business intelligence in a digital world

In contrast to the pre-digital world, we’re now drowning in data, creating an increasing need for business intelligence to guide effective decision-making.

Success in today’s technology driven world directly correlates to the quantity and quality of information possessed… that of the client, the competitor and the market. Informed decision-making ultimately leads to greater access, opportunities and technological advancement. The difference between renewable success and ultimate failure is a fine line that most ambitious businesses are willing walk. Very often, the most important factor in determining success comes down to who engages in factual and informed decision-making and who follows hunches and suppositions. [source]

Why business intelligence?

In marketing, we used to emphasize doing market research — a periodic activity to better understand our customers/ clients and how to satisfy their changing needs. Marketing intelligence got short-changed in discussions (and in textbooks) mainly because we didn’t have many sources of information and those sources didn’t provide much in the way of insights.

Mostly, business intelligence in the marketing area consisted of forecasting and maybe a little pricing. We often just drew a trend line over our sales graph, and maybe one over our expenses and called it a forecast.

Everything is different in a digital world and the amount of data available for analysis increases exponentially with the addition of data from IoT devices that transmit data about everything from customer behavior to traffic patterns.

Using business intelligence as a marketing tool provides insights that put you well ahead of your competition.

IBM contends 4 Vs surround the problem of Big Data — which increases by 20% per year:

  • Volume — currently the data equivalent of around 47 YEARS of HDTV programming exists
  • Variety — about 80% of that data is unstructured (non-numeric) including images, video, and text. Analyzing this data is challenging, even with powerful data handling tools like Hadoop. It comes in a ton of formats from thousands of sources.
  • Velocity — it’s coming hot and heavy, especially from IoT devices.
  • Veracity — actually, I added this to IBM’s 3 Vs because you’re never quite sure if the data you’re getting is accurate, so cleaning data is an important first step to analysis.

Making business intelligence work for marketing

Much of what you read on business intelligence focuses on configuring data and extracting it from databases using SQL, Hadoop, Python, R, Excel, or any one of a growing list of tools to handle big data. These tools troll through massive databases to quickly return data associations based on user queries. Let’s call this BI engineering, which is common in the field.

Very cool.

That way, if you want to know what other products a particular consumer bought or looked at since becoming a customer — bam, you’ve got it. This is the basis for user recommendations that create tremendous customer value when using Amazon or Netflix. Instead of searching thousands of items, you get great suggestions that both increase purchases that benefit companies while helping customers find things they’ll like.

Check out the demographics of people who bought XYZ using a BI tool and you quickly get information to help you market your product more effectively.

Also VERY cool.

Of course, there’s also a creepy side of business intelligence for consumers — that companies track your every move, every search, every click.

In today’s economy, information is power!!!!

Dashboards in business intelligence

Increasingly, firms use interactive dashboards providing trend analysis along with user customization options. Google Analytics is an example of one such dashboard for, while it provides detailed data correlations that commonly affect business decisions, it also provides options to customize queries.

Maybe an example will help.

I can use any of the standard Google Analytics views — they call them widgets — to answer questions like, how many visitors to my site came from Twitter? or How many people who bought a product came through AdWords? By tagging my content properly, I can even discover the ROI of each social network.

Goole Analytics allows me to put together a dashboard of my favorite widgets. I can even create different dashboards to answer different types of questions.

Here’s a performance dashboard I created for one of my clients:

business intelligence

I collected reports that I find help guide our decision-making all in one place.

We can see how successful we are in meeting goals, which content contributes to traffic, any performance issues, such as slow load time and difference across countries.

If I want to dig a little deeper, I can choose a second dimension, such as source, demographics, interests, etc to see how multiple factors determine performance.

I use multiple dashboards for different aspects of my client’s website.

I can even turn clients loose to set up their own dashboards or manipulate parameters on the existing ones to answer their own questions without having to reprogram. These can be done ad hoc or set up for inclusion in the dashboard moving forward.

Going beyond dashboards in BI

To gain true value from business intelligence, you need to move beyond simple correlation with your BI function and that requires someone who’s a subject matter expert, possibly paired with a BI engineer. Several experts say 70% of firms don’t have adequate data analytics capability to efficiently manage marketing. But, as challenging as it is to even get BI engineering right, stepping up to model building, which combines marketing concepts with data, is the stuff of only the top marketing firms.

You see, there are limitations and problems with using dashboards.

You can only view correlations between 2 variables

Correlation ISN’T causation. We call these spurious correlations when the variables correlated actually have no logic. The typical example of this is the correlation between the length of women’s skirts and the economy or the coincidence that led to an athlete doing things or wearing things for important competitions. Thus, shortening women’s skirts has NO effect on the economy.

Sometimes data is just plain wrong. For instance, if you’re trying to figure out who’s buying products in your store using loyalty card information to match characteristics of buyer with products, you get totally invalid data when I let a friend use my loyalty card or pick up items for the school bake sale.

Fixing these problems may prove impossible, but modeling helps provide more nuanced insights to guide decision-making.

What is modeling?

Modeling involves using established relationships among several variables or creating them based on theory. Econometrics is one way of building models based on financial information. Some marketers used these as building blocks for marketing models. This sometimes creates a problem when you assume that a particular financial variable is a good surrogate for a behavioral or cognitive variable. For instance, assuming purchase implies preference for a brand isn’t always true. It may be that the brand is cheaper or readily available and preference will change if either changes.

Marketers are getting better at building their own models based on sound marketing theory. The problem is, these models aren’t making their way into the decisions made by practitioners. That’s sad.

For instance, I developed a model to explain why patients don’t follow the directions of their physicians and another to explain why it’s so hard for people to make changes they know are in their best interest, like losing weight. Neither of these models are used in practice today.

Building models

Big data really helps in building new marketing models based on a company or industry.

Instead of looking at how a single variable affects an outcome variable of interest, a firm can now look at a number of variables to determine how (or if) each one impacts the outcome variable. But, don’t just throw all your variables into a soup, hoping something will emerge as life emerged from the primordial soup. You’ll end up with too many spurious correlations or overshadowing an important variable by another (when 2 variables are correlated all the relationship is assigned to the first, which may not be a controllable variable).

This is why model building is part art and part science. You can start by using stepwise regression to build a model, but go beyond that to consider whether this is the BEST model. Are there non-linear relationships that you didn’t model? Are some variables in the model nonsense? Do variables you think SHOULD be in the model get excluded?