Identify distinct groups of customers and how they behave with your products / services to help you better tailor your marketing mix 

You know from experience that your customers behave in various ways ...
but could you define and describe the number and the distinct characteristics of these types of customers in terms of how they relate to your services or products ?

ICORE is an engine that integrates different techniques (described here) to help you make the right product offer to your customer


Customer behavioural segmentation will analyse your customer data and come up with a « typology » of your customers based on their transactions, contacts and/or all other behavioural information you have about your customers.

The successful typology will help you decide :

  • On what type of customers the focus should be
  • What marketing mix and message would be best for a specific type.
  • How broad your focus should be
  • In which competitive space is it most profitable to operate

Here is an example of a common typology based on a very wide set of information about the customers (recency & frequency of purchase, spending amount, number of products, number of contacts, evolution of spendings over time):
 The Cream of the Crop

The Ideal customer spending a lot on your products or services

168.307 customers                                412 €

 The new wave

The New customer showing signs of a « Cream of the Crop to be »

105.562 customers                                  228 €

 The Informed Customer

The customer that requires a lot of communication …information greedy

27.637 Customers                                    207 €

 The Average customer

The Customer with no peculiar characteristicwith a very common behaviour

250.321 customers                                  202 €

 Total Customer base

1,158,893  customers 

 average spending: 178

 The new recruit

The New customer  with lower spendings

230.422  customers                                135 €

 The Runaway Customer

The Customer spending less and less over time.. About to leave you or fallen asleep ?

150.118  customers                                   88 €

 The Sleeper

Inactive customer, buying very unfrequently and for small amounts

125.606 customers                                    84 €

 The Bad Recruit

A lot of complaints, returns, low spendings

38.517  customers                                    71 €


Every customer is assigned to a specific group of customers who all tend to be similar on the mutliple measures chosen for the Analysis.

Projection of customers on a 3D plot of TENURE- RECENCY- SPENDING

We can see, on this 3D graph, how 4 types of customers are represented. each dot represents a customer. The 3 axes chosen here are: average spending, tenure (length of relationship with the company) & recency (number of days since last purchase which is the the axis forming the "depth" of the cube).

We can see how the 4 distinct groups of customers are represented as clouds of customers. The red one (ideal customer) is clearly higher on SPENDING. The green cloud is the sleeper group, lower on spending and higher on recency.

The blue cloud represent the new recruits with low length of relationship (tenure)

The white cloud represents the average customers; customers « in the middle »

The analysis will assign a colour (group) to each customer based on their specific measures and how close they are from each core profile on each measure chosen by the Managers.


Looking at the same 3D plot from a different angle, We can see how the green cloud (the sleeper group), is higher on recency and also further away on the length of relationship axis. 

The blue cloud represents customers with a lower length of relationship (on the left of the graph) and a lower recency than the other groups as it is placed lower on the recency axis 


The typology will reflect the choices of customer information (measures) to be used.

The measures needed are discussed & selected together by the marketing managers & the analyst in function of their strategy & of course limited to available info on the customer. An Initial workshop will precisely define what the marketing managers want to analyse with the segmentation… the type of info selected will have a big impact on the final typology.

You can focus the segmentation purely on transactions (recency , frequency, monetary value) or include a more holistic view of your customers behaviour (contact channels, type of products bought, discount based purchase… ).

The analysis will separate your customer base into distinct groups of customers behaving in the same way. Each group is specific and distinct from the others, all your customers will be assigned to one of these groups in function of their behaviour.

The main insight will be a summary map of your customer base (see table above) 

Each segment is then described individually, a good understanding of the specific characteristics of the segments individually will allow to set-up bespoke Marketing actions that will be more efficient as they will be based on a true understanding of customer types.

 
 

The method used strives at a 3-fold objective:

  • Make sense to the business (it must tell a business story in order to enthuse Marketing Managers) 
  • Show homogeneous clusters, customers that look very similar inside the same group but clearly distinct from other groups
  • Show sizeable clusters (allowing a roll out to a full marketing campaign)


Identify popular product baskets to help you optimize store/catalog design & and make successful bundled offers

Association Analysis can help you find the most popular product baskets which regroup products/services commonly bought together.
The results of this analysis gives you a list of preferred « product baskets » among all possible combinations of « baskets » that have been bought by your customers over a defined period. These baskets are also called RULES, and are specified as this:

Product A => Product B (Product A triggers or influence sales of Product B), this is a rule (basket) containing A & B.

You might have a feeling that items in your store or in your catalog could be rearranged and specific products should appear closer to each other in order to leverage sales.
For example, sales of product B are stagnating, knowing from the association analysis that product B is often bought together with A and C, it would be a good idea to place B in the vicinity of A & C in your store in order to leverage sales for B.
You can also imagine to make bundle offers on B: « buy A and B and get 25% off the price for B » .
The same principle can be applied to the laying out of products on a website or in a catalog.

