At Schneider National, a US$3.5 billion logistics and transportation services provider, Travis Torrence has seen his job as a dispatch analyst change since he joined about two years ago, writes Thomas H. Davenport in a Wall Street Journal article. The company launched a new dispatch system that optimizes matches between arriving shipping containers and available truck drivers, which means Torrence spends less time on any one particular container—figuring out who can take it and how quickly—because the system does that for him.
Torrence and the dispatching algorithm are like colleagues, notes Davenport, a Babson College professor who has been writing about Schneider’s investments in analytics for some years. The system identifies the match and then Torrence checks the data and augments it. He can check to see if a driver can pick up a container earlier than the system suggests. He can ask if a customer due to receive a container can be flexible on delivery times. And he can consult weather reports to confirm whether the system’s estimated delivery times make sense.
As a result, Torrence can complete about twice as many dispatch jobs in one day with the new system than he could with the system that was in place when he first joined Schneider. The machine does not obviate the need for staff, but it is possible that a downturn in demand could lead to staff reductions, Davenport says.
Like a growing number of systems embedded in a business process, Schneider’s logistics optimization process represents an evolution in the way enterprises are incorporating algorithms into their operations (see “The New Tool for Process Change”).
Algorithms as Competitive Assets
Companies used to become leaders in their industries by establishing an unbeatable brand or by having a supply chain that was more efficient than anyone else’s. Now leading companies are investing in algorithms to improve their position.
While executive committees still build market value and gain on their rivals through traditional means, algorithms, powered by data, have emerged as a better way to change the business and gain a competitive edge (see “The Drivers Behind Algorithms’ Rise to Stardom”).
Algorithms should be part of corporate strategy. These critical assets give meaning to the growing mounds of data that enterprises have been amassing. Used well, they provide insights that can make a business process more profitable or competitive and spotlight new ways of doing business and new opportunities for growth.
5 Traits That Lead to Success
Algorithms are pieces of code that use data to influence a business process, and analytics is synonymous with a business algorithm that provides guidance or insight derived from data. For example, a retailer analyzes customer behavior to determine the best way to use pricing and promotions to increase sales. A mall owner studies pedestrian traffic patterns to offer lease prices based on proximity to hot spots. An oil rig operator analyzes data from the sensors on a drilling rig to determine when it’s time to replace equipment before it breaks down. A machine manufacturer correlates warranty claims to issues that can lead to future product improvements. And a port uses GPS and other data to better manage ship and container truck traffic for loading, offloading, and transport.
Organizations employing algorithms that deliver results have several traits in common:
- Strong data collection and management. The types of data that businesses can collect has grown dramatically. Examples include data from sensors in machines on the factory floor, on the road, underground, in the air, and at sea; from a company’s online customer interactions; from third-party data sources; and from social media streams. Algorithms make great use of this data, but companies must prepare and organize the data for it to be valuable. Dirty data doesn’t work.
- The right analytical models. To be effective, an algorithm has to include the right mix of mathematics and assumptions about the world. If an algorithm includes an idea that is wrong, such as that housing market prices will always rise and never fall, it can lead to faulty results. Such was the case among some Wall Street firms that relied on computer models during the run-up to the Great Recession, Ed Sperling reports in his Forbes magazine article “IT’s Role in the Recession.” The thinking behind the algorithms has to account for realworld situations and changing conditions. That’s why it’s important to revisit them and adjust.
- Investments in training. The people who use the results of an analytical process need to know what to do with them. That’s as true for a veteran company executive as it is for a frontline worker or a person with a PhD in mathematics. The algorithm to optimize shipping container management at Schneider National is effective because of the training the company provides, Davenport notes. That benefits both the employee’s career and the company’s bottom line.
- A culture that prepares and adapts to changes that algorithms bring. Organizations that report the best results have put analytics at the center of their business strategy and have appointed an executive to lead the effort, says Christopher Mazzei, global chief analytics officer at EY.
- Attention to risk management. The pressure to improve results never wanes, and with companies holding so much data, executives often think that there must be value there, says John Lucker, Deloitte’s global advanced analytics and modeling market leader. That could well be true, but smart firms also consider two risks when they pursue a strategy for analyzing all that data.The first risk is a matter of rights: ensuring that they comply with privacy rules and that the terms under which they use data for analysis are allowed according to agreements made when they collected the data. For example, if a customer shared data to get a coupon offer, the company must honor its terms for collecting that data and must not use the data for purposes outside the scope of that agreement.
