One of the fundamentals of running a successful business is understanding your customers’ purchase habits. How often do they buy, what do they buy, and will they buy again in the future are all important considerations. In fact, it could be justified that this knowledge is the most important aspect of continued profitability. With so many internal operations (and dollars) tied to customer behavior, any improvement in understanding them should lead to greater margins when that knowledge is applied appropriately.
Discovering insights from Internal Data
Working with Expected X, a mid-sized manufacturing company needed assistance planning their annual marketing activities. Operating within a niche market, they determined that all potential customers were either already working with them or a competitor -- new, unserved customers were few and far between. Therefore, rather than focus on acquiring new customers (or attempt to switch competitors’ customers), they decided to allocate their marketing budget towards upselling and better serving their current customer base (i.e. market penetration).
After this, they were at a loss. Where to begin? How do we determine our best sales opportunities among our customers? Their initial plan was to immediately engage a market research firm to help them gather additional customer information to use as a guide for preparations. While their logic was sound in choosing to investigate the situation rather than take a shoot-from-the-hip/gut instinct approach, they failed to first understand what their own data could tell them -- a common mistake companies make prior to implementing a market research project. Thankfully, Expected X was there to intervene by recommending a “SecondaryFirst” approach: analyzing internal data for insights before approaching an external market research firm.
Connecting Research Objectives to Applications
CRM data is often a treasure trove of knowledge regarding clients and their interactions within an organization. Unfortunately, many companies do not have the resources to effectively mine this data. This is particularly the case with small and mid-sized businesses which don’t have dedicated marketing, research, and/or data analysis departments. Expected X fills this void.
Working together, Expected X and the client identified two main business questions:
- Which customers are most likely to purchase a product within the next year?
- What product is most likely to be purchased next?
By answering these questions and applying that knowledge toward formulating the annual marketing plan, the client could approach high-potential clients with a specific product. Not only would this save them marketing spend (a fairly small budget for most companies this size), but it would also save them prep time for manufacturing the product by knowing beforehand what it is likely to be (the company’s products were custom setups not mass produced).
Modeling the CRM Data
Data can be dirty as is often the case when it is not maintained for the purpose of analysis. This instance was no exception. However, a wide spectrum of analytical techniques exist -- some more robust than others. Expected X took a two-model approach:
- Modeling Customer Lifetime Value to predict purchase probability within the next year
- Modeling purchase patterns using a simple Markov model to estimate the product with the highest probability of being purchased next
By interlacing both model outputs, the manufacturer could set thresholds for targeting their most valuable customers (e.g. customers with an 85% probability of purchasing this year who will most likely buy Product XYZ). When the model was tested on newer purchase data not used to build it, it proved to be ~81% accurate in predicting purchasers vs. non-purchasers. Additionally, the Markov model predicted the next product purchase correctly ~68% of the time.
Finding Measurable Value in Research Outcomes
So how does this translate into value for the client? Before working with Expected X, the client had invested nearly its entire research budget into surveying its customers annually and used satisfaction scores as a proxy for future purchase intent. Needless to say, this did not prove fruitful -- it proved costly.
The solution provided by Expected X was able to reduce marketing spend ~11% by reducing/eliminating marketing activities to low-potential customers and reallocating “on-the-fence” customer spend to high-potential customers -- all at a fraction of the cost of doing business with a major research firm.
What can you learn from your data before investing in costly research?