Beware of False Profits, Which May Come to You in AI clothing
With all of the hype around AI, don’t overlook the importance of Human Intelligence to ensure your analytics efforts are addressing the right problems
With apologies to Matthew 7:15 for the tacky paraphrase, companies today must remember to look beyond the science of AI and machine learning alone to identify areas where analytics will help drive revenue growth.
Recent McKinsey research identifies “Analytics Translators” as a key role to help companies derive value from their increasing investments in data science, AI and deep learning by translating insights into action. They also talk about the “perfect union” between creativity and analytics that is cross-pollinated in market leaders today. At MarketBridge, we call this the “Human Intelligence” side of the equation.
In the rush to build data science skills and capabilities, companies must not lose sight of the need for creativity and a deep understanding of the business and its customers to both identify the best focus areas for AI, as well as to effectively “translate” analytic insights into business outcomes in new and creative ways.
Case in point
We recently worked with a large distribution company which had observed a slow decrease in customer spend, for a specific product, across many of their accounts. Over time, these accounts were exhibiting small decreases in both order frequency and volumes in the category, but in many instances these small category decreases were masked by offsetting increases in other categories within the account.
By the time this SKU-specific decline was brought to the account manager’s attention, it was often too late – the customer was sourcing from other competitors, and remote sites were not conforming with centralized purchase agreements.
The category owners wanted to develop models that would help predict which accounts were likely to decline by a certain percentage in that category in the next quarter. Armed with that data, they could then point their account reps to those accounts sooner to try to identify and mitigate the root cause of any possible decline wherever possible.
We built a machine learning model, first flagging those accounts who had exhibited that type of slow decline, and then using an algorithm to pick out the features that predicted the decline.
Once that model was developed and validated, the entire population was scored monthly on their probability to exhibit this slow decline, and those accounts that had a higher risk profile was flagged for treatment by both sales and marketing (i.e. category promotions, rebates, account reviews, etc.) to try and retain that category business.
One of the key indicators the models identified was that non-seasonal reductions in order frequency and average order size for these items were predictive of longer-term category decline. While not a surprise, the process of systematically evaluating the portfolio and identifying and ranking those accounts at greater risk put a much greater focus on the issue. Armed with the outputs from that model, the client was able to implement marketing and sales programs that helped reduce decline in those higher-risk accounts by identifying “early warning signals” much sooner and flagging those accounts for engagement.
But this “decliners-only” view of the issue did not allow the client to easily see another important dynamic in the marketplace. When we looked at overall customer activity within the category, we uncovered another important reality.
There were a number of customers who had significant increases in their order frequency, volumes and average order value within this category. As a category owner, one might be tempted to think that the firm was doing very well with those customers since their order volumes, frequency and order values were increasing. They were using more of our product, they are engaging with us more frequently, and our category revenues are going up. What could be wrong?
In fact, what the analysis showed was that many clients who exhibited these apparently positive behaviors ultimately contracted to zero. This binary behavior—going from a great account to nothing—was much more serious than small declines and shifts, and could have been totally missed, if the right question hadn’t been asked. When we looked deeper, it became apparent that the increase in overhead and effort associated with managing this category prompted buyers to give up trying to manage it, and instead hand it off to a value-added third-party managed services contract.
Even though the client provided similar solutions, their decliners- and category-only view of the problem blinded them to the opportunities within their own portfolio to migrate the customer to a different solution—until it was too late.
So, what are some of the best practices you can put into place to avoid falling into a similar trap? Here are some suggestions based on work we have done across numerous clients and best practices we see in the marketplace:
1. Start each analytics initiative with a “Delphi Method” brainstorming session
We like to start each analytics initiative with a brainstorming session, where we bring together a number of cross-functional stakeholders to review the problem statement and hypothesize and prioritize scenarios that should be investigated in the analytic process.
In the example above that team included category managers and marketers, account reps and category product specialists, market intelligence personnel and customer service and support. Using a structured process such as Delphi allows you to identify, capture and prioritize hypotheses across customer and category “experts” which can and should be tested in the analytic process.
2. Identify external data sources—including talking to customers—that provide additional insight into customer behavior
In the case above, we incorporated a third-party data source that tracked search data for those clients with increasing spend. We were able to identify select clients that were actively searching for managed solutions as their category usage increased, and we were able to incorporate that in the next set of models we developed which were focused on target accounts for the managed solutions category. This helped the client to identify those accounts who were actively searching for managed service solutions.
An even more obvious way to gather external signal is to actually call customers. This can be formal “qualitative research”, or it can be just picking up the phone, or talking to the sales reps who talk to customers every day.
3. Expand beyond line-of-business organizational boundaries
In many instances, the focus of an analytics initiative is limited by “who pays for it”. In this instance one category is looking to maximize their LOB revenue vs. looking at the portfolio of business with the customer and understanding how to maximize the overall market basket with that customer. Looking at relationships between LOB’s and product usage and purchase patterns is key to identifying relationships that may not be readily apparent when viewed through the single-LOB lens.
Integrating data science expertise with business domain knowledge and creativity is key to driving value from your analytics investments, and ensuring that you are focused on solving the right problems. Combining AI and HI, data and content, creativity and analytic rigor—while often challenging—will yield much greater returns.