21 Feb Quantifying Digital Experience (DX) into an Index (DXi) using Machine Learning (ML)
The author Dr.Srinivas Kilambi is the Founder and CEO of Sriya DXI LLC.
Friends, I will attempt to introduce this unique concept, its advantages, unique features, actual case studies through a series of posts. Here is the first of the series:
Digital Experience of an user (DX) on a web site or a mobile app or a social media site is very subjective and is attempts are being made to capture DX currently through surveys or through data visualization. Both of these methods are partially successful at best and leave the final analyzer with too many questions unanswered which needs or merits the intervention of expensive data scientists.
So why are surveys and data visualization methods not very successful? Let us start with surveys. how many actual fill any survey, very few. Also, those actually fill a survey usually have a very strong positive or mostly negative bias and want to express that bias through a survey. So surveys represent a very small percent of the actual population and are also very biased/skewed and hence may not be a representative of the actual broad population user DX. So any inferences based on surveys could be very risky and even faulty. Now what about “data analytics (DA)” tools. they indeed are a semi effective way to understand DX but are confusing to most analysts and hence need data scientists for final interpretation. Why? because they first capture data into a set of 10+ independent input variables like total # of visits, time spent on page etc and a set of 5-10 derived parameters like bounce rates and then show the data in a visual format like bar graphs, pie charts etc and allow some visual customization like trend analysis etc. So analysts need to look at maybe 20 parameters and their trends and then try to understand all this into one context namely, “Did the DX rise or drop” from last week/month? No simple and easy answer, in fact most of the times “DA” gives no answer but just vague trends. Imagine there is a CXO level meeting in a company and the CEO asks what was the user DX this month? what was it compared to last month? and relation to sales?. What answer will he/she get?, Something like this: Well based on a 3% of the people who answered our survey we have 25% like, 40% average and 35% dislikes; we have 20 total DX parameters of which 5 are up; 5 are flat; 5 are down big time and 5 are up and down. Imagine the fate of that company, the CXOs and most important the share holders who depend on these CXOs? Not very bright. right?
Most current methods are very aggregate oriented and present an aggregate view of all user’s DX for a time period. This means that every visitor forget the visit of a visitor is not an unique data point in the model. This means if a visitor came 1 time and spend 30 mins and came 3 times and spent 15, 10 and 5 mins respectively aggregating 30 mins is treated same. More than that, if there are 100 visits totally 1000 mins then 10 is the time per visit in most models which is a gross over simplification. Also, these models look at 1 parameter at a time independently and we all know that in real world nothing is truly independent and all parameters have some correlation with each other.
So, Is this really the right way to look at DX? Is there no other way to understand or quantify DX? Does every visit matters? Can we get uniformity and conclusions rather than vague trends? Can we get a score which means same to a CEO, CMO or a CIO/CFO. Good news, the answer is a YES to all these problems/questions and my next set of posts will answer them 1 at a time.