Call Us Today! 1.866.937.2224 | Support | Our Blog | About Us | Login|Help@allari.com

5 Steps to Go from Data to Decisions

Many companies have made the investment in BI and Analytics technology but the processes to fully utilize and benefit from the technology is in it's early stages. This post outlines a 5 step process that provides a way to capture the expected return from the technology investment.

In a previous post, on building a data-driven organization, I overviewed Analytics by sharing the information I learned from the book, Behind Every Good Decision further referenced as BEGD.  In this post I wanted to outline the author's  approach, which consists of 5 simple steps, to go from Data to Decisions. 

State of Analytics

Gartner has forecasted that BI and Analytics would grow in 2016 by 5.2% from the prior year to reach a $ 16.9 Billion market. There are two shifts taking place that are interrelated. One is that the focus is shifting from the technology to the process and the other shift is from IT led, system of record, reporting to  business led, self-service analytics. The table below from Gartner shows this and some of the other differences between the modern and traditional BI/Analytics area.

Gartner_Modern_BI_Analytics.jpg

In a survey of 448 senior executives by the Economist Intelligence Unit (EIU) the results indicated the desire and intent to use BI/Analytics were far ahead of the actual positive results. The survey report titled, Broken Links: Why Analytics Investments Have Yet to Pay Off found that although 70% of the executives rate BI/Analytics as "Very" or "Extremely Important" to their business, only 2% say it has had a "broad positive impact"

Some companies have seen returns up to 1300% on Analytics investment.  So the problem isn't the investment in technology but rather the execution of the processes that will use the technology to gain insights and make the changes resulting from the insights into a positive return.  In addition, Analytics is much more successful when it is embedded throughout the business rather than being a pure IT function.  Both the process and the "democratization" of analytics is only in its infancy at most companies.

Gartner says that as analytics has become increasingly strategic to most businesses and central to most business roles, every business is an analytics business, every business process is an analytics process and every person is an analytics user. 

Our previous post on Turning Data into Knowledge highlights the success of the Cincinnati Public School Districts with Analytics after focusing on the process and embedding it into the most important user - the teachers.

B.A.D.I.R. : 5 Steps to Go from Data to Decisions

Successful Analytics = Data Science + Decision Science

  • Data Science - The technical track that derives insights from the data.
  • Decision Science - The business track that aligns stakeholders so that valuable insights produced using data science can be inserted into the organizations decision-making process and converted into action.

Just like many other aspects of technology, Analytics requires the technical skills  but also requires the soft skills to understand the business, understand the questions that need to be answered, and present and sometimes sell insights that will drive business impact.

B.A.D.I.R is an acronym for the framework developed by the authors of Behind Every Good Decision, Piyanka Jain and Puneet Sharma, that stands for Business question, Analysis plan, Data collection, Insights, and Recommendations. The two tracks, Data Science and Decision Science, progress through these 5 steps and the authors believe analytics fail when any of these steps are not followed or skipped altogether.

BADIR_steps.jpgFrom Behind Every Good Decision: How Anyone Can Use Business Analytics to Turn Data Into Profitable Insight

Step 1: Business Question

I think all answers are either out there somewhere in the universe or somewhere in our own internal universe waiting to be discovered. Most of the time it's a great question that leads us to that discovery. In both science and business, a well defined and well thought out question is critical.

Quite often, data is considered to be the beginning and the requests come in for data. To exlore that let's use an example request, "give me conversion data by country, product and feature". 

On the surface, fulfilling this request is easy enough and can be done to the letter, though it might not necessarily be to the intended spirit of the question. Most likely it will take several more request iterations to get the data sets actually intended. This is fine, but if there is more information on the problem/question upfront, it may help you to help the person making the request go through fewer iterations.

Of course, the other possibility is that you hand over the requested data making note that revenue appears to be trending down. The requester, in turn, replies with a little anger that they already knew this and are trying to figure out the reasons why. Sometimes the request given is not always the same as the expectation- seems to happen a lot when the request comes from up above.

Start with the traditional What, Who, Where, When, Why and How questions to help identify the problem in its context. Understanding the context, the impacted segment and potential reasons as understood by the business may provide a quicker path to the resolving the problem. This is true for both the data science (context, impacted segment, potential reasons) and decision science (timeline, stakeholders, actions) paths as shown in the diagram below.

Business_Question.jpgFrom Behind Every Good Decision: How Anyone Can Use Business Analytics to Turn Data Into Profitable Insight

The above diagram from the book, BEGD, shows an example for a fictitious online pet supply company. The process that started with "Give me conversion data by country, product and features turned into the following well defined and well though out business question:

What are the reasons for the conversion drop after the dog leash online checkout feature was launched in the UK? What actions can QA, PD, and PM take to address the bleeding?


http://info.allari.com/download-free-ebook-4-trends-todays-it-leaders-must-master

Step 2: Analysis Plan

The Analysis Plan step includes 5 parts; analysis goals, hypothesis, methodology, data specification and project plan as shown below in the diagram from BEGD. Each of these parts need to be communicated and reviewed with the stakeholders.

