NextGen Scoring: Schufa aims to standardize scores and simplify understanding
Despite Schufa's transparency initiative, its score values are still complex. Over 50 scores are now to be replaced by one in the interests of explainability.
As part of its transparency initiative launched in 2022, Schufa wants to massively reduce the breadth of its scores and make them easier to explain. To date, the credit agency has used well over 50 of these creditworthiness indicators. These are now to be replaced by a single one, with around 10 to 15 influencing factors, in the interests of traceability. In the future, Schufa wants to be able to illustrate in its online score simulator how, for example, concluding a contract will affect the credit rating. So far, it can only say: "It depends on the individual case."
Schufa provided an initial insight into this ongoing Next Generation Scoring (NGS) project on Thursday. When the company launched its score simulator, it was easy for all sides to see that it could not be "completely satisfactory", explained CEO Tanja Birkholz, also referring to criticism from politicians. It is important to know that there is not just one Schufa score, but six industry scores alone. In addition, there are others for specific transactions and "customer-specific score models", resulting in over 50. The best known of these is the bank score: 62 percent of the values are industry scores with such financial institutions.
According to Birkholz, the previous system was geared towards "very good selectivity" and a "special forecasting ability". It was therefore primarily intended to express how high the default risks are and at the same time create over-indebtedness protection for consumers. The factors vary depending on the model: each individual factor can be weighted differently. As a bridge, the credit agency has therefore created the Schufa basic score, which has been shown in the self-disclosure since 2006. However, as a "weighted average", this "does not actually play a role" in decisions, explained the Group CEO. However, it was the inspiration for the simulator together with the bank score.
"One score for all"
With the NGS project, which has been running since 2022, Schufa is now pursuing the claim: "One score for all." Only the very individual models are excluded. It should be easy to see online how the score changes if a consumer takes out another loan or applies for a credit card with a new bank. So far, six different influencing factors have come into play in the latter case alone, reports Andre Muhle, head of the decision-making strategies department at Schufa. These include the number of credit cards, the age of the oldest and the number of stored bank contracts and current accounts. Each of these factors is "statistically highly significant". When they interact, it becomes complex. From a scientific point of view, this statistical procedure is correct, but ultimately leads to an unsatisfactory situation for all sides.
With NGS, Schufa now wants to make this structure more comprehensible by removing the aforementioned interactions as far as possible. In the case of a credit card application, "a maximum of two to three influencing factors" remain, Muhle illustrated using a bar-line diagram for concluded bank contracts. The proportion of consumers who have four to five is around 30 percent. A risk group is calculated to determine how many end customers are unlikely to be able to meet their obligations. Up to five contracts, there is no great risk, but the sixth contract leads to a deterioration in the score.
As a solution to simplify the model, Muhle said to focus on monotonic risk progressions. Almost 90 percent of consumers had not taken out a loan in the previous year, which indicates a below-average risk. If someone had entered into one or more such contracts, this would result in "monotonically increasing risks". This is easier to understand.
Schufa gives AI the cold shoulder in scoring
The new statistical approach will be "validated once again" and also scientifically assessed externally, Muhle assured. There is "not yet the final hook on the final model". The previous industry scores were developed in 2016 based on historical data from 2014, he explained. Aspects such as the elimination of the so-called positive data in the telecommunications sector had been considered meanwhile. In NGS, Schufa is now taking into account "what has happened in the market in recent years", i.e. above all the increase in comparison portals as well as more microloans and e-commerce.
This alone "would have already led to a significant improvement" in the score, Muhle clarified. On the other hand, many aspects "that we had identified as being difficult to explain" would now no longer apply. For example, the number of credit cards. This reduces the higher selectivity again, so that Schufa will ultimately maintain the level for indicating the risk of payment default. In general, "we continue to operate in the Schufa data cosmos", emphasized the insider. External sources are not included. In principle, the variables would continue to include installment loans, credit cards and mortgages.
Birkholz emphasized that Schufa would not use complex models with artificial intelligence (AI). These would be associated with a further "loss of explainability". The well-known logistic regression, with which the probability of an event occurring is estimated because of a given data set of independent variables, would remain. The company has also started to give customers "insight into the complete algorithm" upon submission of a confidentiality agreement.
NGS to launch at the end of 2024 - transition period of several years
Birkholz wants to be on the market with NGS by the end of 2024. She anticipates that it will then take two to three years for banks & co. to understand the changes and go through the necessary certification processes with the supervisory authorities. Until then, it is quite conceivable that a financial institution could use both score models in parallel: "one productively, the other for checks". The Schufa simulator will also be gradually adapted. Since May, it has included new scenarios to give users a feeling for the effects of changing a current account, for example. In September, a free identity theft protection service will also be added to the website, and the subscription packages are also being constantly expanded. In addition to English, Turkish will be added as a second foreign language. All functions will also soon be offered via a smartphone app.
According to her boss, Schufa has not suffered a decline in demand for scores following the ruling of the European Court of Justice on problems with credit checks. Customers have been approached, "a few have made changes to their process routes". The credit agency generally sees itself more as an intermediary whose credit rating "should not be a knock-out criterion" for contracts with consumers. Following reports of mass fraud with stolen account data when purchasing the Deutschlandticket, "one or two companies" have also come forward. Those affected apparently realized what could happen if they did without lengthy credit checks and Schufa information.
(nie)