The second risk involves competitive advantage. When companies decide to monetize their data by creating a new product or service, they must ensure that they get more than they give away. Not all companies take the time to consider this risk after spending so much effort to create what they consider a valuable algorithm. They are eager to realize that value by selling the insights to others. “You really need to think about whether you should do this and what rights you allow somebody to have to use the information,” Lucker says.
Steps Forward on a Continuum
Algorithms can increase customers’ willingness to pay.
Using shopper’s opt-in data, a mall in Southern California analyzes pedestrian traffic to discern movement patterns, answering questions like: Where do people go after they get something to eat? After they visit the electronics store? Which customers visit anchor tenants, and what other shops attract their visits? The answers enable the mall to determine leasing rates based on the relative hot spots revealed in the data. Spaces that are likely to attract more customers get higher rents.
It’s a well-trodden path for technology firms to acquire startups for the innovations they have developed and the talent they employ; expect the same kind of behavior for takeover targets when it comes to algorithms. New algorithm-driven business units, such as General Electric’s billion-dollar investment in a division to develop predictive maintenance and other analytics offerings for industrial customers using data signals from machine sensors, are another path for asset investment.
Algorithms have a lifecycle.
An algorithm starts to go stale as soon as it’s deployed. It is important to revisit the model and the assumptions upon which it is based to refresh or add data sources. New circumstances that deepen understanding about market conditions, the economy, an industry trend, consumer tastes, company goals, and other factors can color the variables that go into an algorithm and the questions it is designed to answer. Business processes related to an algorithm can change, too, creating another reason to revisit it and adjust if needed.
The great thing about algorithms, of course, is that they enable the user to measure their success. Did the decision to invest more marketing dollars in the western region succeed? Look at the data. Compare projected versus actual results. Pour the findings back into the model to improve the algorithm.
As a field, algorithm development is advancing. Startup BeyondCore has developed an analytics application that automates the discovery of important correlations in financial and operational data (to make data insights available to those “beyond the core” group of data scientists). Its system, which uses machine learning as part of its algorithm, is not bound by a preset limit of variables. In other words, its creators designed it to be open to new variables and more data.
Machine learning is an advancing field. These are the algorithmic models that guide driverless cars, perform automatic speech recognition, and execute other advanced computing tasks that approach human capabilities. Scientists constantly revisit the algorithms to improve their accuracy and to add or improve functionality
The effectiveness of a predictive analytics algorithm is only as good as the goals you define.
Use key performance indicators (KPIs) to judge the success of your algorithm strategy just as you would with any other initiative. Grow revenue. Increase yield on agricultural fields. Elevate market capitalization. That KPI data and historical results will tell you what’s going on.
Planning for such evaluations must begin well before that day, however, says Eric Siegel, author of Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die.
“The project must be designed backwards by starting first with the end goal,” Siegel says. “What actionable purpose will the analytics serve? For example, a predictive model could be trained to target marketing based on predicted purchases or cancellations or to target fraud auditing. Ideally, some specific, existing, large-scale operation must first be agreed upon for optimization in this way, and there must be a critical mass of buy-in to alter the existing processes by acting on predictions.”
Then comes testing. To conduct a proof-of-concept project, it’s critical to evaluate it over a set of data that would be like that used for a full production deployment in order to determine the model’s validity, Siegel says.
And then we’re back to the beginning: making sure there’s leadership support and the ability to integrate the results of analytics into existing business processes—or launch new processes to take advantage of the analytics results.
The Work Never Ends
Algorithms are not pieces of projects, steps to take along the way to make your enterprise more efficient. They are core assets that make your enterprise work. When algorithms are well managed, they point out strengths and opportunities, weaknesses and failures. They show what is working and when it is time to refine the effort and try again. They are not just an IT responsibility; they require C-suite commitment. They call for careful tending and management.
In fact, Davenport says, the work on algorithms never ends. At Schneider National, the work has gone on for years. “At Schneider, they’ve been using analytics to optimize their dispatching for a long time,” he says. “They’ve been putting sensors and GPS devices in their trucks. They now have indicators that a driver might be more likely to have an accident based on past driving behaviors. There are all sorts of things that they have done. If you want to be a successful analytical competitor, you just can’t stop. You can’t do a few things and say, ‘OK, well, you know, we’re pretty much done with that.’”