How Salesforce Einstein AI makes BI predictive at FinancialForce

While business intelligence (BI) is useful for looking at past trends, where it gets really interesting is when it becomes possible to accurately extrapolate the future. That's one of the big selling points for adding artificial intelligence (AI) to BI — the ability to apply the brute power of machine learning to predictive analytics. With the help of Salesforce Einstein, cloud ERP vendor FinancialForce is now doing exactly that to help it manage churn, improve collections and pick out its best sales prospects.

The SaaS vendor is currently refining these applications of the technology for its own internal use, but plans to productize the results to offer to customers during 2018. Internally, the business has seen significant impact on its ability to target sales and customer success resources effectively, according to Joe Thomas, Analytics Solution Evangelist at FinancialForce, who told me on a recent visit to London:

We've been pretty impressed with the technology.

This is changing how we run our business ... Behavior is changing within weeks of introducing this technology.

Feeding data to Einstein

FinancialForce is able to take advantage of the Einstein technology because its entire product suite is built on the Salesforce platform. This also means that it can analyze data from its own financials, supply chain and professional services applications, alongside CRM data from Sales Cloud, Service Cloud and other Salesforce applications. This is a potent combination of customer data that makes it possible, for example, to analyze prospects based on how quickly they're likely to pay their bills, or to see the impact of various aspects of an implementation project on the likelihood the customer will renew in a year's time.

All of this is enabled by feeding the FinancialForce data into Einstein Discovery in a two-step process that applies machine learning to reveal the patterns in the data. First of all the machine learning has to be trained to find the patterns. Thomas says it took about 40 hours to load the entire historic dataset into Einstein Discovery to get started, and then it's just a 20-minute update every week to bring the training up to date with fresh data.

The second step is to run the reporting using live data, as interpreted by the trained system. This means the analytics can power real-time dashboards and generate alerts if something suddenly changes — for example, a project suddenly threatens to overrun, or a customer becomes at risk.

Real-time analysis on mobile

A further benefit is that moving reporting to Einstein has taken that load off the application servers, improving transaction speeds. Previously, FinancialForce had used the native reporting within the core Salesforce platform rather than using the separate Analytics Cloud capability, which was formerly known as Wave. But with the addition of AI, it makes sense to go "all-in" on the renamed Einstein Analytics, says Thomas. The reporting itself is "really fast" he adds, as well as timely:

We're not building a data mart. We're grabbing the data in real-time, combining it and building it. On a weekly basis we're retraining the system, [but] we're always seeing real-time data.

The first dashboard, known as Customer Pulse, has been rolled out internally to sales managers and will soon be available to all sales people. They will primarily use it on mobile devices, so that they can have up-to-date metrics in front of them when meeting with customers and prospects.

AI makes BI predictive

The machine learning has turned up several predictive patterns from the data that are helpful for managing customer attrition. For example, if a customer has implemented and is using custom objects that tailor the FinancialForce applications to their business, that's a strong indicator that they'll renew. On the other hand, late payment doesn't correlate to churn — often it's simply an indicator of the size or complexity of the organization.

Another example is the ability to produce a chart of the customer base that shows at a glance which customers are most likely to pay late, the size of each account and when their renewal comes up. This is helping the vendor take pre-emptive action, for example by amending the billing cycle or following up invoices earlier.

In sales forecasting, analyzing opportunities against predictors of customer health can help identify the best prospects to concentrate on.

Human oversight is still needed

Profitability and success metrics for professional services projects are another area where the machine learning insights are expected to prove useful. For example, it can recommend a specific mix of people or start date to improve project health. These are not necessarily results that humans wouldn't be able to arrive at, but it speeds the process by automating the initial analysis to produce a set of suggestions that humans can then evaluate. Sometimes the algorithms produce insights that a manual review would have missed.

On the other hand, not all of the machine's suggestions make sense — because it doesn't have all the information to make the right choice. So human oversight is still needed. For example, it might see a high failure rate associated with one project manager without realizing that person is typically sent in as a troubleshooter when projects are going awry. In one case, says Thomas, it recommended moving the start date of a project back to September — even though it was already November. It's at moments like this that the limitations of so-called artificial intelligence are exposed, he explains:

The machine is artificially stupid. It doesn't know we can't control time.

There's always going to be a training element, a need for human intervention. AI is not going to steal all our jobs, it's going to make our lives easier.

My take

An interesting insight into some useful real-world applications of artifical intelligence in business analytics — tempered by a helpful reminder of its current shortcomings.