The 10 key points of predictive marketing
By Didier Gaultier, Head of Coheris Datamining Business Unit
The success of a predictive marketing project relies on a rigorous process. Here are 10 steps to ensure the effectiveness and success of your project.
1) Access to transactional data
Transactional data (purchases) is the most reliable and important evidence in predictive marketing. Your past purchases can often predict much of what you will buy in the future. Whether these data come from an e-commerce platform, an ERP or a CRM system, they are the keystones in measuring what you want to get: sales.
2) Access to behavioral and "online" data about your prospects and customers.
Your prospects and customers are active. They also receive newsletters and SMS campaigns from you, they visualize your banners, they browse your website or e-commerce outlet, they interact with your call centers, and they make various queries. This invaluable information is certainly not only within your reach somewhere in your business premises, but is quite essential if you want to be able to anticipate the needs of your future customers, predict their behavior, and especially send them the best possible deals, at the right time, and using the most suitable relational channel. The most practical way to proceed at the first time is often to build a data warehouse that is capable of storing at least one copy of all data over a period of time that makes sense for your business.
3) Build a business Datamart
Having access to all transactional and behavioral data from such as a data warehouse is an excellent start, but would rapidly prove cumbersome and impractical in targeting daily campaigns or quickly interpret your customers’ behavior. In fact, information is generally not sufficiently aggregated here, and often the processing times of a datawarehouse are prohibitive.
It is then necessary to build an intermediary database through which you can conduct all sorts of statistical analysis without having to wait for days to see the results. The use of the sampling principle is also part of the rules of construction of the datamart. It should be sufficiently business-oriented so as to contain all relevant indicators for your business, which you will need frequently.
You should also ensure that your datamart is clean enough to be statistically usable (for instance, the elimination of adverse effects of outliers).
4) Use early segmentation
Segmentation is the basic and primary tool to allow the marketer to understand the behavior of its customers. In addition, a segmented database will add more accuracy to all predictive marketing methods that you set up later. They are fairly easy to implement with tools like Coheris SPAD and will save you valuable time later.
The minimum is one segmentation based on transactions and another based on the activity of the prospect and / or its attachment to the brand. But we should not stop there and also carry out a complete clustering of customers to understand how they are grouped logically into clusters with homogeneous behavior within each cluster. The segments will be the starting point for targeting of all of your marketing campaigns in the future.
5) Establish the profile of your typical customers
Within each cluster (following your segmentation), establish a profile of what your ideal customer looks like. It will correspond to the sketch of your ideal and most lucrative customer. This will then enable you to establish campaigns targets and scores.
The exercise is relatively easy to do once a complete typology of your customers has been conducted (see item 4 above), because it is located near the center of the cluster to which it belongs. Ideally it is recommended to define one or more indicators that let you know how far a particular client lies away from an ideal client profile for each of your customers, but this rule can also apply to most of your prospects throughout your database provided you know enough about them.
6) Using association rules to build your campaigns
Association rules is an extremely simple, fast and powerful datamining tool to build effective marketing campaigns, not just to get quick wins like up selling, cross selling, or cross canal. In particular you can identify purchases that are often the basis for a series of successive purchases, and therefore systematically promote this type of purchase in your basic campaigning (with a preferential promotional offer). You can then design marketing messages that will combine the so-called "core" products (or services) associated with their “attached” products and thus significantly increase the response rate of your campaigns, as opposed to campaigns built solely on your intuition or the status of your stocks.
7) Test run your campaigns
The drivers that make a campaign work or not can be extremely tenuous, but the effect on your business and your margins can be very significant. To achieve the best possible campaign, it is necessary to test several versions of creative and call to action, ideally two or three alternatives, in order to keep it simple and pragmatic.
You will be able not only to determine which alternative will give you the best results based on your goals, but in addition, and using datamining tools, you will be able to predict the response and conversion rates of your campaign on the whole database.
If a budget step is necessary before launching the campaign, it will let you know exactly what is needed in terms of investment according to the targeting that you will set, and most importantly, you will know in advance what will be your return on investment. The more any campaign is expensive, the more this exercise is essential to be reassured on one hand, but also to convince your management about the factual benefit of theses marketing campaigns. This is therefore a best practice.
8) Use the scoring to target your campaigns
From the moment you have completed all the previous steps successfully, establishing a scoring for each of your marketing campaigns is within a few clicks from you. A tool such as Coheris SPAD can help you establish this kind of scoring in less time than it takes to explain. Once the scoring done, you classify your target database by decreasing score, and you divide the target market by appetency areas.
High appetency area; proven appetency and profitable area, low appetency area, and churn zone. The next step consist of optimizing your campaigns according to your business objectives and constraints playing with the limits you have found in the different scoring areas.
9) Use the life cycle of your customers
Customers change and so do their needs. The "time" dimension is particularly important in datamining. Identify the different buying stages of your customers’ lifecycle and monitor their transition from one segment to another segment.
This knowledge will allow you to offer them something that fits exactly to their needs through the development of your relationship with them.
10) Manage your marketing pressure
The temptation when you get to this level of control in targeting your market is to multiply the messages and interact with your prospects and customers sometimes beyond what might seem reasonable. Not only should your communication remain acceptable to your potential customers, but in addition, it should be of quality and contain the feel of an important human dimension.
So multiplying automatic messages beyond the reason is not the optimal solution. Too much marketing pressure will only increase your churn, which has a very high cost (generally estimated at more than about 40 dollars per lost contact in B2C, and even more in B2B).
Churn management is generally undertaken using a churn score that may vary depending on your targets, your constraints and your kind of campaigns.
The key rule here is to never target in a too low appetency area, which would be damaging without creating any kind of value, and find the acceptable pressure for each segment of your database. The ideal state of mind is to systematically seek to enhance your database value (don’t forget that your database is one of your key assets) and bearing in mind what really matters to your customers.