What are Einstein Analytics Stories and when would you use them?

When I started working with Einstein Analytics in 2015 needless to say not many knew much about Einstein Analytics. In the past couple of years, the DataTribe as we call it has emerged and we have grown in numbers. I find many more being knowledgeable on Einstein Analytics in terms of visualization and data prep, some have even embarked on predictive analytics and created stories without being stats experts or data mining gurus. But what’s all these Stories in Einstein Analytics about? Why has Salesforce invested so much in extending the reporting and Dashboard capabilities of Einstein Analytics and introduced statistics and modeling? In the beginning, when Salesforce acquired BeyondCore, the original intention was to dig deeper into tables and reports and really get to understand what the data was telling us beyond what we could see with our limited human eyesight. Even after 3+ years with the predictive capabilities, it seems there are still barriers we need to break down in order to make what was originally called Einstein Discovery (but now parcelled up in the fantastic world of Einstein Analytics Stories) digestible. Hence I set myself a new goal for 2020: Spread the word about how to gain huge business benefit by extending reports and dashboards right into the heart of decision-making for each and every user or customer of Salesforce. Now I am not a data scientist, so I’ve teamed up with the brilliant Colin Linsky in writing, editing and validating this content. Colin has a great ability to put Einstein Analytics Stories in context to the business user by making sure that the insight is business-centric and has roots in a business process/decision that could use some predictive decision support.

Let’s set the stage; visualizations and predictive insight

Einstein Analytics is a simple yet very powerful tool kit for business users to support better decision-making by understanding what is going on in your business and to give predictive insights to embed right where decisions are being made. Throughout the blogs I’ve posted, you’ll be very familiar with data management, dashboard building and presenting compelling answers to business questions using Einstein Analytics. In this series of blogs, you’ll be taken on a journey showing you how you can extend dashboards to create more forward-looking insight based experiences.

Let’s be clear, we all make many many decisions at work each and every day. Some small decisions, some large. Some with little or no consequence and some with huge implications if you get them right or wrong! Supporting those decisions using data and insight is a continuum:

  • At the lowest form of support, technology and the user experience is just about showing raw data points and getting the decision-maker to do all of the hard work. That’s easy to do but isn’t providing any insight what-so-ever.
  • Curating the data and create visualizations of the data in the form of absolute values, relative values, comparative values, trends, and anomalies really steps up the help data can provide in the way of insights.
  • Taking a deeper look at the patterns in the data, sifting the strong signals from the weak, particularly when looking at the decision-making around business goals and objectives is the reality of data mining and statistics.
  • Removing decision-making from the end-user and automating the decisions is a final, and sometimes utopian state that usually turns end users and businesses blue with fear. It’s a noble aim, but one that most keep as aspirational.

Many of the blogs I have posted address the second topic above – data management, lenses, dashboards – all of that great stuff that Einstein Analytics is famous for and licenses are bought by the thousands. This series of blogs is all about the third way of supporting decisions – finding and quantifying patterns in your data that are indicative of a good outcome in your decision-making. Einstein Analytics has a set of tools that munch up your data, with reference to your business/decision-goal, and first let you know what the drivers of that outcome have been in the past and then second, allow you to wrap all of those drivers and insights into a predictive model that you deploy back into the business right where people are making their decisions. Let’s leave the fourth topic above to another day when businesses themselves are ready to adopt such transformative interventions.

That last part is interesting…. analytics alone isn’t enough. For any of this kind of decision-support, much like dashboards support decisions, the end-user or system must show the output from these models in a way that makes sense to end-users or consumers. Einstein Analytics creates Stories from the patterns in the data and then deploys the Story as a predictive model right back into Salesforce where people are doing their everyday business.

Analytics leads to Actions – as we call it at Salesforce…. actionable insight. When you inject insights into operational systems you give people the decision-support they need to either direct or confirm their thinking. So, for any project undertaken in the decision support space, you should always have the real world, with real people using real systems in mind. It isn’t about coming up with an insight and running away – this is an end-to-end project that MUST have a business application/process in mind right from the very start. The Einstein Analytics Story is just part of the project – deployment of the Story as a Predictive Model is vital too.

Let’s take a moment to clear up a common confusion in terminology. I often find people using the terms model, algorithm and story interchangeably. Here’s how I see the terms fitting together. Einstein Analytics Plus uses algorithms (statistical modeling techniques) to finds patterns in your data against your target outcomes. These patterns are summarised in a model that is specific to your data. The model is exposed and viewed in Einstein Analytics in a story. The Story is a graphical and numeric representation of the model that was built using Einstein Analytics Plus’s algorithms. So, generally, when we are talking about the “thing” that gets put into production it is a model. That model is stored and deployed in Einstein Analytics via a Story.

Let’s follow that opening with a reminder that Stories in Einstein Analytics do two things:

  1. Investigate any data you expose to the analytical tools and quantify the patterns and drivers against your business outcome.
  2. Package up those Story elements (drivers) into a Predictive Model and creates an embedded scoring and display system in Salesforce.

You can, if you wish, use Einstein Analytics Stories to do just 1. above – go find interesting patterns in the data. What happens in a Story is exactly the same as happens in Einstein Data Insights within the Salesforce platform – it tells you what things have happened in the past and how what’s happened relates to a particular outcome you’re interested in. You don’t always have to deploy your Stories – you can use the tools for research and generally find interesting “nuggets” in your data. However, deploying your Stories via a Predictive Model inside Salesforce is where businesses make a Return on Investment either by increasing revenue or reducing expenditure. Yes, insight into action leads to business performance!

So what is an Einstein Analytics Story?