This knowledge of popular baskets can also help you make a better use of every contact with your customers by offering them the product they haven't bought yet wich they are more likely to buy…
Offering a relevant product to your customers according to popular product associations will maximize your chances to increase sales.
I have also taken part in a project that applied Association Analysis results to boost sales on inbound calls. The Association Analysis can also help identify GAP offers, which are « missing » products in a specific customer list of purchases to make it a complete popular basket.
For example Jean Dupont has bought A & C but still hasn’t bought B, Next time he visits your store, or calls you, ou can promote product B to him knowing that his propensity to buy B is supposedly greater than for another product since he has bought A & C already.

We can also say that GAP offers are a good hook on the customer to maintain the relationship and act as an ideal X-Sell platform by enhancing awareness of other products.

ICORE graphical view of association between products

This graph shows the association between products such as analyzed from the Customer base. This shows the relationship between products in a "network" like fashion where products bought together are close to one another. The size of each circle gives an idea of the penetration rate of the product (% of customers who bought the product) within the customer base. On hovering above the circle the penetration rate appears. The figure shown on the arrow is another parameter indicating the strength of the relationship between two products (the lift) 
ICORE graphical visualization of product associations


Here is an illustration of visualization of the results of Association Analysis for grocery products : (sources :"Visualizing Association Rules: Introduction to the R-extension Package arulesViz", Michael Hahsler & Sudheer Chelluboina from Southern Methodist University)


Visualization of association analysis prefered baskets (selection of 10 baskets)

Here is a selection of 10 preferred baskets :

The larger the circle the stronger the basket in terms of its « preference » among customers.
For example one of the identified basket (on the left of the graph) is : « flour & baking powder trigger sales of sugar ».

Two other baskets are :

« ham triggers the purchase of white bread » and this basket is more strongly "Associated" or prefered than the previous one as the circle is larger.

« processed cheese triggers white bread » but this basket is not as strong.


Enbaling you to better negotiate with your providers by knowing which are your sales driving products and how they influence sales of other products

What are your sales driving products?
What are the products that influence sales of other products ? 
In your negotiation with your providers,You can use a guiding map of your products sales driving force to help you optimize your margins/stocks.

Association Analysis can also be used to identify your « profit drivers » products. You might of course already know most of them through sheer experience but you might discover other very interesting assoication between your products. These insights can give you a very useful source of information and guidance when it comes to negotiating prices and stocks with your suppliers. You can focus your negotiating power towards lowering the price of the low driving potential products as you will know they are not as crucial to boost your sales on other products and by doing so increase their margin. The Association Analysis requires to access individual transactional data.

Association Analysis for describing the strength of sales driving products (LHS) in % of sales of associated products (RHS)

This graph shows another powerful insight of Association Analysis.

Each (RHS) product associated with a driver product in a rule is projected on this graph, however the symbol on the graph represents the driver product (LHS) of the rule ( in order to compare driving products on the graphs).

The horizontal axis shows the % margin made on a sale of the associated product (LHS).
The vertical axis shows the sales figures for the associated product made jointly with the driver product.

The most desirable quadrant is the upper right one were you get high margin product with their sales largely associated (>50%) with a driver product.

For example if we look at this upper-right quadrant, we see that product G and AA drive the majority of sales of two high margin product (margin >50% on the horizontal axis, and % of sales of the associated products > 50% on the vertical one).

Product T is the driver of several products but doesn’t seem to capture a majority of their sales.

Product M drives only 1 product but it captures more than 80% of the sales of that associated product.


Enabling you to preempt any decision from your customers to stop using your products or services

You would like to be able to anticipate any decision from your customer to stop using your products or services...and convince them to keep buying from you
your problem is that you dont know who is likely to stop their relationship with you.

Your past behaviour is a good indicator of what is likely to be your next move … this is the funding principle of predictive modelling…You don’t have a clue who in your customer base is about to leave you soon…but you can give it a prediction.

If you want to know who, among your customers, is more likely to stop using your products or services in the near future, I can help you predict this for every customer in your database using information collected in your systems about each customer’s purchase history & other type of data you could have about your customers. This purchase history could also be combined with other types of data for each customer (Socio-demo, field of activity, geographic) in order to further improve on the quality of the prediction.


Overview of the approach

Retro-analyzing your history database (looking back into past customer data), the Analysis will define and identify these customers who, actually decided not to use your services/products anymore. The Analysis will compare their behaviour with loyal customers behaviour & profile, the analysis will then come up with a statistical model giving for each customer an estimated  %  chances, NOW, that the customer will decide to stop using your services in the near future.

