UPDATE 8/15/2016: One week after writing this blog, news that Salesforce signs an agreement to acquire BeyondCore was released. The timing is ironically magical! I knew BeyondCore was a great solution in 2015 after getting a tip from a peer. What I saw was verrrry cool. My crystal ball for finding awesome unknown solutions is getting better with age. 

To continue my journey that started in Moving Beyond Data Visualization, I finally signed up for a free trial of BeyondCore. At Gartner BI Summit 2016, I saw a compelling demo of BeyondCore that had found hidden insights in the BI Bake-Off demo data set that I had totally missed using Power BI and Quick Insights. I was intrigued. I knew BeyondCore was cited as an up and coming Gartner Smart Data Discovery vendor but I had not tried it. After running a few data sets through it and meeting several existing customers from major banks, McKinsey, and Fortune 500 companies, I am impressed.

One of the conversations that I had with the BeyondCore team was about the differences in Power BI Quick Insights, R and narratives integration,  Tableau’s forecasting, clustering and R integration, Qlik’s narratives, TIBCO’s Spotfire Analytic Models and TERR R, SAS’s Visaul Analytics and Viya, etc.

The automated analytics pitch sure sounds similar… until you actually see how and what they are doing. The BeyondCore approach is really quite different from anything else that I have seen in the market today.

We also chatted about the time it would have taken me to explore all of the demo data and the thousands of combinations analyzed. They told me that they automatically found the insights that I had missed in a few minutes using 15+ patented technologies. The BeyondCore technology explores data at a much deeper level than other automated tools.

Honestly, I might never have found the hidden gems of detailed insights that BeyondCore automation found through manual, drag-drop, build a visualization approach, or with Power BI Quick Insights, or even running savvy data scientist level machine learning tools. These guys had good points but I needed to experience it “hands-on” to believe it.

Manual Analysis

BeyondCore: Finding Signals in the Noise

BeyondCore analyzes millions of data combinations in minutes, for unbiased answers, explanations and recommendations. Unlike manual data analysis, BeyondCore smart data discovery automatically finds and explains statistically significant key metric drivers that truly matter. It also explains what happened in natural language, why it happened, what will happen and how you can improve it.

To identify relationships in data “the signals” and isolate distracting, irrelevant data “the noise”, BeyondCore uses machine learning algorithms (prediction models) for estimating relationships among independent variables (product category, region, date of sale, customer type) and an outcome measure (revenue, units, days).

Actionable Insights: Optimizing Outcomes

Most interesting to me, BeyondCore allows a user to select an outcome (win/loss rate, sales amount, discount rate, days to close) that you want to optimize. Users can select variable combinations and BeyondCore can prescribe what you can do to optimize the outcome at a detailed, granular level.

Although I have seen optimization in prescriptive solutions in advanced analytics tools for years now…the implementation of recommended actionable analytics is new and different.

BeyondCore recommended

Click to see full-size image

Getting Started with a Sample

To get started with BeyondCore, I used the Salesforce Win/Loss example data. You can download it from the BeyondCore Support website.

After logging in I watched the Guided Tour and then uploaded the Salesforce Win/Loss example file. BeyondCore displayed a Prepare Data screen with recommended and automated data cleansing steps. In the sample file, there are few errors that BeyondCore suggests to fix. Seeing as predictive data preparation is an art and a science that often takes up a lot of time, I really appreciated this particular feature.

Data Prep

Click to see full-size image

On the Story Setup screen, I picked Won (deal win rate) at an Opportunity (deal) level to analyze. Then I clicked Create Story to start the automated analysis. When it finished, I was shown a summary of results and a link to an automated voice narrative/visual briefing.  Although the automated briefing is really cool – I jumped right to the detail.

Summary Story

Click to see full-size image

On the BeyondCore Story Summary, the statistically significant single variables that influence the Win/Loss outcome being analyzed are shown on the left side in order of variable influence. That alone was good to know. One golden nugget per se that BeyondCore already identified was that Promotion did not influence Win/Loss rates.

On the center of the screen, BeyondCore shows more detailed variable combinations that positively and negatively influence the Win/Loss outcome. Those are both written out and also visualized with red and green bars. The bright blue color bars indicate other statistically significant values. Here you can see that White Paper as a Lead Source significantly influences Won deals. Exploring further, Referral is also significant.

Scrolling down, BeyondCore shows you the detailed natural language, textual explanations of findings for each statistically significant variable. The story is organized by interactive What Happened (descriptive graphs), Why it Happened (diagnostic graphs), What will Happen (predictive graphs) and How to Improve It (prescriptive graphs).

