SYNTASA DATA SCIENCE SERVICES A 3 : Advanced Attribution Analysis A Data Science Approach Joseph A. Marr, Ph.D. Oscar O. Olmedo, Ph.D. Kirk D. Borne, Ph.D. February 11, 2015 The content and the concepts presented in this document cannot be copied in full or in part for personal or commercial purposes without the permission of SYNTASA Corp. The content and the concepts of this document are of proprietary nature and owned by SYNTASA Corp.
WHY IS ADVANCED ATTRIBUTION ANALYSIS (A 3 ) IMPORTANT? When customers make web purchases it s important to understand what marketing channels drove that purchase and properly attribute purchase credit to them. This understanding enables subsequent advertising spend optimization. Typical attribution analysis utilizes last- click credit, first- click credit, even- click credit, split- credit, and other, arbitrarily chosen models of consumer response behavior. [1] But none of these models are effective at attributing purchase credit, because customer response behaviors vary per customer, per product category and also per industry. These response behaviors are not arbitrary! According to MediaPlex, "...true attribution is both custom and people intensive. You cannot apply a finite set of rules to every company. Nor can you 'guesstimate' the proper credit allocation with any degree of accuracy." [2] We agree. The traditional metrics are no longer sufficient and advanced insights are needed to truly understand what works, what doesn t, and how to improve ROI from marketing dollars. [3] Furthermore, mixed media modeling introduces complexities that render it impractical/impossible for day- to- day decision making. [1] The current choices for attribution analysis modeling therefore boil down to either inadequate, arbitrary models, or to hopelessly complex and impractical models. What s required instead is an attribution analysis framework that accounts specifically for customer, product and industry differences, and that can thereby deliver key, actionable insights from ecommerce data in an understandable way. Enter Syntasa s Advanced Attribution Analysis (A 3 ). WHAT ARE THE BENEFITS OF A 3? Syntasa s A 3 modeling methodology is different because A 3 models are based entirely upon the client s ecommerce data. Actual, measurable ecommerce data therefore drives the client s attribution model choice it is not arbitrary. The resulting A 3 model therefore reflects precisely the clients unique customer and product mixes, and, is situated within the client s specific ecommerce framework. A 3 models deliver a deeper understanding of marketing channel influences and consumer purchase preferences, thus providing marketing professionals with two key benefits: channel ranking and automated customer segmentation. CHANNEL RANKING The most important advertising channels that lead to website visitor conversion (here defined as product purchase) are ranked in importance over time, thereby revealing the highest impact (most influential) marketing channels. Marketers can then focus immediately on these channels. Channel attribution can therefore be calculated per customer, since each customer may arrive at their purchase decision via different channel touch sequences. The following figure displays the results of a channel ranking for two customers who arrive via different channel sequences:
It s clear from the figure that different advertising channels (labeled A through H in the figure) impact conversion differently, for different customers. Each advertising channel is therefore graded differently, based on actual ecommerce data and not on uninformed choice. That conclusion stands in stark contrast to conventional attribution models, which treat all advertising channels identically. For example, the popular last click model fails to differentiate between advertising channel effectiveness, but considers only the temporal placement of a channel: the channel immediately preceding the conversion event is assigned the majority of credit for the conversion. An arbitrary credit assignment is therefore imposed upon the chain of advertising channel touches preceding conversion. Those credit assignments are incorrect, but, this situation is regrettably all too common: In practice, the multiple touches a customer makes before a conversion are rarely taken into account when measuring campaign effectiveness across communication channels. [4] The channels that are the most effective at producing conversions should emerge from detailed analysis of a client s empirical website data [5]. Decisions about channel effectiveness should be made based on analysis of these data, and should inform attribution analysis. Advertising spends can then be adjusted accordingly, for maximum revenue impact.
AUTOMATED CUSTOMER SEGMENTATION That individual customers respond differently to advertising is obvious. And yet, attribution models in current use do not recognize this fact. All customers are treated equally by these models, and each customer is also assumed to respond equally to various advertising channels. These assumptions contradict the goal of personalization. If customer personalization is therefore the goal, then why are outdated, inapplicable and ineffective attribution models being used to pursue that goal? When our A 3 methodology is applied to a client s ecommerce stream, website visitors become segmented into groups automatically, ranked by the effectiveness of each advertising channel at producing converters: Each website visitor s advertising response behavior embodied in their actual advertising channel touch sequence is correlated with their purchase likelihood. Purchase credit can therefore be attributed across marketing channels in a rational, personalized way. As a result, Syntasa s A 3 methodology enhances personalization because attribution of purchase credit can occur at the level of the individual customer or at whatever level best fit our client s overall marketing strategy. HOW ARE A 3 MODELS CONSTRUCTED AND HOW ARE THEY USED? Syntasa employs advanced data science techniques and targeted numerical experimentation in concert with actual client website purchase data, to reveal the most effective advertising channels. Once revealed, the channels are ranked according to conversion effectiveness. Attribution profiles can therefore be personalized per visitor, and purchase credit assignment computed based on this individualized profile. When the contributions from all purchasers individual attribution profiles are aggregated, the resulting picture illustrates advertising impact across all marketing channels. Syntasa s A 3 methodology is general: It applies equally well to an organization, to a particular brand category, or to individual products. Furthermore, because our methodology is data driven, we can link attribution to actual purchase behaviors as they occur, and, update our model in real time to reflect changes in purchase circumstances, marketing campaigns and even the seasons. Our clients no longer need to remain shackled to outdated, ineffective and possibly misleading attribution model. Syntasa s A 3 methodology fits customized attribution models to our clients unique product mixes and market positions, and, can update these models as frequently as our clients wish.
REFERENCES [1]: Kaushik, A., Web Analytics 2.0, Wiley Publishing, 2010. [2]: Anthony, M., "Getting It Right: The Road to Genuine Marketing Attribution", MediaPlex, 2012. [3]: Latham, S., Observations on the Attribution Market, 7 July 2014, (http://www.attribution101.com/) [4]: Hongshuang, A.I., and Kannan, P.K., Attributing Conversions in a Multichannel Online Marketing Environment: An Empirical Model and a Field Experiment, Journal of Marketing Research, Vol. 51, No. 1, pp. 40 56, 2014. [5]: Xuhui Shao and Lexin Li. 2011. Data- driven multi- touch attribution models. In Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining (KDD '11). ACM, New York, NY, USA, 258-264. NEXT STEPS For more information, or to schedule a demo of SYNTASA s Advanced Attribution Analysis, please contact: Grant Wagner VP, Sales & Marketing grant.wagner@syntasa.com 703.596.0100