You will then be able to make an intelligent guess of who among your customers might not be purchasing from you  in a decided timeframe. That will allow you  to focus your marketing efforts on these customers who most need it. You can then avoid to spend money on marketing initatives for customers who would not need it.





Enabling you to leverage ROI of your Direct Marketing Campaign

The predictive modelling approach is applicable to different purposes: trying to predict for each customer in a defined population the probabibility to be interested by a new product (X-sell), trying to predict the percentage  of chances that a customer doesn't pay off his debt, and in fact any event you may think of interest as long as this event is measurable at customer level in your history database since we are using the past to predict the near future. 

Enabling you to acquire new profitable customers

You want to grow your customer base ?
You want to be able to spot and profile your best customers in order to target most look-a-like profiles outside your current customer base…

The  aim of the Analysis is to allow the Marketing managers to set up a Direct Marketing acquisition campaign which would target customers with a specific desired profile (matching the acquisition strategy pursued by the business).

First, the typical customer to be acquired will be defined within your current customer base using Socio-demographic data (in your database or bought to Data Providers). This first descriptive analysis of customer profiles can already be very insightful.
Second, a regression analysis will be done using socio-demographic data (third party data) and give a typical "ressemblance" profile with the "ideal" customer to be acquired for a particular acquisition campaign.
Third, the DM campaign will be based on this "look-a-like" profile produced by the Analysis in step 2 by focusing the campaign only on Customers outside your customer base that show a similar profile to the "ideal" profile and by doing so maximize the chances to look like the typical customers you are targeting in the campaign. This third step requires you to "buy" from Data vendors data needed to score customers outside your base, a pragmatic method can help you minimize the costs of new data acquisition by using just the right amount of data for the analysis and a savvy approach .

The first step will typically give you a Socio-demographic Map where "typical" customers for each of your product will be projected.In this illustrative example, we have 3 models of car (A,C,X) projected among socio-demographic attributes like:
age, job status, gender,education, housing. the products (car models) are located on the map according to the "average" location of the customers who own them. Analysing this map can already give you useful insight about who is buying what
and give you some ideas on adapted above the line Marketing campaigns, promoting one or the other product. for example, Projecting the type of prefered media or magazine can help you direct your campaign in the right media or place your commercial boards
in better adapted areas.   

The second step, the regression (scoring) analysis, will give you a "look-a-like" score to a "likely buyer". Using this score you can better predict who has got the profile closest to a "likely buyer" profile for the Acquisition Campaign at stake. For example, if you want to acquire new customers to buy car model A.

You will produce a model giving you the "best" socio-demo profile of the typical customer for A, hence maximizing the response rate of new potential buyers for model A. Depending on the quality of the scoring process (the strength of your prediction which depends mainly on the quality of predictive measures used, implying the capacity of socio-demographic attributes to predict the ideal profile for the campaign) you will increase your efficiency compared to a "blind" or "naive" approach such as illustrated in the graph hereunder.

The horizontal axis represents all the customers available for the campaign (all adult people living in an area, city, country)  rank ordered from left to right by their descending score of "ressemblance". The vertical axis represents the expected response rate to the campaign
(in this case it is the likely buyers hit ratio).
The Orange middle line is your expected response rate at different possible cut-off percentages of the available population selected for the campaign.
The blue line is the response rate you would get if you would pick at random customers to be selected for the campaign. the blue line lies at a 1.5% (theoretical) response rate meaning that we expect the total percentage of new ideal customers for A to be 1.5% in the total population.

The objective of the graph hereunder is to give an illustrative  example (realistic but still theoretical) showing how much we can optimize the selection of customers for the acquisition campaign by selecting the top scores and so reduce the amount of "wasted" contacts to customers who wouldn't fit the "likely buyer" profile as much.
The gap between the blue line and the orange line is determined by the strength of the scoring to actually identify likely buyers of A. This of course will vary from campaign to campaign, influenced by various factors (quality of predictive socio-demo measure used, type of target profile definition, macro-economic & competitive context).

If we decide to mail the top 20% scores "ressembling" the most to the typical buyer of A in a population of 250 000 people (say we had built the analsyis using the population of a city of 250 000 adults), meaning you would mail the 50 000 best "scores" customers from the population,  you would then expect to hit about 1875 likely buyers (3.75% of 50 000 selected by the model).
 
If you would take a "naive" approach you would expect to hit only 750 likely buyers (1.5% of 50 000 people selected "at random") with the same campaign size.  That is a difference of 1125 likely buyers not hit if you dont use Analytics. Depending on the conversion rate (actual sales rate) you can expect from likely buyers and the margin on model A, these missed 1125 likely buyers can represent a substantial missed sales opportunity.