William F. Nowacki | 11 January 2019
Successful businesses have long relied on predictive modeling and analytics to understand how to unlock opportunities and to invest for development as well as anticipate future areas of growth.
Those that take a holistic and expansive view of how data can inform every part of decision-making will give their organization the flexible platform it needs to adapt and succeed.
Traditionally, organizations would employ decision models that took into account just a small number of data points (or signals) to help predict outcomes. Examples include models that help a bank decide whether or not to give you a loan, or that help an online retailer recommend another item. The explosion of data, however, (90 percent of which has been created in the past 2 years alone1) offers a great opportunity both to expand the reach and increase the depth of predictive modeling. It also poses a serious challenge to all businesses. Namely, if you are not expanding the scope of your predictive analytics what opportunities and risks are you missing?
Over the past couple of years, we have been developing the KPMG Signals Repository in anticipation of the growing need for 'big data fueled' predictive analytics. Our two-part process takes into account thousands of relevant signals allowing our machine-learning analytics to execute with greater accuracy and predictive power. And the more data the repository receives, the smarter it becomes as more signals and causal correlations are identified.
In the relatively short time that we've been using KPMG Signals Repository it has already become clear that this super-charged predictive modeling and analytics solution can make businesses more efficient, and help specifically tailor products and services to customers and markets – generating greater revenue as a result. Here's how KPMG Signals Repository works and a couple of the interesting ways we've been using it to help organizations.
Start making sense of big data
The Signals Repository provides time-relevant, location-relevant context to both understanding what happened in the past and what's likely to happen in the future. It involves two key components. The first is the repository itself – a data platform that constantly receives and processes pieces of information ranging from weather to real estate to doctors to stop signs to capital markets to college football games. The second component is the scoring engine – essentially a machine learning analytics platform that enables data scientists to rapidly explore tens of thousands of signals while engineering and selecting features, building models, and ultimately solving business problems.
The KPMG Signals Repository has a number of internal functions that – when assisted by a person – enable rapid system learning and improvement. Consequently, models expressed within the scoring engine 'learn' and improve over time. Here's how it can be applied.
No more crying over spoiled milk
Imagine you're the head of the dairy department for a large grocery chain. You have responsibility for the profit and loss of the entire dairy department at each of your 3,500 stores across the US and Canada.
Having been in and around dairy for 20 years, you know several things to be true: milk, cheese and eggs are high volume items and it's hard to keep them in stock. As such, you've worked with your forecasting team to refine your automated inventory replenishment system to ensure that as quickly as stuff is flying off the shelves, the delivery team is restocking. With trucks going in and out of stores every other day, you've learned to shove as much onto each shelf as it can hold.
In your 20 years, you've also learned another thing: spoiled milk and eggs stink. Not to mention dairy is a low margin category and even a modicum of spoilage can put the category in the red.
Then one morning, your phone begins to ring. Random stores are reporting that sales of milk, eggs and cheese are off. Way off!
What you didn't know – what you couldn't see is that elementary and secondary schools around the country are closing down for spring break. Some closed this week. Others will close next week. And others the week after. All information available publicly.
But what if your inventory management system had been connected to the Signals Repository? And what if KPMG had helped your forecasting team incorporate features – predictors – that were attuned to local events schedules?
You'd no doubt have been prepared for the sudden, dramatic downturn in demand for dairy in each of your stores and would have avoided the loss.
The goal of KPMG Signal Repository is to understand true cause and effect – given situational context – thus allow people like the category manager to make better decisions.
The discovery of a relationship between school schedules and localized demand didn't happen by accident. Rather, it was identified through exhaustive machine learning experimentation. Once identified, it was codified and deposited in the Signals Repository Archive where it can be used by others trying to solve similar demand planning problems.
Build a better shopping mall
KPMG Signals Repository is just as useful when it comes to major planning and investment decisions. A local KPMG member firm recently advised a real estate investment trust (REIT) on new shopping center investments. The REIT had been around for 40 years and so already had a refined predictive model that drew on 39 variables. Using the repository we were able to apply 400 new signals to better inform their planning, including ease of access to the car park and the ideal square footage for secondary units in the shopping mall (80 feet if you're interested). However the signal that proved particularly useful had nothing to do with the shopping malls themselves. We found that the best opportunity for a return on their investment would occur when the mall's trading area had the fewest four bedroom homes that had seen price decreases over the past 24 months.
If that sounds really obscure, well, it is. But that's the power of big data predictive analytics. It listens to all internal and external inputs then combines them (with the help of human sense-checking) to provide a decision-making road map unlike any organizations have seen before.
Is your business ready to receive the signals?
It's becoming clear to all of us who work in the field that advanced data analytics has the potential to transform many different parts of an organization. Certainly it can improve logistics and sales but it can also be employed to bring greater accountability to sustainability, inform investment decisions, mitigate insurance risk, improve employee productivity and even anticipate competitive threats. Yet the best deep data insights will only be useful if the organization has a consistent strategy on to how it uses data. Those that continue to silo or `protect' their data because they only see it being relevant for specific parts of the business will fall behind. Those that take a holistic and expansive view of how data can inform every part of decision-making will give their organization the flexible platform it needs to adapt and succeed. If you haven't considered yet how prepared your business is to be data-led, it's time to start.