A little over three years ago, I wrote a series of posts about the pressures online retail was placing on traditional brick-and-mortar stores. From behemoths of the past, like Borders, to those at risk today such as Sears, companies that failed to react to broad changes in consumer expectations have closed their doors or are struggling to keep them open.
Let’s be fair, though. Physical retail isn’t doomed. It’s a piece of the puzzle in the omnichannel world. It’s just not the centerpiece anymore. Companies that are succeeding are paying close attention to customer data, learning or predicting what their customer’s actions will be and driving customer experience to an entirely new level. A level that is the new baseline for how to compete and succeed in retail.
And if you’re in the B2B space, I hope you’re paying attention to the growing role of machine learning and predictive analytics.
Over the past 20 years, I’ve helped many companies implement both B2C and B2B sites. The challenges are different in some ways, but the expectations of those interacting with a site have been converging for years. And as companies like Amazon continue to expand in the B2B world, you can be sure that the demand for a “smarter” experience will grow.
Most companies in the B2B space are using web analytics to track the basics; page views, visitors, bounce rate, and similar metrics. Today, that is the low bar.
The true value that can be extracted from the way in which customers interact with a site isn’t lurking in an obscure Google Analytics report. It’s buried in the wealth of data that can be collected from customer interactions. For companies both small and large, the world of data science can be intimidating and knowing where to start can leave one’s head spinning.
But ignore this at your own peril.
B2B market: 3 things to know
OK, so a full discussion of data science is beyond what’s possible in a blog, but here are some key points that all in the B2B space should know:
- Being able to respond quickly and accurately to changes in customer behavior is critical. The days of taking a week or longer to peruse reports and figure out what your customers want are long gone.
- Predictive analytics and machine learning are here to stay and companies that employ these techniques will outmaneuver those who don’t.
- Your commerce platform may or may not be able to gather all the data you need. It’s a piece of the puzzle and there are other sources to consider.
So how can B2B companies use predictive analytics and machine learning? Here are a few use cases that I’ve seen.
Machine learning and predictive analytics: Customer classification
Understanding behavioral tendencies of customers and grouping them with similar customers can be an effective way to focus marketing and merchandising efforts. One classification that most have likely seen is the “VIP Customer” but too often, B2B companies simply decide, using a single metric, what constitutes a VIP.
One implementation I’ve seen simply classifies any customer that purchased over $100 as a VIP and placed a little badge on the customer’s online profile. That’s not really classification; it’s closer to gamification which gives the customer some sort of sense of achievement and hopefully boosts their loyalty.
Real classification is based on a set of data features measured across all recent customer activity. In other words, whether someone is a real VIP customer or not, is going to change when compared to the behavior of the entire customer base.
Knowing customer classification and being able to use a classification model to predict a group to which a new customer is likely to belong, can help greatly in determining which promotions to show to a customer while they’re on the site or via email. In a recent machine learning implementation we did for a B2B firm, we grouped customers into several segments by using data from a recent time period:
- VIP Shopper: The highest tier of customers based on value of conversions and number of conversions.
- Engaged Shopper: A large number of visits, an active cart, numerous product views, and at least one conversion.
- Window Shopper: Several visits, numerous product views
This information was not shared with the customer – it’s not a public distinction, rather it’s an internal indication of how to interact with a customer. Having this information allows marketers to display targeted promotions on site during a visit. For example, if a customer falls into the “Engaged Shopper” classification and has an active cart that is above the AOV, a coupon for free shipping or a discount could be displayed to move this shopper toward conversion.
It’s also important to note that models aren’t static. They must be recalculated frequently. In the recent implementation, the models would recalculate every few hours to ensure that they were as current and as accurate as possible.
Prescriptive product and content placement
There are many reasons buying behavior may change. Seasonality is one such reason. Other reasons that may drive short term behavioral changes are weather, news, shortages, and manufacturer promotions. Using models to detect conditions that are “outside of expectations” and adjust site merchandising to respond to time sensitive anomalies are becoming more commonplace.
This technique can be used to alter search results, home page item placement, category item placement, and guide users to product information when a search for a competitive product is conducted.
Does a given product perform better when visible at the top of the home page? Some do, some don’t. Being able to detect optimal placement for products given a recent history of activity such as conversions and cart additions can be accomplished through machine learning. When introducing a new product, similar models can be used to predict a best-placement for optimal performance based on product attributes.
Improving personalization with machine learning
Personalization on most B2C sites has a long way to go. And on B2B sites, it’s still in its infancy. While it’s possible to group shoppers into a cohort and show them similar items, machine learning makes it’s possible to make these cohorts smaller, approaching a more unique experience for your B2B customer. This can be especially impactful when there are multiple B2B buyers from a single customer.
To begin moving toward a more unique experience, machine learning models can be developed starting at a high-level, then progressing to more granular levels that deliver unique insights into a customer.
For example, beginning with geolocation as a factor, a B2B seller of industrial HVAC equipment should feature different products for customers in Minnesota vs. those in Florida. The buying seasons are not only different, but events in Florida such as an impending hurricane may influence buying behavior in the short term. A properly designed model can help spot these changes and alert the B2B marketer to changes that may require a change in site merchandising for a geographical region.
To add on, a company could develop models that can predict the optimal sort order for search results for a customer, the ideal products and categories to feature for them, along with suggested promotions based on their recent behavior.
Looking forward
It’s a fascinating time to be working in commerce and with data in particular. The convergence of low-cost cloud-based computing and the abundance of data available from a wealth of sources (not the least of which is a company’s B2B site) and provide a lot of actionable intelligence with an investment orders of magnitude lower than a decade ago.
That not only puts machine learning and predictive analytics within reach, it positions the technologies to become a core part of how you relate to your customers and how you operate your business.
Those that ignore the call today, may likely be tomorrow’s Borders.
