report-consumer-friendly-scoring
Dieses Dokument ist Teil der Anfrage „Gutachten des Sachverständigenrats für Verbraucherfragen“
16 About this report
II. Scores and scoring
Among the best-known forms of scoring are credit scores, the realm of consumer policy is on novel forms of scor-
which are assigned to individual consumers by private ing.7 Especially as a result of progress in what is known
credit reference agencies. Credit reference agencies col- as narrow artificial intelligence (Ashley, 2017; Goodfel-
late a wide range of consumer data, such as information low, Bengio and Courville, 2016; Jentzsch, 2018; Nilsson,
on a consumer’s credit history, records of so-called pay- 2009; Russell and Norvig, 2010; Witten, Frank, Hall and
ment irregularities and personal information about the Pal, 2016) there is new scope in the field of machine
consumer. Some credit reference agencies also include learning for automated data analysis based on pattern
data concerning the area in which the consumer lives recognition.
(geo-scoring).5 From this collection of data the agency
derives a probability rating or behavioural prediction on These new forms of scoring, the effects of which are still lit-
the individual’s creditworthiness or the likelihood that tle known and which regulators have largelynot yet analyz-
the loan will be repaid (see, for example, Schröder, Lang, ed, are one of the focal points of this report. One example
Lerbs and Radev, 2014; Unabhängiges Landeszentrum that is examined in detail relates to the telematics-based
für Datenschutz Schleswig-Holstein and GP Forschungs- system of pay-as-you-drive (PAYD) insurance cover, which
gruppe, 2014). Creditworthiness is expressed in a score is already an integral component of some motor insurance
which can be used as an aid by banks, for instance, when products. In this type of system, insurers – or, as the case
deciding whether to grant loans or by online traders may be, their contracted data-analysis agencies – record,
when determining which payment options to offer a par- by means of a smartphone app, for example, details of
ticular customer. Depending on the score, in other words policyholders’ driving behaviour, including data on jour-
the degree of creditworthiness, of a given prospective ney times and routes, and communicate to the driver and
customer, online shops offer easier – albeit riskier from his insurer a score indicating how safely the vehicle has
the shop’s point of view – payment terms, such as pur- been driven. Especially ‘good’ drivers, in other words those
chase on account, to some customers but not to others. whose scores exceed a certain threshold, then receive a
In e-commerce, scores are used primarily for the purpose discount on their insurance premiums. The scores in this
of fraud detection, to distinguish notoriously bad payers case serve not only to predict drivers’ behaviour but can
from customers with default-free records. Otherwise the also be used by insurers for the express purpose of modi-
latter would have to help foot the bill for losses caused fying that behaviour. At the present time there is no such
by the former, which, despite fraud detection, amounted thing as a pure PAYD tariff, in which the rate of premiums
in 2017 alone to more than 2.5 billion euros, according depends entirely on the registered score. Instead, these
to the German e-Commerce and Distance Selling Trade are always offered as an optional addition to a motor in-
Association (bevh).6 surance policy with conventionally calculated premiums.
In the debate on consumer policy, scoring has featured Other examples that are described in detail in this report
predominantly in the context of credit checking (see, are from the realm of health care. Many statutory health
for example, Bala and Schuldzinski, 2017; Oehler, 2017; insurance funds reward their members with a credit note
Unabhängiges Landeszentrum für Datenschutz Schle- or other inducement if they collect scoring points in a bo-
swig-Holstein and GP Forschungsgruppe, 2014). Against nus programme by engaging in any of a predefined set
the backdrop of the progressive development of digital of healthy activities – preventive measures such as phys-
technology and the use of algorithmic decision-making ical exercise, which may be recorded, for example, by a
methods in other areas of business, such as health and fitness tracker, inoculations and attendance at health
motor insurance, however, the focus of today’s debate in courses (Baun and Nürnberg, 2015). According to an
5 According to information from the German credit bureau SCHUFA, however, this occurs only in “a few exceptional cases” in which no other information is available.
SCHUFA website, https://www.schufa.de/de/ueber-uns/daten-scoring/scoring/scoring-schufa/
6 https://www.bevh.org/nc/veranstaltungen/details/datum/2017/oktober/artikel/betrug-im-e-commerce-und-datenschutz/fe_
pw/?cHash=b38d2f5410c449712c2e5c2a4c0b0b1b&sword_list%5B0%5D=betrug; accessed on 3 September 2018.
7 Development of “the latest technological processes such as machine learning” for fraud prevention, among other things, is also benefiting credit reference agencies.
Schufa website, URL https://schufa-wegbereiter.de/de/digital/innovationen-labor/innovationen-aus-dem-labor.jsp; accessed on 17 August 2018.
