|
|
| [Home] [Projects] [Contact] |
|
|
|
|
|
Prospecting strategy: turning marginal lists into profitable ones... Towards your business objective: targeting the right customers... What works and what doesn't: causation vs. correlation... Market basket analysis: order starters and predictive filtering... Improved focus: customer segmentations... Matchback audit: fractional allocation ... Contact strategy optimization: putting it all together... Project descriptions A zip code penetration model will identify zips with the highest prospecting potential. They are among the easiest models to deploy. Demographic and lifestyle data available for both house customers and prospects is used to develop these models. Current customer penetration is computed by zip code, which are carefully weighted in the modeling process to reflect mailing intensity and responsiveness. Zip code models are powerful prospect selection tools on a list-by-list level, turning marginal lists into profitable ones. Prospecting into top-ranking zip codes is like preaching, if not to, at least among the converted. A predictive model ties precise marketing decisions to a specific business objective. The business objective could be increasing prospect response within store retail trade areas, maximizing a campaign’s total revenue or up/cross-selling opportunities, identifying customers with a high risk of attrition, determining paths to higher lifetime value by customer segments, identifying best customer look-alikes among prospects: the list goes on... Experimental design is the gold standard in establishing cause and effect. Marketing tests going beyond the occasional hold-out panel or A/B exercise aren’t as popular as they should be. Multivariable testing can provide considerably more insight than an A/B test of the same size. And not only do multivariable tests establish causality and guide marketing knowledge discovery, they are also the foundation of incremental response modeling, a predictive modeling technique also known as uplift modeling. Multivariable designs must be developed carefully to avoid inconclusive results, post-test salvage operations or wasteful cell sizings. The next best thing to a designed marketing test is the observational study, where, for example, emailed records are retrospectively matched to records that weren't emailed in order to study the impact of the email campaign on, say, web buying behavior. Market basket analysis identifies patterns in customers’ purchasing habits. This analysis is particularly powerful when tied to historical data. It helps answer questions such as what longer term inference can be drawn from the content of a new-to-file’s first basket. Does buying item A and B together suggest long-term growth potential? Or is this an indication that the customer is unlikely to ever buy again? A large and complex customer base is more tractable when divided into a few well differentiated persona segments. A customer segmentation (also known as cluster analysis) is necessary to define and test marketing actions. It is often a preliminary step in analytical projects. Segmentations are developed using merchandise and channel affinity, recency status (whether active or lapsed), amount of repeat activity and price point information among other customer behavior data. They are very well received by clients: a good segmentation is intuitive, provides valuable insights and drives a high degree of confidence in the associated recommendations. Matchback Analysis and Fractional Allocation What is the impact of catalog mailings on web demand? Full-funnel attribution modeling is definitely an interesting topic, particularly given today’s complex tangle of channels. However, properly estimating the percentage of total web demand actually driven by catalog mailings over time is definitely the lower hanging fruit. The “last contact takes all” allocation methodology does indeed simplify processing, but can be misleading for marketers at circulation planning time. Furthermore, fractional allocation doesn’t need to be highly sophisticated to be accurate. More importantly, it relies on neither complex business rules nor arbitrary customer behavior assumptions. As I’ve explained to several marketers and service bureaus, fractional allocation is the simple application of weights derived from careful analysis of promotion history and synergies between sourced and unsourced demand by channel over time. Ideally, these weights are calibrated using hold-out panels, but the methodology also provides valuable results without them. There is no need to agonize over diminishing keycode capture rates, or force web buyers to key in a source code at check-out. I can provide these weights, and walk the mailer or service bureau through the programming necessary to deploy this straightforward attribution approach. Given the upward trend in mailing costs, marketers are moving away from RFM towards modeled contact strategies. A contact strategy optimization project will provide mailing intensity recommendations by customer segments. Contact strategy optimization encompasses geographic modeling (such as zip models), response modeling (probability of response along with expected spend per customer), customer segmentation (focus on cluster homogeneity to provide focused contact recommendations) and test design (how should the recommended, say, 10 catalogs mailing be spread over the next 12 months for a given segment, or should a new booksize and different spread be considered). In addition to methodically building each stage of the project, careful thought must be given to the deployment of the recommendations. This is why I map the recommended name selection rules to prior selection criteria to instill confidence in the new contact intensity recommendations. Doing so simplifies the client’s tasks of circulation planning and forecasting. Most projects are completed between 1 and 4 weeks after the necessary data is made available. All projects are self-contained, unless I am asked to take on production assignments such as scoring models on a monthly basis. Project deliverables include:
In addition, I can complete a detailed write-up of my analyses, and
|
|||
|
|
| [Home] [Projects] [Contact] |