Analysis_Plan_Step.jpg
From Behind Every Good Decision: How Anyone Can Use Business Analytics to Turn Data Into Profitable Insight

Analysis Goals

Business problems and questions can be large in scope so narrowing the objectives or breaking it into parts, where each has its own set of analysis goals is important.  These goals should follow the S.M.A.R.T methodology - Specific, Measurable, Attainable, Relevant and Time bound.  See our post, The Art and Science of Goal Setting for more info on this methodology. The following is an example from the book:

Business Question: What are the reasons for the conversion drop after the dog leash online checkout feature was launched in the UK? What actions can QA, PD, and PM take to address the bleeding?

Analysis Goal: Determine drivers of conversion, and segments where conversion in the test population is lower than in the control group.

Hypothesis

Analytics can be compared to Detective Work as it primarily uses the deductive approach as shown below. In most types of business analytics we start with what we have experienced and know, then brainstorm some hypothesis (educated guesses to cause and effect related to the Business Question and Analysis Goals) that we test and confirm or reject based on observations or data.

Deductive Approach

On a side note I used Sherlock Holmes as the feature image for both Analytics post as he is the icon for detective. However, he was very unique and in most cases used the inductive approach shown below. He starts with the data/observations that only he can seem to make and determines a pattern and then solves the crime.  There are analytics cases, especially for the more advanced predictive analysis, where the inductive approach is used but for the vast majority of cases we are still relying on people (non Sherlock Holmes type) to start the process with Question and Hypothesis before using the data.Inductive_Approach.jpg

The brainstorming should include those closest to the business problem and question. These hypothesis would be prioritized so the top ones could be tested. In the books example they prioritized three: software bugs, a problem with the untested IE6 browser code, and problems with the new Chrome browser.

Methodology

Some common analytics methodologies are used to help solve business problems.  BEGD has a chapter on the seven most common ones used.  The table below describes each at a high level.  The book provides more details on when to use each and typical use cases which are more detailed than I can cover in this blog.

Analytics_Methodology.jpgFrom Behind Every Good Decision: How Anyone Can Use Business Analytics to Turn Data Into Profitable Insight

Data Specification

The data specification describes the data that will be needed to prove or disprove the hypothesis and support the chosen methodology.  It's important to only collect relevant data. The keys are to determine the granularity of the data, weekly, monthly, geography, etc. A data field should serve as a unique ID. The specification for the book's example is shown below. They needed transaction level data to determine which transactions resulted in conversion issues.

Data_Specification.jpgFrom Behind Every Good Decision: How Anyone Can Use Business Analytics to Turn Data Into Profitable Insight

Project Plan

The project plan will tie together all of the information gained and serve as the road map for the analysis execution and steps that follow.  

Step 3: Data Collection

This step is quite often the starting point where the first 2 steps are skipped or skimmed too quickly which result in less successful returns on the analysis and an unfair reaction that analysis does not work. The main parts of this step are fairly straightforward:

  • Data pull which is the collection of the data following the data specification from the previous step.
  • Data cleansing and validation to avoid the GIGO (garbage in, garbage out) syndrome the data needs to cleaned for usability and validated for accuracy.

Step 4: Derive Insights

This step is the execution portion of the process where data meets methodology. At the highest level, regardless of the methodology chosen there are three parts shown in the diagram and described directly from BEGD below.

Derive_Insights.jpg

  • Review patterns: It helps validate patterns in the data if there is a real problem and if there are unusual patterns in key variables.
  • Prove or disprove hypotheses: Look at each hypothesis, one at a time, and examine the relevant data needed to prove or disprove each one. This will help eliminate some hypotheses and identify the ones on which you should focus your energies.
  • Present findings: Finally, present your findings in terms of quantified impact to guide prioritization of the hypotheses for analysis.

Step 5: Recommendations

This may be the most important step as it is required to take action for positive impact.  The main purpose of the process is to turn data into insights and then insights into actions.  Usually the recommendations are done by presenting to an audience so the authors of BEGD suggest trying to achieve three things with your recommendations:

  1. Engage the audience by presenting a short, concise, insightful set of recommendations without getting bogged down in detail.
  2. Be perceived as an effective business partner by presenting credible recommendations.
  3. Drive the audience towards actions that create impact by solving the business problem

Summary

I've tried to present an outline of the book that I found a lot of value after reading.  I don't know the authors or have any connection to their organization but found their ideas and methods both very well presented with great examples and enough detail to make an effort at adapting or gave me a starting point for further exploration of details such as the analytic methodologies. I shared a lot of material directly from the book as I felt just suggesting you read it would not be as effective as sharing highlights from the material.  I leave you with one more diagram from the book that summarizes the 5 Steps.

5_Step_Summary.jpg

http://info.allari.com/download-free-ebook-trends-it-managers-must-master