About this report 17
opinion delivered by the Federal Insurance Office (Bun- Although other scoring processes in the health sector
desversicherungsamt – BVA), however, there is a need to other than the Vitality Programme are still few and far be-
take a close critical look at the actual beneficial health tween, growing public acceptance of self-measurement,
effects of the defined activities (Bundesversicherung- including by means of wearable devices, which is dis-
samt, 2018). The statutory health insurance funds are cussed in The Quantified Self and Self-Tracking (see,
bound by the provisions of section 20(3) of Book V of for example, Lupton, 2016, and Selke, 2014) seems to
the German Social Code (Sozialgesetzbuch) to operate indicate that this could change in future. Providers are
within narrow limits when selecting eligible activities. already cottoning on to the fact that many people now
Both private health insurers and other non-statutory record and analyse their physical performance data for
insurers are freer to shape their own scoring-based pre- the purpose of improving their fitness levels. This prac-
vention and fitness programmes; the Generali insurance tice is encouraged by statements such as the one made
company, for example, is planning to offer what it calls by the start-up business Dacadoo to the effect that a
the Vitality Programme in 2019 within its range of private person’s health is improved by the use of apps which
insurance products.8 Part of the programme is based on convey the user’s state of health in the form of a numer-
health scoring, whereby both participation in preven- ical value.13 This development is part of a general ten-
tive measures and transmission of vital parameters earn dency in preventive health care to focus increasingly on
points, which are redeemed with vouchers from partner early detection and preventive health-promoting action
companies and lower insurance premiums. Anyone who as a supplement to curative treatment (see, for example,
signs a non-smoker’s declaration, for example, earns GKV-Spitenverband, 2017). A study commissioned by
4,000 points in a year; anyone buying “healthy foods” the Federal Ministry of Health found that there was still
(fruit, vegetables or fish) from a cooperating online shop a lack of evidence in the form of robust studies which
can earn up to four times as many points in a year.9 would allow a conclusive identification of beneficial
health effects of fitness apps, particularly of the longev-
What is not clear are the criteria used to weight individual ity of any such effects (Albrecht, 2016).
activities and the extent to which the score is actually a
valid basis for conclusions regarding the improvement of Regulatory measures such as the e-Heath Act of 1 Janu-
a person’s state of health. Generali itself sets out its stall ary 2016, designed to establish a modern IT indrastruc-
clearly, emphasising the scientific basis of the Vitality ture in the health sector, and the loosening, which took
Programme (“Vitality is a health programme based on effect on 10 May 2018, of the ban on remote treatment to
scientific findings”10) and advertising on that basis for the allow the practice of telemedicine, for example in video
health effects of participation in the Vitality Programme consultations, are also indicative of a general increase in
(“Vitality members have lower health costs”11). However, the use of digital services in the health sector (see also
impact studies showing evidence of actual improvements Gigerenzer, Schlegel-Matthies and Wagner, 2016). On the
in people’s health resulting from participation in the Vi- one hand, this is a gain for patients such as those living
tality Programme in Germany in particular, including in rural areas with few doctors and even fewer special-
comparisons with a randomised control group, have yet ists; on the other hand, these technological solutions
to materialise, which has prompted critics to describe Vi- create a data problem on an unprecedented scale.
tality as a kind of cashback scheme using health data.12
8 Because of regulatory requirements, these are confined for the time being to the fields of disability and occupational disability insurance and term life insurance.
9 URL https://static01.cosmosdirekt.de/CosmosCAE/S/linkableblob/home/213750.1525232169000/data/Antragsinformationen- zur-Generali-Vitality-Mitgliedschaft-
data.pdf; accessed on 4 September 2018.
10 URL https://www.generalivitality.de/vmp/so_funktioniert_vitality; accessed on 4 September 2018.
11 German text at URL https://www.generalivitality.de/vmp/so_funktioniert_vitality; accessed on 4 September 2018.
12 URL https://www.sueddeutsche.de/wirtschaft/versicherung-wer-sich-bewegt-zahlt-weniger-1.2920176; accessed on 4 September 2018.