The practicalities are that Stories are created in Einstein Analytics’ Analytics Studio. They are only available when the user has an Einstein Analytics Plus or an Einstein Discovery User license. Stories can be created from any dataset and aim to answer a business question that you have defined. That business question (or operational decision) must be represented by something in your data that is a historic representation of the decision being made one way or another. Either something happened or it didn’t, either something happened a lot, or it happened only a little. You get the idea – we’ll revisit this principle in more detail in the next blog, but for now, you just need to be thinking about a real-world decision that people make. When setting up a Story, you ask a question in the form of “maximizing or “minimizing” an outcome. That makes sense – you either want more of something or you want less of it, hence you’re making decisions as to what to do next! Now you have a “target” as a reference, the rest of the Story building process assesses all of the other data points in your data set against that target and bundles up the correlations and statistical differences into one single presentation – the Story itself.

Using the data available in an Einstein Analytics data set, and a collection of some fantastic statistics and Machine Learning algorithms Einstein Discovery identifies (and packages into your Story):

  • What happened. Tell me what relates to significant differences in my target. Show me the most important drivers first and then let me know any subtle interactions between any of my data points that drive particular differences in my target.
  • Why it happened. Not all things are equal and with some clever maths Stories can disambiguate the actual importance of any one piece of data in relation to the target. Lots of statistical algorithms are either black box or just give you a high-level view of the relative merits of different parts of your dataset – Einstein Analytics Stories gets right down to the individual categories and values in your data and tells you how important each and everyone is for determining why a particular result has occurred.
  • What will happen. This is rare insight from modeling tools – you select aspects of your business that you think has made a difference or you suspect could make a difference and the Story will let you know whether or not there really would be a difference in the target between one condition or another. You can use this part of a Story to plan new interventions or help be creative on what to do next.
  • How to improve it. This part of a Story is pretty much unique in the business world. As the Story contains both “Why it happened“ and “What will happen“ analyses, for every single record in your dataset, you can see not only why it gets a certain score but also what you could do to improve the chances of the decision going in your favor. We love this part of Einstein Analytics Stories as it is in plain English right in front of the end-user right when they are making their decisions about how to handle the case or customer. Priceless.

The goal for Einstein Discovery has always been to be simple to use and the results to be easy to understand. Therefore, the stories are narrative, you are guided through the story output. Frankly, business users, today have too much data available and too many combinations or places to investigate for them to find insight, so doing this without a tool like Einstein Discovery to build you a Story will create a lot of human bias. Einstein Discovery takes the number crunching and human bias away and instead provide all the insight with narrative explanations and actionable recommendations.

Traditionally companies would hire data scientists to noodle on a business question and many weeks (or months!) later come back with an answer, actually, it can take a lot longer than a few weeks because the world constantly changes and model needs constant tweaking and improvements. In fact, the vast majority of models never make it into the hands of the business user because the business or the operational environment changes so quickly that the model is no longer relevant.

While it is still important to have data scientists to solve those very complicated questions, having tools like Einstein Analytics Stories makes it easier for the business user to get predictive insight without having to create all the models using code or unfriendly Statistics programs. In Einstein Discovery, with a few clicks, you can create a story that answers a business question. Sounds simple, right? Well it gets better, with a few more clicks you can deploy your model directly in Salesforce and all your Account Executive is able to see the likelihood of winning a deal or your Service Rep can see a predicted CSAT for the open cases.

Before ending this series’ opening blog, I think it’s relevant to point out that Einstein Discovery is not the best tool for anything predictive, in fact, there are some things that Einstein Discovery doesn’t do. This is what you cannot use Einstein Discovery for:

  • Affinity recommendations like “because you bought this you may like this”
  • Anomaly or fraud detection
  • Natural language analysis.

Let’s finish up by summing up what Einstein Analytics Stories let you do:

  • Identify factors related to raising or lowering business outcomes
  • Statistically validate findings
  • Find and explain patterns in your data that humans wouldn’t have the time or ability to discover
  • Help disseminate insights to the rest of your organization
  • Drop forward-looking views of decisions into the applications that end-users use in their everyday work
  • Prescribe things you can do to improve your business decisions

Now you know what Stories are, and hopefully peeked interest in having a go in your business, in the next blog, we will look at how you prepare an outline for building your own insight-driven decision-support system.


2 thoughts on “What are Einstein Analytics Stories and when would you use them?”

  • Avatar 1
    Piyush Gupta on February 16, 2020 Reply

    Thanks for this informative blog Rikke.
    I have written one recently myself, pls do have a look:

    https://medium.com/@metal.piyush/citizen-data-scientist-vs-academic-data-scientist-case-study-salesforce-einstein-discovery-f958fced7f97

  • Avatar 2
    Robin Barnwell on July 22, 2020 Reply

    Hello, what are your thoughts on this question? I’ve gone through it loads of times and arrive at D but all answers are plausible….

    A consultant is working with a credit card company that needs help with ongoing fraudulent transactions. The company provides a representative sample dataset for the consultant to analyze in Einstein Discovery. The story’s initial assessment shows that a third-party payment app is the source of these fraudulent transactions.

    However, the company rejects this assessment outcome, stating they have not had a partnership with this payment app long enough for it to be a concern.

    What is the recommended next step to improve the story outcome?

    A. Make adjustments to the story to better demonstrate that the third-party payment app is the culprit.
    B. Use the credit card companies domain knowledge and exclude the third-party payment app from the story.
    C. Explain to the company that the story has returned unbiased results and the initial assessment is accurate.
    D. Ask the credit card company for a more comprehensive dataset to analyze.

Leave a Reply to Robin Barnwell Cancel reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.