Technology has upended industries, capabilities, and functions, and has the ability to automate a great number of tasks. The evolution of technology has meant the automation of more sophisticated tasks, including things like resume screening for HR, legal reviews of documents, and automated marketing communications.
Internal IT organizations are tasked with optimizing internal and external processes, vendors, and software to optimize the return on technology investment. This works well in operational and staff capabilities where processes are streamlined, and the inputs and outputs are well understood.
Marketing organizations, however, are a different breed. Marketers’ needs change; their ability to adapt to competitors, build loyalty, and test new campaigns and offers, across all digital and non-digital media. This causes their technology needs to be more complex. Methodologies, understanding of fickle customer interests and behaviors, and changing channels (Facebook rules, SEO changes) mean that the routine process of installing sophisticated software and tools and then moving on to the next initiative doesn’t apply.
Analytics are key to the success of any contemporary marketing organization, and data sciences and predictive modeling are more important than ever. However, buyer beware inside an IT organization! There are many IT data science providers who sell themselves as adept at changing customer behaviors as they may be at, say, fraud detection or actuarial sciences. They look at data sciences (Hadoop, machine learning, statistics, etc.) as a solution that by itself changes everything. CIOs who are challenged with building a customer-driven infrastructure (data warehouses, data lakes, marts, platforms, stacks) must understand the limitations of the tools and the needs of marketers if they are to provide technology, resources, and leadership in this area.
Here are seven things that CIO’s need to know about marketing analytics when evaluating internal and external partners that support marketing:
- Marketing data is messy! Integrating social, email, transactional, CRM, and web data is difficult, as each channel has its own tactical limitations on collecting and sharing personally identifiable information. For example, your web log files have great data, but unless you know who is using the IP address you don’t know who the customer may be. Add in situations like hand-entered data by customer service reps and the challenge is even larger. Anyone helping here must know fuzzy matching, reverse appending, data quality, and external toolsets that will fill in missing data. When done correctly, an email address can become a name and address that will match an order form.
- External data about customers is critical. You may know that a customer spent $150 with you, but, could they spend a lot more. Are they a millennial? Do they have kids? Are they interested in your product category? There is so much external data about customers available today that any provider should be knowledgeable about this. Not only that, a competent data scientist will know how to optimize the use of this data in algorithms (whether machine learning or statistics processes). They should know if different models will perform better, what to do with missing values, and how to maximize the inferences that can be made. For example, sometimes a missing value means nothing (like, the record didn’t match), but other times, it could mean something (“not a homeowner”). These are in-the-weeds issues that impact the quality of predictive analytics solutions and they require marketing expertise.
- Customer Journeys and context matter a lot. When a data scientist is building a solution or prediction, he/she must understand the universe they are talking to. For instance, you never combine a “prospect” universe with a “customer” universe. Prospects have very little data, no purchase data, and few interactions with their brand. Any predictions on this group are broad guesses, meant to get the odds in your favor. Customers, however, know your brand, know your products, and want to hear from you. Optimizing what you say, and thus the predictive analytics process, is critical. Also, you have so much more data to use in the prediction modeling.
- Methodology is not strategy. The latest tools and advancements in data management mean that different data is available for use in analytics. However, a data scientist must start with the modeling strategy. Who am I modeling? Lapsed customers? High value? What am I trying to do? Get one more visit from a lapsed customer? Increase average order size? Where can I use promotion history to see if this customer responds to offers in the past? If a data scientist does not have experience in understanding the customer lifecycle, they will fail. Great new data sources and processes cannot, and will not, compensate for lack of modeling strategy and experience.
- The data is more important than the technique. Data Scientists must be skilled in understanding the data, what it is telling them, and how it should be modeled for optimal prediction and performance. Various techniques that are standard in different contexts (fraud detection, etc.) may be helpful in marketing analytics, but it does depend on the competence of the data scientist doing the work. Is Machine Learning (ML) more accurate than statistical procedures? Maybe, maybe not. We use both strategies based on the data and the application. In some cases, ML can produce superior results. In other cases, regression modeling out-performs. In either case, neither works well if there is no thought behind the input data before it goes into a procedure.
- Text analytics and sentiment analysis only apply to a very small portion of a customer base. It can be in a call center, comments on a website, social feeds, or guest satisfaction feedback. The data in these areas is invaluable from a strategic standpoint, and can be mined and analyzed for specific learnings, insights, and sentiment. However, for a great majority of businesses, this is more of a research exercise to report on strategy than it is a personal customer insight that can drive communication. Why is that? Typically, less than 3% of a customer base participates in any feedback that has text associated with it. This means that the time and effort applied can only be used for engaging 3% of your customers, not the other 97%. Again, this can be useful for research and trending, but not generally for predictive analytics.
- Applications drive predictive modeling strategy. If the application is hot leads for a sales team, that requires specific journey modeling, appropriate use of external data, and, turning anonymous consumers into known consumers by using data integration technology. If the objective is to personalize and optimize email communications to customers, then predicting who is likely to buy, adding segmentation, and version testing will be required. There is no one-size-fits all modeling strategy. Experience and track-record are important in evaluating potential suppliers.