13 URL https://info.dacadoo.com/de/unternehmenslosungen/life-health-insurance-solutions/; accessed on 19 June 2018.
18 About this report
A score that provides information on a person’s own A similar situation exists with regard to the new telemat-
state of health can certainly be regarded as a means of ics-based tariffs for motor insurance. On the one hand,
consumer empowerment, as it reduces the information they can lead to greater safety and better traffic flow, less
asymmetry between doctors and patients. It is question- information asymmetry and more efficient markets. In-
able, however, whether the same applies to the informa- surers advertise that continuous recording and analysis
tion asymmetry between consumers and companies; on of speed and acceleration data encourage careful driving,
this point, the German Ethics Council takes the view that thereby contributing to greater road safety.14 The analysis
such asymmetry is more likely to be widened by the use of individuals’ driving behaviour can also serve as a basis
of big data, which potentially enable companies to find for more risk-related rewards and discounts (Baecke and
out more about their customers (German Ethics Council, Bocca, 2017; Bian, Yang, Zhao and Liang, 2018), which
2017; cf. Weichert, 2018). Both research and implemen- can be especially beneficial to young drivers, who are
tation, however, are still at quite an early stage, and be- otherwise charged very high premiums. On the whole, so
sides highlighting the opportunities, it is worth sounding the argument goes, driving analysis allows a more pre-
an early warning of the risks which may not surface until cise actuarial cost calculation for motor insurance (see,
later and which can easily be overlooked by consumers for example, Baecke and Bocca, 2017; Bitkom, 2014; Kraft
because of the immediate benefits. and Hering, 2017). Another socially desirable potential
benefit of more careful driving as a result of scoring con-
In contrast to the discussion of these relatively new sists in a reduction in congestion and environmental pol-
applications of scoring, the debate in the field of con- lution (Kraft and Hering, 2017; Litman, 2005).
sumer and market policy on the role of credit scoring
in the financial sector has been going on for many dec- Concerns are expressed to the effect that constant record-
ades. There is generally a good stock of literature, on ing and analysis of driving behaviour can lead to increas-
the basis of which the macrosocial advantages of credit ing surveillance by commercial insurance firms (see, for
scoring may be summed up as follows: the use of credit example, Stiftung Warentest, 2014, and Verbraucherzen-
scores reduces loan defaults; it lowers transaction costs trale Bayern, 2016). Last but not least, the criticism is quite
and therefore has a major impact on the efficiency of often made that, while telematics-based deals benefit
financial markets (Schröder et al., 2014). In addition, consumers, their main beneficiaries are insurers them-
credit scores can help to reduce information asym- selves, which exploit the increased opportunities to ad-
metries that exist between borrowers and lenders and dress consumers directly, through push notifications on
to prevent credit rationing, because they give lenders smartphones for instance, as a means of customer reten-
the vital information they need about prospective bor- tion (see, for example, der Weidner and Transchel, 2015).
rowers when it comes to granting loans (Schröder et al.,
2014). In the realm of online shopping, credit scores play To put it plainly, this report does not contest the fact that
a major part in the detection and prevention of online scoring performs an important function in business and
fraud (Bolton and Hand, 2002; Marschall, Morawitzky, society. The real question for the SVRV is how scoring is
Reutter, Schwartz and Baars, 2015). On the other hand, and should be designed. Scoring-based business models
there are legitimate concerns about data sovereignty, are normal today,15even though they are applied in var-
for example, or about discrimination against particular ying depth, and can bring many benefits for individual
groups, as we explained in previous reports, such as consumers and for markets in general. They also entail
SVRV, 2017a. risks, of course, some of which are already obvious, while
others are only just beginning to emerge and, given the
rapid speed of technological development, cannot by
any means be definitively assessed.
14 https://www.cosmosdirekt.de/veroeffentlichungen/versicherungstipp-telematik-198254/; accessed on 19 June 2018.
15 See also footnote 1, which refers to applications of scoring that are not dealt with in detail in the report, namely micro-targeting by onlne shops, robo-advisers
helping with the selection of financial products, applicant scoring, predictive policing and, in particular, the individualised control of social media by their providers.
About this report 19
This report highlights the key challenges of scoring-based In a longer-term perspective, this report also considers
business models and makes recommendations on polit- the development of so-called super scores, in other words
ical measures that can strengthen the position of con- scores that not only assess consumer behaviour within a
sumers. At the heart of the report is the concept of con- limited area such as finance, mobility or health but as-
sumer-friendly scoring, and our task is to describe that sess it across the board. We shall look closely at Chinese
concept (see also Mittelstadt, Allo, Taddeo, Wachter and pilot projects for a system of social credits in which, from
Floridi, 2016) and to discuss the following questions: 2020 onwards, all citizens of the People’s Republic are
to be assigned an individual score that will take account
• What does fairness mean in the context of scor- of behaviour patterns in all areas of their lives (Kostka,
ing-based business models? 2018). In view of the differing political and legal systems,
• Which data should be included in scores, and this model is not transferable to the Western world at the
which should be excluded? present time, nor will it be for the foreseeable future, but
• Which statistical quality criteria should scores it nevertheless provides food for thought on what is tech-
meet? nically feasible and what is socially acceptable and unac-
• Which assessment criteria are relevant to con- ceptable. Germany’s President Frank-Walter Steinmeier
sumer-friendly scoring? spoke in similar terms on 15 February 2018: “There is no
• What does discrimination through scoring threat of such a thing [as the Chinese system] happening
mean, and where and how does it occur? in Germany, but it goes to show how important it is that
• Which elements of scores should be known, we engage in detailed discussion on the social implica-
which should be made transparent and compre- tions of technological developments.”16
hensible, and which should not?
