How companies look into the future

Predictive Analytics

How companies generate added value with Predictive Analytics


Targeted data evaluations enable forecasts of future business events. Predictive Analytics now helps people in many different industries to improve their planning of production conditions, cash flows or warehouse stocks, for example. Our specific examples show the potential hidden in the predictive modelling of data.


Text: Felix Raymann, Images: ©Alamy, Strandperle, Swiss National Bank, 24




Predicting future events and trends with the aid of historical data – that’s what Predictive Analytics (PA) is about. Amassed company data contains a lot of information which is very difficult to extract manually but can be quickly found and used by an automated analysis. “Many companies from practically all sectors can benefit from Predictive Analytics. Anywhere where repetitive processes take place, the collection of data over time produces a lot of usable information which can be used to generate forecasts”, says Martin Gutmann, Head of Analytics & Data Consulting at Swisscom. This can be the case almost anywhere – for example, in retail, in dealing with IoT or Machine Learning, in the calculation of cash flows, personnel planning and risk identification, or in industrial production and machine maintenance.

A logistics system yields lots of data which can be used for resource planning. Every day which a product spends in the warehouse costs money. Without optimisation, warehouses are usually too full, since companies do not want to risk delivery delays. “We are in contact with diverse SMEs which want to optimise their warehouse management using Predictive Analytics. System-supported forecasts enable warehouse stock to be planned far more accurately, which can result in tremendous savings”, emphasises Gutmann. Warehouse management is just one example of many. The two following examples show how Swisscom uses Predictive Analytics successfully in specific cases:


Example 1: Optimising the cash flow using forecasts

The Treasury department occupies the role of an in-house bank within Swisscom and is responsible, among other things, for the entire Group’s liquidity planning. The department approached Swisscom’s data analyst team with the desire to plan the cash flow better. Optimisation of the cash flow holds a lot of potential for saving. Every day, an average of 30 million francs enters the account as a result of invoices sent to private customers. But the responsible employees in the Treasury department do not know in advance on which day exactly how much money will come in. Knowledge of these incoming payments is important because the liquid assets should not be too high, but also not too low. The liquidity minimum is currently set at 50 million francs, which incur 1.6 per cent interest. The fact is that if the daily incoming payments can be more accurately predicted, the liquidity minimum, and therefore the costs, can be greatly reduced.





The Swisscom data analysts are therefore currently creating a forecast for each day as to how high the probable incoming payments will be. These results are obtained as follows: by analysing the databases from the previous months, the analysts are able to find connections and influencing factors between incoming payments and other factors. To do this, different data sources are used and variables are defined. For example, the analysis takes into account whether public holidays or holiday periods are included in the time period examined, whether a day under investigation is closer to the beginning or end of a month, which weekday it is and so forth. The more such factors are taken into account, the more accurately the payment flow for a particular day or for multiple-day cycles can be predicted.

Until recently, Swisscom Treasury was still creating manual forecasts in order to plan the cash flow in advance. Thanks to experience and 20 years of historical values, they previously achieved a deviation of plus/minus 5 million francs in relation to the average amount of 30 million francs paid in each day. These calculations were done on an ongoing basis using rudimentary Excel tables with only two variables. The new predictions using algorithms are not only faster but also more accurate: with the new method, the deviation can now be reduced to 3 million francs per day, which means an annual interest saving of around 30,000 francs. Although that is not an enormous sum, the example shows that Swisscom has been able to optimise planning with just a small project and at very little expense. The chief gain, however, is that the entire manual analysis workload is now rendered unnecessary.

The system continually learns over time: Swisscom Treasury provides the actual figures for incoming payments each month to the data analysts, who use the data to improve the analysis system. Further potential is available, and further optimisations would be conceivable in future for Swisscom Treasury – for example, analyses not only for the overall number of invoiced bills but for each individual customer. However, this would also involve much more extensive data processing.


Example 2: Holistic customer reporting

Customers can sometimes be a “black box” for companies: an unknown individual whose characteristics and needs are largely unknown. Many companies therefore have various departments diligently gathering data and manually creating reports in order to find out more about their customers. At Swisscom Natel Pay, the employees concerned are trying to replace laboriously created reports, which can only be carried out selectively, with automatic real time analysis. Predictive Analytics is being used in order to find out more about general customer behaviour at Natel Pay. Instead of simply accumulating individual characteristics, the aim is to create an overall picture. The project is only just beginning, but the tests are already revealing that the more is known about customer needs, the better the department can cater to the customer and act accordingly. Swisscom always handles this data carefully; after all, data protection remains the highest priority when handling customer data.

At Natel Pay, Predictive Analytics is used on various levels: with what is known as Carrier Billing, customers have the possibility of paying for digital services or products by mobile phone bill. In order to be able to create forecasts, previously hidden correlations must be uncovered in the data sets: which customers use which services? Are 25-year-old Android users more reliable payers than 40-year-old iPhone users? Is there a correlation between the type of mobile phone subscription and the type of product bought? The calculation model tries to give answers to many such questions. Depending on the result, more suitable products can be offered, discounts can be granted for preferred services, and spending limits can be imposed on customers.





In order to be able to answer all the questions, the data analysts have deployed several dozen variables and searched for correlations between all these variables. The variables contain diverse information, for example on demographics, user behaviour or previous transactions. Any employee can easily relate two variables to each other using an Excel table. It becomes more difficult, however, with dozens or even hundreds of variables. It is here that the potential of automated models lies: they can uncover hidden connections and unforeseen correlations.

“In order to understand the processes and find the complex connections between all the influencing factors, we first carried out an exploratory search, together with experts from the business, for possible variables which we could use to perform the calculations”, explains Martin Gutmann. It is also necessary to find out which data sources are suitable and how the data can be used. “Processing data is spadework and generally takes up the most time”, says Gutmann.

Debt losses are also analysed. For many companies, they are among the biggest unknowns. Payments for already delivered products are not made – this has been experienced by all mail-order companies which offer payment by invoice. In order to keep debt losses to a minimum, spending restraints can be imposed on individual customers. But which customers should be subjected to such a limit? Those who delay payment? Those who don’t pay at all? After all, loyal customers should not be driven away simply because of a single omission. For example, some customers regularly only pay their invoices a few days after the payment period has expired. These customers do not adhere strictly to the rules, but are nonetheless reliable and do not have to be alienated by unnecessary reminders. This not only reduces the administrative workload and therefore costs, but also improves the customer relationship in the long term. Using the Predictive Analytics method, such cases can be evaluated not only more accurately but also much faster and at less expense.


Fast analyses, continual learning

The examples described show how predictive data analyses generate a tangible benefit. Compared to manual analyses, Predictive Analytics is not only much faster and more exact, but also more objective: “For example, when employees create forecasts about future sales figures, psychology always plays a part. This means that subjective views or apparent correlations may influence the result”, says Martin Gutmann. The more data are available, the better the applicable model will be.

Data-supported analysis can sometimes dig up surprising correlations which cannot be discovered manually. However, in order to be able to interpret the causal connections, specialist knowledge from the relevant department is needed. Martin Gutmann knows this: “Because the experts’ view is important in order to obtain meaningful results and continually improve the system, the analysts are always in close contact with the business.”


“Analytics and data are the key to measurably better solutions.”



Martin Gutmann

Head of Analytics & Data Consulting





Newsletter

Subscribe to the newsletter now and stay informed about trends, industry news and benchmarks.





More on the topic