Think about the last time you heard an entrepreneur speaking regretfully about a business failure. Many people who go through such circumstances admit they never saw it coming and wish they’d done things differently to prevent what happened.
You can’t gaze into a crystal ball to see what the future holds, but data analysis may give even better insights. Here are four methods that could predict business failures before they occur.
1. Cognitive predictive analysis
Cognitive analysis combines artificial intelligence (AI)with high-powered data analysis. The “predictive” part of the name means this method assesses the likelihood of something happening.
Coverage from McKinsey & Co. highlighted the potential of this kind of predictive analysis. It gave the example of an oil producer that repeatedly faced problems related to compressors on off-shore platforms. A breakdown of a single platform could cost the company millions of dollars per day, and engineers tried for years to pinpoint the cause of the failure, but could not reach a definitive conclusion.
Eventually, the company came up with a solution that analyzed sensor data with information on 1,000 parameters. A resulting algorithm could predict equipment failure several weeks in advance. The company then adapted by preparing repair specialists and other personnel to start fixing the broken equipment as soon as it stopped working. This approach shortened the overall downtime by more than half.
Another example of this method in action relates to using data to predict the failure of car parts. Dr. Stephen Norris, technical lead at a company called We Predict that utilizes this approach, explains how it works. “The predictions can be at the car-line level, brand level, subsystem level or the component level. We can forecast for every single combination or permutation in the data the behavior of a vehicle or component when it reaches the end of its warranty.”
Having this kind of information can help automakers avoid preventable expenses. For example, if the data shows that the rate of warranty claims went up substantially after switching to a new supplier for a particular component, the brand may want to rethink that partnership before associated costs balloon out of control.
2. Anomaly prediction
Anomaly predictions—also called outlier predictions—allow you to pinpoint occasions where the data output strays from the norm. People often engage in anomaly predictions by using time-series analysis. It means the information shows details associated with particular intervals in time, such as every day or every hour.
Anomaly predictions look at history to predict future values. An anomaly on its own does not necessarily signify a good or bad event. It just indicates that something is different from the usual.
Pinterest uses a real-time anomaly detection system based on time-series analysis. It deals with hundreds of millions of time-series events every second. Engineers then get real-time data about circumstances that could cause failures. They use that newly acquired knowledge to take decisive action and keep the social site functioning as it should for as many users as possible.
Anomaly predictions are especially valuable for e-commerce brands. They can minimize site downtime, plus show issues that require prompt action, such as small but continual declines in revenue. Today’s consumers have less patience for dealing with website outages or other blunders that disrupt their experiences. Depending on anomaly predictions could help you notice characteristics about your store that could lead to future failures, giving you time to fix them.
3. Weibull analysis
Weibull analysis, also called life data analysis, determines how your product will perform throughout its lifespan by analyzing a sample set of failure data. Weibull can characterize a wide variety of trends that other statistical distributions may not show. Those include decreasing, constant, and increasing failure rates. This approach can show companies how to avoid problems later.
One transportation company began experiencing new parts failures with fuel subsystems on some vehicles within its fleet. Although the buses were not new, this particular problem was not historically an issue. Moreover, four failures required emergency repairs, adding to the urgency of diagnosis. A Weibull analysis ultimately revealed that a brass elbow was failing due to clogging.
Addressing the problem required changing the bus maintenance schedule. Before obtaining the data from the Weibull analysis, maintenance experts replaced the brass elbow at the same time they serviced a different component, and they made that choice due to convenience. However, when the team used the data to replace the brass elbow at a different time, it stopped the failure from happening.
A Weibull plot records the percentage of product failures over an arbitrary period. However, making a Weibull analysis work for you means determining a time-based measurement that makes sense for the respective product. For example, if the item is a tire, the number of road miles makes sense. If it is a washing machine, the number of wash cycles is likely a more appropriate gauge.
4. Real-time consumer trend analysis
History is full of products, marketing campaigns and other corporate efforts where the people involved felt sure they had their fingers on the pulse of the next big thing, but were wrong. Marketers, product designers and other people working on brand or item development must still rely on experience to avoid pitfalls. However, real-time consumer trend analysis can help fill knowledge gaps.
Maneesh Kaushik, the global insights director at PepsiCo, uses a platform that gives a real-time view of the food and beverage options that are currently generating buzz in the marketplace. It tracks more than 1,000 ingredients and dozens of benefits that matter to people. For example, the data might show that consumers are particularly interested in a drink that features an exotic fruit, plus promotes alertness.
Kaushik explained, “That means that we can go back and tell PepsiCo that, over the next few months, you will start seeing this benefit becoming more and more important to consumers, or we will start seeing these ingredients start gathering scale.” Such relevant data can prevent business failures by reducing the chances business professionals invest heavily in a product that consumers are not interested in trying.
Brands can also find success if they figure out how to personalize product recommendations throughout the shopping experience. Wish.com discovered that its shoppers did not always know the brand or the product they wanted to buy. Those consumers based their purchasing on browsing first.
Wish.com deployed a customer data platform that captured more than 17 billion events per day. The associated information included browser cookies, past order data and product SKUs. The tool gave customers accurate recommendations so they could shop without searching. If customers don’t find what they need at websites, they’ll leave, and profits may plunge. Wish.com doubled its year-over-year conversion rate using this method.
Help your business perform better
Even though you can’t know with certainty what’s ahead, these future analysis options give you trustworthy information that could make failures less likely to happen. Consider using one of them soon to help your business thrive.