by Richard Boire, Senior VP, Environics Analytics
In the world of Big Data and Artificial Intelligence, we are all aware of the tremendous hype around these themes, some of it arguably very exciting and relevant but some of it bordering on the excessive. This excessiveness is sometimes demonstrated by certain opinions regarding the notion that older analytics methodologies and approaches no longer apply as we embrace more automation and more advanced forms of analytics such as artificial intelligence and specifically deep learning.
Why is this? The simple answer is human nature and the fact that people and organizations no longer want to be considered as Luddites when it comes to new technologies. But we often forget that most business problems can be solved using simple analytics and not involving the latest Big Data tool or deep learning algorithm.
There is no question that organizations are right to pursue these new technologies on their business. But do organizations explore these technologies at the expense of utilizing simple solutions that can be produced quicker, thereby solving more of the pressing business problems and issues?
Many of these so-called “simple” solutions yield tremendous benefits because they are utilized in situations or business scenarios where no analytics has been done. Seasoned analytics practitioners, including data scientists, would agree that successful analytics solutions developed for a first-time business problem will yield tremendous benefits, particularly if there is no prior solution.
The intellectual challenge for practitioners is to attempt to identify business situations and problems that can accrue those type of huge gains while using simple analytics solutions.
What do we mean by simple? Simplicity within the analytics context comprises two criteria:
- Ease in creation of analytical file, and
- Use of simple business rules or algorithms in the development of solution.
In creating a simple analytical file, the use of perhaps two files, a customer file and a purchase file, are typically all that is required as source files. Our objective here is to extract simple structured data and minimize our efforts in extracting semi-structured and unstructured information. From this data, we can then create the necessary inputs whether it is a targeting tool such as RFM or a model, or the generation of key business reports.
It is important to remember that 80%-90% of the analyst’s time is spent creating the analytical file. The remaining 10%-20% is spent developing the solution. Once again, the approach to developing these solutions may be straightforward business rules or predictive models using traditional machine learning techniques. Unless there is automated machine learning software that can encompass the use of deep learning algorithms, the more traditional type approaches will be used if simplicity is an objective.
Let’s take a look at some practical examples of simple solutions in practice.
For many organizations, the ability to target the right customers remains the No.1 analytics and data science problem. Yet, simple RFM techniques can accrue huge gains as we create an overall customer index based on recency of activity, frequency of activity, and amount of activity. Note the word activity can mean many things depending on our business objective. If our objective is profitability, then in many cases it can simply relate to purchase activity. Yet, if it is customer engagement, it could be activity on a website such as recency of last click, # of times clicked to that website and the average duration or time spent on that web site.
As a part-time professor at several colleges within the Toronto area, I often mention to my students that one of the first questions they should ask from their new employer is whether or not the organization has a best customer program. Can the organization identify its best customers, as well as those non-best customers who look like best customers? These may seem like simple initiatives but simplicity often gets overlooked especially when a more complex challenge dealing with the latest technology flavor of the day becomes the latest marketing initiative.
Let’s take a look at a simple decile report ranked by some predetermined measure of value which reveals the following:
|% of Customers Ranked by Value||Average Customer Value||% of Total Value Captured in Interval.|
In the value decile report above, we have identified the top 30% as being our best customers. Each decile within the top 30% contributes over 10% of the total value from the entire customer base with the remaining deciles (70%) contributing 10% or less. The ability to produce a report like this can motivate marketers to initiate a program where in effect different deciles or segments within this report can be tested within a campaign. Both push and pull type marketing campaigns can be generated with specific initiatives and activities based on whether someone is a best customer, looks like a best customer or is not in any of the aforementioned groups. As disciplined analytics practitioners, these activities against certain segments would be continually evaluated and adjusted accordingly.
Within the marketing world, there are ultimately three core objectives as it relates to customer management: acquisition, migration, and retention of customers. Does marketing know where to prioritize its initiatives?
The common marketing refrain is that we focus on all three. But simple analysis may indicate that there one objective should be the priority. For example, a simple KPI report might reveal the following:
|Period 1||Period 2||Period 3||Period 4|
|% of active customers that increased their spend||25%||22%||19%||15%|
|% of active customers who have cancelled or defected.||2%||4%||7%||10%|
In this simple report above, clearly there are migration (increase in spend) and defection problems that may be stemming from the same issue. But the larger problem here would be defection which has increased fivefold over 4 periods. The great utility of KPI reports is not to solve problems but rather to identify problem areas that need investigation. In this case, the analytics would need to probe deeper into why defection has increased dramatically.
Another very useful, albeit simple, report is the customer cohort report, which tracks customer segments and their behaviour over a period of time. Let’s take a look at a report which tracks new customers at different time periods.
|New Customers 3 years ago||Year 1||Year 2||Year 3||New Customers 2 years ago||Year 1||Year 2||New Customers 1 year ago||Year 1|
|Retention Rate||50%||40%||35%||Retention Rate||48%||41%||Retention Rate||70%|
In this report, it is clear that the new customers from a year ago are exhibiting different behaviours than the other new customer cohort groups. This type of insight would then warrant more analytics on what is causing both increased retention, yet reduced spending for this cohort group within their first year as customers.
In many organizations, the issue of customer retention is usually a corporate priority where organizations will be willing to devote resources in the development of predictive analytics solutions.
These solutions, in effect, would be targeting high-value customers who are most likely to defect. Using two tables or files (the customer file and a transaction/purchase history file), powerful models can be developed without even venturing into the social media ether. This would enable marketers to target this high-risk high-value group which would involve differing strategies towards different risk groups. The use of social media information alongside other external information could be used to augment our communication strategies towards these groups.
There are just some of the many examples of how simple analytics can be used within an organization. In all these exercises, the common theme is simplicity in arriving at a given solution. Despite the fact that in some cases sub-optimal solutions can be produced, the fact that we can develop more analytics solutions in effect yields larger benefits overall to the organization. The challenge going forward for practitioners is when to apply a simple solution versus a more complex solution and what are the trade-offs – something that is not often discussed by the consulting experts.