Dreamforce, Salesforce's massive annual user event, takes place next week in San Francisco, drawing tens of thousands of people to the Moscone Center to get a better understanding of the latest and greatest from the world's top CRM vendor. Given the current mania around all things machine learning and artificial intelligence (AI), I'm expecting Salesforce's own AI platform, Einstein, to be front and center at the event.
It was at Dreamforce 2016 when Salesforce first announced Einstein, and even though two years have passed, customer feedback remains tepid at best. Earlier this month I interviewed a CEO of a mid-size media company about his experience with Einstein, and his response was, "It doesn't work." I don't believe there's anything wrong with Einstein itself. In fact, I think it's quite a good product. Salesforce is one of the most innovative companies in the world, and I'm sure its AI is very strong. The problem is one of the data, and as the old axiom goes, "Bad data leads to bad insights."
The Bad Data Problem
The first issue is that Salesforce has no ability to aggregate the data from its massive customer base. The product was built in an era where the paranoia about the cloud was at an all-time high, so it was built in a way that each customer's data is completely isolated from every other's. A good way to think about it is that there is no single Salesforce cloud. Rather, each customer is its own virtual private cloud. This hampers Salesforce's ability to aggregate data and provide cross-customer analytics. A data set that spans the entire Salesforce customer base will lead to much better insights than a single customer.
The second issue -- and perhaps the bigger problem -- is that much of the customer inputted data is bad data. Sales reps, marketing people, and others hate the process of inputting data into Salesforce, and this creates a double-edged sword for most businesses ...one where both sides of the blade are bad. If the salesperson is a poor performing one, they'll likely just input data as quickly as possible, doing the minimal amount of work, which leads to bad data. If the salesperson is a highly diligent, A-type individual, they can spend hours doing nothing but entering data into the CRM system. I've talked to some sales managers that have told me that some of their best people spend up to one full day a week entering data into Salesforce, while the worst ones enter nothing useful and are hiding behind the old "Do you want me to sell or do data entry?" excuse.
Salesforce has built some automated tools enabling data to be ingested automatically when emails are sent or meetings take place. In theory, that should save some time, but this does require a clean database. For example, a sales rep might send an email to a big client with multiple subsidiaries and when Salesforce tries to import the information, it runs across three people with the same name and 100 instances of the company, prompting the user to take action and classify the record correctly. Again, a bad rep will just click the first record, making their job easy; the process-driven rep, on the other hand, will have to go through the painstaking task of making sure it's correct.
Solving the Problem
I recently ran across a company, called People.ai, that helps with this problem and had a chance to interview its CEO Oleg Rogynskyy. He started the company because he had spent years doing sales jobs and sales management and always struggled to build high-performance teams with repeatable processes. When he looked at what the barriers were to that, he realized it was from a lack of good data. Rogynskyy then interviewed over 100 sales leaders and recognized a pattern: None of them were able to gain an understanding of what their teams were doing from the data in Salesforce. Solving this problem was the ideation behind People.ai.
People.ai is an AI-based application that runs on top of Salesforce. It automates the inputting of data by ingesting phone, calendar, and email activity information; creates a constant feedback loop; and provides recommendations on what the reps should do next. The goal behind the product was to fully automate the insertion of data and then to simplify sales reps' jobs by giving them visibility and guidance for their next best actions. For sales managers, People.ai can show the reps' activity, which reps are dropping the ball, and at what step in the process. Now sales managers can, as Rogynskyy stated, "track the hell out of the sales process and build a high-performance sales force."
For example, many sales reps drag along sales engineers to every meeting because they are not trained on the product well enough, which is a waste of these engineers' time. They should be brought in for proof of concepts, to help with complex late-stage demo presentations, or other times when technical knowledge is required. Initial meetings, competitive discussions, and the like is something most sales reps should be able to handle by themselves. People.ai's activity-based dashboard would proactively show which reps are taking up too much of the engineers' time. If a sales rep is taking along an engineer for a first-level meeting, the manager can be notified and corrective action can be taken. This makes everyone more productive.
People.ai differs from the automation tools that Salesforce offers in that it uses AI to be precise about how information is entered. This makes it more accurate and significantly faster than people doing it manually. From the initial research Rogynskyy did, he learned that only about 20% of data that could be entered into Salesforce is actually in the system, and with that, only 12% is in the right location. The AI-based algorithms that People.ai use are now tracking well over 90% accuracy and giving organizations considerably more data to work with.
Also, unlike Salesforce, which is a set of virtual private clouds, People.ai is built on a modernized cloud architecture where certain anonymized customers' metadata can be shared (without disclosing proprietary, competitive, or other types of sensitive information) to help People.ai improve its accuracy for all of the customers on the platform. When the company first launched, Rogynskyy said, one out of every 100 activities required manual intervention to be classified correctly. Today, it's about one in 6,000 activities, and by end of the year that number will be one in 10,000. People.ai primarily uses unsupervised learning to fuel its product but with a small bit of supervision sprinkled in to fill the few gaps it has.
There is currently a tremendous amount of rhetoric around AI and people being concerned it will kill off jobs. People.ai is an excellent example of an AI use case where, rather than eliminate jobs, the technology can augment the job of a sales rep or sales manager and do the heavy lifting to make everyone more productive. AI should be thought of as everyone's friend, and if you're a company that's heavily invested in Salesforce, People.ai should be something you look at to get the most out of both AI and your data.
Follow Zeus Kerravala on Twitter!
@zkerravala
Related content:
