Customer acquisition is the process of acquiring new customers for business, or converting existing prospect into new clients. This could involve finding customers who previously were not aware of the company product, were not candidates for purchasing your product, or customers who in the past has bought from competitors. Some of these customers might have been old customers, which could be an advantage since more data might be available to them. This could also be a disadvantage since they may carry some bad experience as a result of poor quality of service or product. Importance of client acquisition varies according to the specific business situation of an organization.
This process is specifically concerned with issues like acquiring customers at less cost, acquiring as many customers as possible, acquiring customers who are indigenous and business oriented, acquiring customers who utilize newer business channels etc. Once the prospect makes first purchase he becomes the customer and that is where customer acquisition ends. This includes follow-up non-purchase encounters. Subsequently, customer may request a service or other activities that would strengthen chances of him making a repeat purchase and that further increases customer acquisition costs.
Customers acquisition only makes sense if their future customer potential, or discounted future value exceeds his acquisition costs, and sometimes in order to acquire more customers companies broaden acquisition effort which may have a negative effect in lower response rates of their acquisition campaigns. Sometime acquisition strategies are two-fold. Company may want to increase awareness through some sort of mass-communication channel offering free-trials, or introductory offers on their products or services, or they may work on brand positioning differentiating themselves from competition offering specific expectations. End goal of either strategy is to provide product/service experience at the right price which would lead into increases satisfaction which is emotional category that would most likely lead into repeat purchase. Part of successful acquisition efforts is right pricing which can be “penetration pricing” – price low to acquire customers and raise prices later, or “introductory” pricing that is given to a specific market segment and then taken away once their retention figures become steady.
The traditional approach to customer acquisition is the most likely combination of mass marketing and direct campaigns based on customer’s knowledge of the particular customer segment that was being targeted. So, typical scenario would involve the purchase of third-party data that meets overriding characteristics of target segment based on age, gender, affinities, marital status together with some risk profile characteristics. Obviously this data would include target segment contact details, such as addresses, emails, or phone numbers. Since names and addresses are of limited value for data mining one can try to overlay this data with census information that contains average income for that area, age, etc. Shallowness of data is the main problem with customer acquisition, so great effort is usually spend on overlays and enrichment of third-party data. There are no such problems for data mining applications for existing customer since you have an existing relationship in place and subsequent wealth of affinity and purchasing behaviour information.
After constructing of customer acquisition data-set, marketing campaign (test campaign) would then go out with appropriate marketing message that would hopefully entice the prospect to try out a new product or services. Main purpose of this initial “test” campaign is to track and collect response measurements which are needed in order to create acquisition predictive model. As the minimum there are two basic binary response behaviors – where customer response is either a “yes” or “no”. Also, there could be a variation of positive response behavior. E.g customer may not just purchase the product, but also sign up for additional services, insurance, etc. In which case more than one variation of target variable can be constructed. There could also be response in terms of a purchase of a product different originally offered.
This response measurement is critical in a future acquisition efforts. There could be some uncertainties with non-response. Sometimes it is not clear if non-response is flat-out rejection or simply due to the fact that recipient may never receive marketing solicitation due to change of address, or because marketers used wrong marketing channel for specific customer It is important to say that better model can be built if one can distinguish between rejections and non-responses but that that is not always possible. There are several ways of dealing with non-responses! If it is not clear that is rejection – one can drop it out from modeling data set. Problem is that this approach can throw a lot of potentially valuable data. Another approach is to turn it into flat rejections – and problem here that this can carry unknown level of bias into the model most likely resulting in potential misses of likely responders. Last approach is to make non-responses missing values and then using some sophisticated imputation methods to infer or predict if non-responder would reject or accept the offer had he seen it.
Once the data have been prepared consisting of “input” variables which are mostly demographic with some “overlay” censors data, and the “output” or “target variable” with binary values (0=rejection, 1=response) modeling can proceed. Modeling is simply logical or mathematical expression of patterns in data that lead to one or the other levels of target variables. If all responders to acquisition campaign happen to be of male gender and all females are non-responders then simple “rule-based” statement such as “if gender=”males” than response=”yes” can be used to score new data and it would be no surprise that all males would be given highest response score and all females lowest.
This is a gross simplification of how modeling works – but in essence modeling techniques attempt to do just that – to find certain values or certain variables that the best differentiate between different levels of target variables. And it is obvious how this approach can generate the value; out of new customers data model will quickly find those who look like previous responders giving them high likelihood of response and to those who have attributes similar to historical non-responders probabilities of response will be low. And of course - there is always going to be those in the grey area who have mixed characteristics and for whom allocated probabilities will simply mean – “not too sure”. And so, next acquisition campaign will have those with high chance of response and those with low probability will be dropped.