Retail in the Age of Disruption (Part Two)

This is the second entry in the CityStream Solutions blog series discussing Four ways that Big Data, Data Science, and Predictive Analytics must evolve and innovate. Navigate all Age of Disruption series entries: (1) | (2) | (3) | (4) | (5)

From the “future state” network model to the “pendulum” network model

When developing a strategic network of warehouses, distribution centers, and fulfillment centers, the old way was to compare possible future states. We would look at 1, 2, or 5 distribution nodes and compare the truck trips (by length and frequency) that would be necessary to supply the stores. Then we could look at the cost of diesel fuel at $1.50, $2.50, $3.50, $4.50, $5.50 per gallon. When fuel costs more than $3.50 per gallon, it becomes cheaper to operate 3 distribution centers, because the costs of the extra warehouse (more real estate, more employees, another branch manager, etc.) are outweighed by the costs of the extra truck trips (they are longer and cost more as fuel prices rise). Sophisticated, data-driven models could produce a bunch of future states based on a bunch of variables and help a company select the most cost-efficient model.

But we have learned from recent history that fuel prices don’t slowly ratchet upward. It sure looked like prices were only going one direction, passing $1.50 in 2002 to hit $3.25 in 2009. But then the Great Recession hit, and prices dropped to $2.25 in 2010. As recovery took hold, it skyrocketed back up to $3.65 by 2013. But now, the Saudi-Russia feud combined with stay-at-home lockdowns all over the globe have combined to push prices back below $2.50. With this price volatility, it’s impossible to use future-state models. A plan for opening a new distribution center when prices hit $3.00 would have turned out to be disastrous, not that we could have seen it coming from 2007.

Big data analysts and data scientists must keep advancing their models. We must move from “future state” to “pendulum” to help companies understand how their real estate footprint can adapt to a volatile pricing market. We must add the idea of unreliable resources: electrical power, highways and rails, ability to import from abroad. There is no single future state to optimize for in a world that isn’t just changing but cycling, swinging like a pendulum between high and low prices, open and closed public spaces, available and unavailable resources. Predictive analytics must be capable of finding “average” efficiency among several futures instead of peak efficiency for one future: multi-objective optimization will be an important tool in the data analysis kit.

Instead of a single challenge, there are now three problems for analysts to solve. First, modeling for the high extreme (diesel at $5 per gallon). Second, modeling for the low extreme (diesel at $1 per gallon). Third, predicting how much time we think we will spend at one pole, at the other pole, and in between. The third one is the trickiest, as it involves advising clients on making long-term decisions that are hard to quickly re-calibrate or overhaul. Analysts will have to step up their game to continue to provide value to their clients.

Continue on to part three here.

Questions? Comments? Contact us.