Knowing before customers turn elsewhere, machines go down, employees quit: The ability to anticipate and drive better business outcomes is becoming a decisive competitive factor. That’s the benefit of predictive analytics in a nutshell.
Suppose a company wants to find out why it is losing customers, what customers it can propose additional products to, and which channels promise the highest sales rates. What does it do? It looks at performance in its various regions of operation, divides the customers into segments by product, and addresses the most critical issues at stake.
A telecommunications company with 20 products, for example, seeking to analyze 20 regions and 10 customer segments using 10 score types such as churn management, channel affinities, and upselling, would need to develop 40,000 predictive models to be able to “address the different constellations at a fine-grain entity level,” explains Sven Bauszus, global vice president and general manager of Predictive Analytics at SAP.
Quite often, though, companies don’t know what to do with the detailed information the data patterns ultimately provide them. “Customers have the sophisticated processes and modern technologies but then often lack the resources to verify the results effectively and incorporate these new findings into their operations,” explains Bauszus, pointing out that too many customers still lack an “end-to-end approach.”
BARC: Predictive Analytics is a Future Must for 94% of All Companies
Without a doubt, predictive analytics is major concern going forward. According to the German market research group BARC (Business Application Research Center), 42% of the 210 experts recently surveyed in DACH region rated the topic as either important or very important. And nearly all (94%) of the enterprises questioned were convinced of its importance for future business success. Companies that had already deployed predictive analytics in relevant projects particularly praised the improved planning reliability (48%), better management of processes, the opportunity to develop new business models, and better decision-making support.
Seven Practical Use Cases
Predictive analytics can support day-to-day operations in a variety of ways and, as Bauszus emphasizes, is “not limited to the industrial sector.”
1. Fraud detection
From duplicate or incorrect invoices to manipulated balance sheets – rules and algorithms automatically detect irregularities, and halt suspicious transactions for manual verification.
2. Forecasting service dates (predictive maintenance)
The goal is to “replace corrective or reactive maintenance approaches with more preventive maintenance strategies,” explains Bauszus. Algorithms constantly analyze the machines’ behavior in conjunction with historical data to calculate the most suitable time for the next inspection and to ensure the right spare parts are in stock when needed.
3. Reducing waste (predictive quality)
Parameter settings defined for each stage of production enables companies to identify and remove faulty products from the production line early on, thereby reducing waste. This presumes, however, that the company knows the appropriate steps to take in response to the sensor data patterns indicating the defect. For more information, see .
4. Spotting unhappy customers (churn management)
Imagine a mobile service provider. After a year of no complaints, its customer service center is suddenly bombarded with calls mainly from young tablet and smartphone users. The trend pattern warns us that those customers are at risk of leaving. Based on this insight, the service provider can now preemptively retain their loyalty, or prevent churn, by proactively offering targeted discounts or new service packages.
5. Identifying upselling potential
A customer who has already taken out car insurance might also need liability insurance. But which customers are best reached by telephone? Which respond more often to e-mail? Who prefers regular letter mail? And where would “upselling” be pointless? Algorithms analyze customers’ past behavior to calculate the individual upselling potential and the most worthwhile means of approach in each individual case.
6. Improving payment moral
Outstanding receivables from suppliers and business partners are a dilemma for any company. Because when too many claims are outstanding, your liquidity starts to suffer. However, every company at one stage or another has become familiar with the conditions under which their customers reliably pay their debts on time. Discounts for settling accounts particularly quickly are one such example. Predictive analytics takes advantage of this insight to enable efficient cash forecasting.
7. Retaining staff
Already five years at the company, top qualifications, never switched departments: There is nothing to suggest that this employee might be thinking about looking for a different job. A closer look at his resume, however, reveals that he has never worked for more than four years at any one employer, and that he’s recently started taking further training courses. A forecast model can predict the likelihood of staff movements and gives the HR department enough time to react accordingly.
Weed Out Noisy Data
The challenge in the above examples and in other areas of application is always the same: “noise reduction.” Bauszus explains: There is a lot of ambient noise in your day-to-day business data. The trick is to extract those sounds, or patterns, that are relevant to the business issue at hand. For example, to determine the ideal time for maintenance of a machine, data scientists first use regression models to gather and correlate the readings of hundreds of smart sensors. The data from the top 10 sensors, or key influencers, is then trended and analyzed to predict when maintenance should be performed.
With SAP BusinessObjects Predictive Analytics, companies can choose whether they want to use the preconfigured automated forecast models or let their data scientists develop their own pattern recognition algorithms in expert mode. The standard solution serves as core starting point for predictive analytics. Fresh incoming data, for example in the case of certain target variables such as “fraud,” helps the system identify trends and patterns against past data, enabling it to continuously “learn” and leverage the insight gained. “As such, it [the system] provides forecast models representing the best compromise between training data sets and real data sets,” explains Bauszus. In expert mode, data scientists can set their own parameters, reuse existing SAP HANA libraries, and select algorithms.
Learn more from the following whitepaper:
Top image via Shutterstock