I am sure most of you saw the latest announcements around Salesforce Einstein. At his Dreamforce ’16 keynote, Salesforce CEO Marc Benioff talked about a world where Einstein will make artificial intelligence (AI) capabilities pervasive across the Salesforce Cloud.
Einstein leverages all data within Salesforce — customer data; activity data from Chatter, email, calendar and eCommerce; social data streams such as tweets and images; and even Internet of Things (IoT) signals — to train machine learning models.
Having seen data-driven analytics help customers at PROS to achieve their desired outcomes, I see a few interesting challenges for some of Salesforce’s use cases, which include predictive sales, marketing, commerce, and services recommendations.
- If data is not accurate or timely, it compromises overall value. Data quality and accuracy of data within Salesforce is still a challenge. Even though special tools exist in the master data world for data quality and assessment, getting it right within the transaction system (CRM, support, marketing) will not be easy.
- It takes a lot of legwork to get data in the right shape. Simply having the ability to call Einstein’s machine learning APIs is not going to be enough. You need real-world domain experience in marketing, pricing and sales.
- While unstructured data processing has greatly improved, analysis of structured data is still more relevant in an enterprise setting. Salesforce does not have easy access to all the structured data it needs for Einstein. As long as a majority of enterprise companies still host major business processes outside of the Salesforce ecosystem, significant effort will be required to merge and analyze both data sets, which is essential in drawing valid conclusions.
To further elaborate on point #3, let us look at a simple use case example that Einstein is promising to solve: upsell and cross-sell recommendations.
At PROS, we combine all past historical transactions (typically in the order management module of an ERP system) with Salesforce or other CRM data. We base our recommendations on actual customer buying patterns and comparisons to the transactions of similarly segmented companies. The PROS approach delivers product recommendations that were purchased either in tandem or in place of other products by that specific account. That means there is a greater likelihood that customer will buy the recommended products, resulting in potentially larger deals. Sales/CRM data alone is not enough to uncover these opportunities.
In short, Einstein offers a piece of the prescriptive analytics puzzle, but it still has CRM-specific limitations. With two full days of Dreamforce to go, I hope to learn more about Einstein. Let me know what you think.