What Happened (Descriptive Graphs)

The descriptive graphs are shown as a bar chart. Average outcome is a dashed line. The variables that significantly influence the Win/Loss outcome are filled bars. Translucent or faded bars are not statistically significant. You can get much more detailed analysis by adding cards shown at the right to see combinations of influencing variables.

BeyondCore Descriptive

Click to see full-size image

Why it Happened (Diagnostic Graphs)

By clicking Diagnostic on the top right menu, I explored the increase or decrease to Win/Loss by variables in a waterfall diagram by hovering over each bar. This chart is also segmented on the bottom axis to show overall contribution to the outcome.

BeyondCore Waterfall

Click to see full-size image

What will Happen (Predictive Graphs)

Predictive graphs have several options that I used to experiment with variables. This kind of What-If feature is rarely found in data discovery tools today even though traditional BI tools have included it for years. One difference from typical What-If analysis is that BeyondCore’s What-If capability is based on predictive and prescriptive algorithms. Other predictive options include Bulk Predictions for scoring data sets and Exploring Regression Terms.

For my hands-on evaluation, I went ahead and tried the What-If analysis comparing Enterprise Suite and Personal Desktop deal Win/Loss rates based on different discount levels. Here is a sneak peek of my analysis. Oddly 15% discounts resulted in more deals being lost than 5%, 10% or 20% discounts for Enterprise Suite. I chose to add that interesting insight to my Story.

BeyondCore What-If analysis

Click to see full-size image

How to Improve It (Prescriptive Graphs)

My favorite area of BeyondCore is the prescriptive recommendations.

Prescriptive recommendations are truly actionable, intelligent insights that can make an impact on chosen outcomes = a game changer.

The prescriptive recommendations take a couple minutes to run. When the process finishes, another set of recommendation cards was shown. To explore recommendations, I clicked on a card to review details.

Recommendations

Here are the results from picking the recommendation card: Predicted impact of 133,000 if change Lead Source from Channel Partner to Referral when Primary Competitor is Vista.

BeyondCore Prescriptive

Click to see full-size image

Although there are more things that I could have tried like scheduling analysis, bulk predictions or integrating the BeyondCore REST API into an app for embedded intelligence, I’ll save those topics for future articles. Right now, I want to wrap up with the export feature to share analysis.

Sharing BeyondCore Insights

BeyondCore makes sharing stories incredibly easy to do. Now that I have found fascinating hidden gems that can improve sales effectiveness, increase win rates, optimize discount rates and marketing spend on the right lead sources, I’ll export my results into a PowerPoint presentation.

In the top right menu, there is a Download icon with several options: HTML, Microsoft Word and Microsoft PowerPoint. I chose PowerPoint. Looking at the exported content, I noticed BeyondCore created a PowerPoint presentation with my Story graphs – and – the textual summaries in slide notes. Nice.

BeyondCore PowerPoint

Click to see full-size image

Learning from Mistakes

What I did not tell you right away… on my first couple attempts, I failed in my BeyondCore trial. I initially uploaded AdventureWorks data and Tableau Super Store data that was not in predictive-friendly format. I rarely read the manual first or follow directions. I just dive right on in and start playing!

Don’t make my mistake. After realizing what I did, taking a moment to understand predictive data formatting and trying it again with a provided sample data set, I then succeeded. I also had great results with UCI Machine Learning and Kaggle data sets.

My initial failure made me question the quick insight results in data discovery tools. Data discovery tools like Power BI, Tableau, Qlik, TIBCO Spotfire, and so on use data in dimensional formats that are not optimized for predictive modeling. From what I can tell, those tools are not automatically preparing data, pivoting, binning, asking about outlier handling, decision granularity, target outcome, etc. to provide accurate predictive model results. They also don’t give me detailed multiple variable level prescriptive recommendations.

Are data discovery built in automated predictive results reliable for decision making? It depends…they are nice for basic outlier detection, trends, correlation and descriptive level insights.

The bottom line is BeyondCore complements data discovery tools. A modern enterprise analytics strategy would likely have both kinds of solutions as Gartner mentions in published Smart Data Discovery research.

To Learn More about BeyondCore

To get the most out of a free trial with BeyondCore, I strongly suggest starting with the Getting Started doc and data set to learn the user interface and features. Then you can copy your own data into the example file templates or prep your own. Note there are data limits on the free trial but not with the production version. If you want to test with millions of rows, you need to contact them directly for a different license.

I have sent feedback to BeyondCore on where I got a little lost during my evaluation. Their responsive product team is already working on my suggestions in an upcoming version. If you get lost, send them a note.

To continue learning about BeyondCore automated prescriptive analytics, data preparation, scheduled analysis, bulk predictions, integration with applications, and more advanced capabilities that are too extensive for one article, check out the following resources.