Marketers Should Use Modeling to Compress Large Data into Small Answers

Let’s just say it: Most people don’t like to get involved with analytics. Even people with good aptitudes for analytics don’t care to read long dissertations unless they have to. Modern human beings are too busy to dig deep into every matter that they encounter. New generations are becoming exceedingly impatient and they won’t sit through any boring moment. All this is a real burden for professional data players, because everything they produce must not only be valuable, but also be simple and succinct. Further, they must consider the possibility of their work being misunderstood or misused, because not many users would pay attention to boring details of backgrounds, assumptions, limitations or cautions. Analysts must deliver answers to questions, not mounds of data. Data modeling is the best way to compress large and complex data into concise answers to questions. Why model? The history of applying modeling techniques to marketing goes back to the 1960s. Smart folks at large publishing companies asked themselves, “Why should we send an expensive mail piece to everyone in the database? How about reducing the marketing costs by predicting who would respond to campaign offers before sending out anything?” You can imagine how much money you would save if you reduce the mailing quantity by 80% to 90%, while obtaining similar responses by model-based targeting. That was the birth of response modeling, and they achieved great success with old mainframe computers that were about a million times slower than their modern equivalents. Interesting point to remember is that the mathematical part of modeling hasn’t changed much since. Even neural net algorithms that drive machine learning date back decades; it’s just that the computing speed finally caught up with the ideas and theories. We develop models for marketing purposes, as they increase targeting accuracy; reduce costs by contacting less; enable marketers to stay relevant to target customers; bring consistent results; reveal hidden patterns in data; help marketers reach marketing automation sooner; expand the target universe; and supposedly save time and effort. You might ask yourself, “Wait, contacting targets through digital channels is so much cheaper than sending mail pieces or catalogs. Why bother with modeling now?” Opportunity cost is very much real, and you should not contact everyone 10 times a week just because you can. You may be training your target audience to ignore you, or worse, encouraging them to opt out. Also, modern day modeling is not just about ranking target lists, but also about filling in the gaps by converting “unknowns" to “potentials.” Yes, it is true that modeling is statistical guess work, but I’d take “70% chance of a snow shower tomorrow morning” over not knowing anything at all. What is a model? There are many different techniques of modeling but, basically, they are algorithms composed of various predictors, producing answers in forms of model scores. Formats of the resultant scores may vary depending on the design. Sometimes they are just numbers with many decimal places, but kind modelers would generally convert them into score groups for end-users. Sometimes, lower numbers are better (as in one out of 20 model groups is the best), but other times, the higher number is better (nine out of zero to nine possible grades, for example). Modelers may further simplify them by converting them into simple yes/no answers, drawing the cutoff lines for you. Imagine movie recommendation scores by streaming services. Sometimes they express them in a percentage scale, a five-star system or just yes/no answers. Regardless of the methodologies or formats, models compress an enormous amount of information into simple scores. In the age of information overload, that is the most notable benefit of modeling. Simple answers are what modern decision makers need. Even machines work better and faster with compact pieces of information, as the machines will have to work extra hard to utilize raw data at the time of decision-making. Further, model scores can be input variables to other models. Excuses for not modeling If modeling is such a great thing, then why doesn’t everyone employ modeling techniques all the time? Here are a few common reasons: The universe is too small, especially in b-to-b marketing. But don’t forget that the number of variables is increasing at an alarming rate, even if the size of your targeting universe may be small. Even for 100,000 sales prospects, you would need help from machines to plow through over a hundred attributes attached to each of them. Consumer data often include over 1,000 variables for each individual target.Predictable data is not available: This could have been a good excuse in the last century, but not anymore. The real excuse would be that collected data are in such bad shape that they can’t easily be utilized. Figuring out the patterns in unorganized or unstructured data is a job for machine learning. 1:1 marketing is not in the plan: That’s not good, as every channel is, or soon will be, a 1:1 channel (i.e. an interactive channel). In 1:1 marketing, you must know who to engage and what to offer through what channel. To answer these questions, you would need targeting models and various propensity or look-alike models. A tight budget and lack of resources: Advanced analytics requires high level resources. But you don’t have to build a fancy analytics team internally right away. There are many outsourcing places for modeling and, if you can’t afford that either, check with third-party data providers, as they maintain many off-the-shelf models featuring various personas, propensity models and behavioral models. In data-rich environments, all decision makers must be aware of modeling options and, at the very least, benefits of modeling. Not everyone has to be an expert level data analyst, but lack of understanding shouldn’t be an excuse for avoiding advanced analytics. After all, the difficult mathematical parts will be done by the professionals and machines. Just ask them the right questions.