blog – Page 2

A WHITE BOX MARKETING-MIX-MODELING APPROACH TO UNDERSTAND THE KEY DRIVERS OF YOUR BUSINESS

June 28, 2016/in blog /by designmaster

By Neeraj Kulkarni

A recent Fournaise Group global marketing effectiveness study showed that 70% of CEO’s think marketers misunderstand (misuse) the business definition of terms like ROI and have lost trust in their ability to prove effectiveness of marketing spend to drive business goals. Most marketers are still challenged by their strategy and analytic teams working in silos and producing individual channel performance results for mass media (TV, Radio etc.) and online media as if each works in isolation. This approach can leady to faulty reasoning and can lead to millions of dollars being allocated in the channels which don’t yield the highest marginal ROI. The reality is that consumers today are exposed to several media touch points and sales channels. The brand funnel is no longer linear but highly connected with various channels playing a part in the decision making process. Marketers are looking to get clear understanding and solutions to the following 3 key questions:

What are the key business and marketing drivers that impact sales? How do we quantify their synergistic effects to compute contribution and marginal ROI for individual channels?
What is the optimized marketing allocation that will drive brand purchase intent and business objectives?
What are the strategic decisions that need to be made based on the model findings to hit or exceed your brand and business goals?

Current approaches & shortcomings

Marketing mix modeling (MMM) is use of advanced analytics and statistical techniques specifically developed to estimate the impact of marketing activities on sales and then forecast the impact of future sets of activities on business goals. However too often marketing mix models are developed by utilizing only data inputs which are either media or marketing related activities and some seasonal factors. However they rarely take into knowledge inputs from key stakeholders in the organization on future outlook of business, product innovation, management changes and marketing activities. Such naïve modeling approaches are backward looking and depend on past marketing activities to be predictive of future sales or brand behavior and rarely reflect forward looking organizational beliefs and are generally not used for strategic decision making. Due to the dynamic nature of the business and market conditions, this modeling approaches fail to accurately identify and quantify the key marketing drivers which can often lead to serious misallocation of marketing resources.

Customer Intelligence + Experiential Knowledge = Actionable, Predictive Insights

At CIEK, we have developed a marketing mix solution suite which integrates big data and user experience using advanced data sciences and machine learning techniques that are transparent, predictive and highly actionable. This forward looking approach leverages a data strategy which can take in to account data gaps, uncertainties in current business and marketing conditions and allows marketers to test new channels, messages and product mixes to accurately predict the impact of those changes on business objectives. Synergistic impact between cross channel activities are quantified to understand contribution and marginal ROI of marketing activities which can influence media and brand experience planning. Marketers can run scenarios, test marketing plans and run their campaigns with greater confidence. Campaign analysis and learnings act as feedback to the modeling process to constantly validate and fine tune the model results. Strategic recommendations are provided based on the model findings which not only inform media allocations but more importantly provide decision support for strategic planning to drive business goals.

Marketing Leaders – Create Accountability and Drive Growth

The best way to drive organic growth in an organization is by improving current customer lifetime value and developing predictive marketing strategies for high value acquisitions. We all know that. Yet how many times have we seen marketing dollars being wasted over activities which seem to have limited impact on the core business objectives of the organization. Marketers need to drive accountability and utilize fact based decision support to drive their strategies and maximize the impact of marketing activities on their business.

FORECASTING AND DECISION MAKING UNDER UNCERTAINTY- PRESENTATION AT THE 3RD ANNUAL BAYESIA CONFERENCE

June 22, 2016/in blog /by designmaster

By Neeraj Kulkarni

The challenge in attributing value to your marketing mix has never been greater. Today, a variety of analytic solutions related to last touch attribution, statistical time series analysis, and market media planning tools is being used to optimize the marketing mix. Most of these solutions, however, fall short due either to reliance on imperfect information related to direct attribution or failing to take into account management assessments, brand studies, product recalls and other business uncertainties.

We have developed a novel analytic modeling approach that integrates Bayesian network modeling and traditional mixed model regression techniques to solve this problem. It incorporates available marketing data, management decisions, stakeholder user experiences, business trends and market research studies to provide a robust and predictive multichannel marketing optimization process that enables us to identify and forecast of the contribution of various offline and online media channels for client decision support and business optimization.

