Understand Diagnostic Insights

Note

The instructions in this unit assume that you have successfully created an Einstein Discovery story. Refer to the instructions in "Create a Story," the first unit in this Trailhead module.

Your Story’s Outcome Variable and Goal

When you configured the story, you told Einstein Discovery to maximize the CLV variable in AcquiredAccount. CLV is the outcome variable in your story, and maximizing CLV is your goal. All the insights in this story show you how different variables and combinations of variables help explain variations in CLV. The top insights in the list reflect the most statistically significant variations in the outcome variable.

Select the Explanatory Variable

In the Variables panel, click Division.

In the breadcrumbs, click Choose Value and select Naval.

The summary tells us that when Division is Naval, the average CLV is 1.75% higher than the global average. Why is that? Let’s find out. Click Why does this to better?

Einstein Discovery displays a waterfall chart.

What can the chart tell us about why customers who are part of the Naval division are different from the average customer?

Hover over the Global Average bar to get more information.

The Global Outcome has a global mean value of 20,136 and a global count of 10,000. What does this data tell us? That the average CLV across all divisions (including Naval) is $20,136, and there are 10,000 rows (also called observations) in the dataset.

Understand the Division Is Naval Insight

Hover over the Division is Naval bar to see more information.

We can learn a lot about Naval customers from this information. Let’s look at the numbers in the following order so that we can understand the building blocks first.

• Frequency is 3.3%. Customers in our Naval division make up only 3.3% of customers overall. How unfortunate, because our Naval customers have a higher than average CLV. Perhaps it's time to try to acquire potential Naval customers? Or perhaps we realize that the Naval market is small and we focus on other divisions?
• Conditional Frequency is 1 (100%). In our case, 100% of the records in the category Division is Naval are in the Naval division. Perhaps this information seems obvious.

Note

Here’s one way you can use this number. Suppose that the number is high, like 1,000. In that case, the effect of being a Naval customer, with no other factors considered, is that the CLV is $1,000 higher than average. Awesome! But you see that the observed outcome is much less than $1,000. This information indicates that Naval customers have the potential to be valuable, but that something else is dragging this number down.

• Coefficient is -267. What does this coefficient value tell us? That the CLV for Naval division would be $267 lower than the mean if there were no other factors involved. This number tells you that the simple fact that division is Naval influences the CLV for the Naval division.
• Precluded Sum is 4,596. The impact for the average customer includes the impact of customers who are in the Naval division and the impact of those customers who are not. Einstein Discovery calculates the impact that customers who are not in the Naval division have on the CLV of customers who are in the Naval division. In our case, the impact of removing all the effects for divisions that are not Naval is to increase CLV by $4,596.
• Impact is 4,338. This number represents the net impact - also shown as Contribution to Outcome at the bottom of this popup. Impact considers the effect of simply being a Naval customer and the percentage of overall customers that are Naval. Impact also adds in the impact of other customers that are not Naval. What is it telling us, in our case? Without the other factors in the Related to and Unrelated categories, Naval customers would have a CLV of $4,338 more than average. That’s a significant number! Why aren’t we realizing that potential? In the next sections, we find out.

We are done with the first-order analysis in the Division is Naval category. Now let’s look at another category.

This category tells you how different combinations of variables work together to affect the CLV.

Hover over Division is Naval and Rating is Hot  to display details.

This data is surprising. Naval customers are supposed to have a high CLV. Customers with a Hot rating are supposed to be profitable. But customers in the Naval division with a Hot rating have an unremarkable CLV. Here is the observation: Rating is Hot occurs 53% of the time globally but it changes to 92.4% when it is known that Division is Naval. Because of these cases, CLV increases by 440.06.

Who knew? Naval customers that are rated Hot do not significantly boost the overall CLV for Naval customers. Why? Perhaps something is wrong with our rating system. It is possible that Naval customers don’t respond well to the approach we take with Hot accounts.

And there are more surprises. 92.4% of our Naval customers are rated Hot, which is an awfully high percentage of customers to be dragging down the CLV. It's definitely worth looking into this more! Make a note to yourself to investigate later with your team.

Let’s look further into the important but non-obvious factors. Hover over Division is Naval and Type is Customer:

This bar shows an impact of -939. Naval customers who are direct customers (rather than partners or resellers or some other type of channel) really drag down the CLV. And 62% of our Naval customers are direct customers. Wow. This one insight makes you realize how powerful Einstein Discovery is. More than half the customers in one of our most valuable divisions seem to be seriously under-performing the average CLV. Time to investigate what's going on with customer channels in the Naval division!

Finally, hover over Small Terms Related To Division is Naval:

Small terms represent the aggregated effect of all terms (variables) interacting with CLV that do not appear in the other bars in the chart. This category reduces CLV by $3,363.

We got some useful information from categories that are related to the Naval division. Now let’s look at the Unrelated category. But why do we look at information that is unrelated? Good question. The information in this section is “unrelated,” meaning that it is not specific to customers in the Naval division. This section shows us factors that have positive or negative effects on all customers. This section also accounts for how frequently a factor occurs for Naval customers, relative to customers in general. Let’s get more specific.

In other words, Einstein Discovery is sophisticated enough to show how bad things happening less often have a positive impact. To see more examples, let's look at our chart.

Hover over the Unrelated Small Contributors bar to display details.

You quickly realize the power of the Unrelated section. It gives you deeper information about why something happened. In other words, it gives you more power to credit (or give constructive feedback to) the right people.

We’re done looking at the Unrelated factors. Now let's move on to the Unexplained section.

Looking at Unexplained phenomena sounds mysterious. Really, it's just the comparison between the predictions made for all observations in the requested subset, and their overall average, compared with the observed average. The bar shows whether the average for unexplained factors was higher or lower.

Hover over the Unexplained bar to display details.

The difference between the actual average CLV (calculated from the dataset), and the predicted average CLV (from Einstein Discovery’s predictive model), is $481.

Resources

• Salesforce Help: Explore Why a Value Does Better or Worse Than Average