforce conference in two weeks' time, Einstein's AI capabilities are to be made available at three separate levels of coding complexity:
- Salesforce administrators and business users will be able to configure functions that use Einstein services in each of the main Salesforce applications, with no coding required. Examples include predictive lead scoring in Sales Cloud, recommended case classification and predictive close times in Service Cloud, and automated service escalation to create Service Cloud cases for unanswered queries in Community Cloud. Marketing Cloud will gain functions such as automated audience segmentation, while Commerce Cloud gains several personalization features. Analytics Cloud and IoT Cloud will add functions such as predictive analytics, smart data discovery, recommendations and automated rules optimization.
- Salesforce developers will be able to call Apex-level services as they build or modify applications on the Salesforce platform, including services that will allow them to train deep learning models to recognize and classify images and text sentiment. For example, they might train an Einstein service to identify images that contain their company logo, and then use that service to find such images as they appear on social media. Initially there are three services available — predictive vision, predictive sentiment and predictive modeling. Ball said developers will be able to harness these without needing to be machine learning experts, although some data preparation will be required, which Salesforce will explain in further detail once the service becomes available.
- Developers with data science skills will be able to build their own predictive and machine learning models using the open-source PredictionIO service running in Heroku Private Spaces.
Because they are native services in the Salesforce platform, the new Einstein capabilities will work with existing applications, including custom fields and objects. Many will be available as part of existing licenses at no additional charge, said Ball, and further capabilities will be added in future releases:
Einstein is a core part of our platform. What you'll see is it's powering AI capabilities across all of our cloud, in every single release. We have a lot of great capabilities coming out in the October release. Some features will be included, others will be an additional charge.
Some functions are generally available immediately. Community Cloud Einstein's automated community case escalation and recommended experts, files and groups as well as Commerce Cloud Einstein's product recommendations are available now at no extra charge. Analytics Cloud Einstein's smart data discovery is also available now, with pricing based on the volume of data and number of users.
Data privacy questions
Also announced today is a new Salesforce Research group, which brings together a team of researchers and data scientists under the leadership of Richard Socher, Chief Scientist at Salesforce. The group's mission is to continue to "push the state of the art in AI" says Socher, which will lay the ground for future enhancements to the platform.
One of the exciting things for us at Salesforce Research is to build this metaplatform that enables other folk to solve their own problems.
Today's announcements build on nine separate acquisitions by Salesforce of AI expertise in recent years, most notably Metamind, which Socher founded and led and which was acquired earlier this year. Others include Implisit, RelateIQ and PredicitionIO.
Einstein is drawing on the full range of Salesforce data to train its predictive models, including customer data, activity data from Chatter, email, calendar and ecommerce, social data streams and even IoT signals. But Ball insisted that this use of data won't impinge on data privacy constraints.
Trust is our number-one value. Einstein is no different. All of the issues apply here.
What's super important to understand is that the entire process is automated. No data scientist is looking at the data.
Data is not being shared between customers. This is a machine learning process.
Ball also confirmed that customers could opt not to use the service, and could opt out of their data being used by Salesforce Research.
This is a significant move by Salesforce, although beneath all the razzmatazz of the announcement, what's actually available initially is somewhat sparse and in some respects not fully thought through. In an email exchange after last week's pre-briefing, long-time SaaS watcher Jeff Kaplan of ThinkStrategies articulated exactly what I'd been thinking:
As with many of Salesforce's past offerings, once again it is promoting a brilliant vision of a compelling solution without a comparable set of defined features, functions, packaging and pricing.
The saving grace is that this is not another Wave, which was first announced to great fanfare two years ago and is only now starting to deliver practical outcomes. The advantage that Einstein has is that it can deliver value in small increments, and therefore I think the impact will be seen much faster. That will make it a real crowd-pleaser at Dreamforce.
But there are a lot of unanswered questions. I feel that the data privacy issues have not been fully thought through — Europeans in particular will want clearer answers on that score. And while Salesforce is eager to get Einstein in the hands of customers and developers, I got very unsatisfactory answers about how this is being rolled out to the vendors' massive partner ecosystem — pricing and licensing decisions there apparently are still at an early stage.
Despite those caveats, this is a significant move into AI by Salesforce and one that will increase its value to customers. The devil will be in the detail, of which more will be on show at Dreamforce, but my initial assessment is that Salesforce has played a strong hand here.