Neeraj presented a client case study for a leading consumer truck rental company to show the flexibility of the approach. His presentation covered the following topics in detail:

Incorporation of data and expertise as key inputs to the network model.
Understanding the contributions and return on Investment for marketing channels in a multi-channel environment.
Running decision support scenarios to predict business outcomes, optimize budgets and meet business goals.

THE 4 C’S THAT MAKE A MODERN DAY DATA SCIENTIST

June 2, 2016/in blog /by designmaster

By Neeraj Kulkarni

Nate Silver has become a rock star. Data scientist has become a buzzword. The title itself has been butchered enough as it lacks specificity and can be perceived as a glorified pseudonym for a data analyst. Regardless of this, I see enterprises, business pundits and “pseudo” data savvy people talk and use (or more like abuse) it all the time. There are zillion descriptions online and ever organization is trying to hire data scientists in every shape and form. Funnily enough on a recent client engagement, the CMO asked me “Do you think it’s time for use to start looking at Silicon Valley to hire a data scientist to help our analytics team?” Now having practiced analytics for over 15 years and having various titles associated with me during that time, I found that question really fascinating. I believe there is so much hype and diversity in description of that role that it has left marketers confused thinking that they seemingly don’t have people with that kind of talent in their organizations which may or may not be true.

I have described here what I feel are the 4 essential qualities of the modern day data scientist. Hopefully it will help marketers identify this talent closer to their home base rather than Silicon Valley (unless they are already based out of Silicon Valley… duh!)

a) Curiosity

“Intellectual growth should commence at birth and cease only at death.”
– Albert Einstein

Curiosity killed the cat. But not the data cat, it actually made him/her stronger. Data analysts will take a request, implement it, and deliver the results that they arrive at using some statistical techniques with certain degree of confidence. A data scientist will first start with a data/business discovery discussion to understand the essential question (and the underlying context and challenge associated) and then interrogate the data with the end goal in mind. It is that underlying curiosity to research and learn more about the business problem which helps he/she in coming up with the most effective solution. It helps them in identifying the right data sets, the right variables and delivering the right insights which are relevant, timely and meaningful for the C suite to implement.

b) Coding

If you can’t code then you can’t be a data scientist. It’s as simple as that. I have found the KDNuggets and the IBM blogs very informative in keeping myself abreast of new techniques evolving in the field of data mining and marketing decision sciences. It’s important for a data scientist to have a learning mindset and learn about new coding techniques like open source languages like R/Python in their free time thru Coursera or Udacity to hone their skills and be better at their trade. Saying that, I have to emphasize here that it is not how efficient you are as a coder (it helps if you are..) that it will make you a great datascientist but it is about how effective and analytical are you in your mindset to attack a particular problem to achieve your end goal and to arrive at actionable insights which differentiate from you just a coder/programmer to being a data scientist.

c) Causality:

The 91% correlation between the two trends does not imply that people marrying in the state of Alabama are generally unhappy with their marriages and commit suicide by electrocuting using power lines. Spurious correlations can lead to inaccurate conclusions. This is true in today’s marketers who are trying to understanding the value of their marketing efforts by correlating individual channels contributions rather than looking at their marketing ecosystem holistically. Having a good data scientist/s will help you avoid this pitfalls. Good data scientists have solid knowledge of statistics, probability theory and statistical software’s. They have the ability to understand the business problem and then pick the right set of variables that can help answer it. Further they can combine data and human knowledge to derive insights which are not only statistically significant but have causal significance in predicting the outcome.

Communication:

Good data scientists has the ability to communicate their insights in a simple, visual and easy to understand manner. It’s not just enough to just have the technical chops, a data scientist must be able to effectively explain how he or she came to a specific conclusion and convince the internal or external customer that their results should be leveraged. The real “aha” moment for me in presentation is when someone from the audience who is non-technical looks at a chart and draws an insight which may not have even called out on the slide. Instead of telling a marketer that 4500 customers in their database have only transacted once, they call out the fact that 70% of the customers have only transacted once. Through effective storytelling, they provide context to the insights and create a buy-in for the data solution to be implemented.

In my opinion, these are the 4 essential qualities of a good data scientist. You could argue that having expertise in text mining, Hadoop, MongoDB etc etc. are important and go on to list 40 more attributes. But to me those things are part of the learning ethics of a good data scientist which they will train and acquire as per the business need. You can find your data scientists right where and don’t always need to run to Silicon Valley. If you don’t trust me, atleast trust Einstein and he wasn’t even called a data scientist.