• Which forms of transparency and monitoring
should there be to ensure that scoring process-
es maintain or improve the enforcement of
consumer interests? Are the existing processes
adequate?
• Which institutions lend themselves to the tasks
of creating transparency and monitoring?
16 German text at http://www.bundespraesident.de/SharedDocs/Reden/DE/Frank-Walter-Steinmeier/Reden/2018/02/180215-Leopoldina-Sachsen-Anhalt.html;
accessed on 1 October 2018.
20 About this report
III. Objectives of the report
The particular relevance of the subject of scoring to There are many processes that permit an automated as-
consumers, society and business is due to three main sessment of whether, for instance, a person represents a
factors: the increasing availability of individualised com- high credit risk or is a good driver with a low accident risk:
sumer data, the spread of methods that can be used by these range from simple rules of thumb – also known as
businesses to process these accumulations of data and heuristic approaches – through standard statistical es-
to profile data subjects and the resultant growing num- timation methods such as logistic regression, which is
ber of applications for scoring. Novel scoring methods, used in credit-scoring practice and constitutes a fairly
moreover, no longer serve merely as a predictive tool simple form of machine learning, to new deep-learning
but are increasingly being used to guide consumer be- methods, such as those based on neural networks, which
haviour too. In the present report we are pursuing three process patterns and correlative connections between
specific core objectives: numerous variables (e. g. consumer attributes) based on
large data sets in largely automated operations.
From the perspective of consumer policy, it is clear that
the less transparent a process is, the harder it is for super-
visory authorities to oversee, which also jeopardises its
Objective 1: Improve the comprehensibility to consumers. Whether the use of ma-
chine learning will inevitably reduce the fairness of scores
information base and or could increase it is the subject of intense discussion.
increase knowledge of
In this report we not only discuss the algorithms that
scoring are currently used in consumer scoring but also focus
particularly on individual and institutional means of
A new feature of the information base is that searches reviewing and regulating these algorithms. Special em-
for consumer data are no longer confined to simple facts phasis is laid on the establishment of minimum quality
such as age and current loan agreements but can target standards for model scoring methods and on issues of
far more detailed variables, such as vital parameters and comprehensibility for consumers.
driving behaviour, with the aid of new methods. At the
same time, inexpensive means of data storage are con- On the whole, the increasing availability of data and the
stantly developing. The growth of business models for spread of new methods are making it possible to devel-
consumer-data brokerage by companies such as acxiom op novel applications of scoring processes that go far
and Oracle, moreover, indicates there is already, in prin- beyond traditional credit scoring. This report examines
ciple, sufficient business interest and know-how to drive these by reference to examples from the fields of mo-
the compilation of extensive data sets on consumers – if bility (telematics in motor insurance) and health (bonus
that became legally permissible – which might enable programmes of statutory health-insurance funds, first
companies to develop far more complex scores that steps towards health scores on the part of private health
were no longer be confined to a single area of activity, insurers and health scores already used by companies
such as finance, but covered many aspects of people’s such as Dacadoo). Whereas traditional credit scoring is
lives. In this way they would no longer merely lay the limited to the prediction of future behaviour, novel scor-
foundations for – or even directly make – decisions relat- ing schemes that analyse exercise habits and, in some
ing to people as consumers in a demarcated sector but cases, vital parameters and communicate the resulting
would determine the individual’s stake in the economy score ‘live’ to the consumer are much more likely to have
and society in general. a behaviour-modifying effect too. The question whether
this is desirable for society as a whole must be discussed.
About this report 21
ents. The data were subsequently processed and ana-
Objective 2: Broaden the lysed by the SVRV.
empirical basis and address Besides these empirical studies, a legal study, described
legal issues in Part E, addressed the data-protection issues in detail,
examined the current rules governing scoring and similar
With this report, the SVRV is fulfilling its mission of cre- practices in the market segments of credit, motor insur-
ating a broad empirical basis for a well-informed and ance and health and added a set of draft building blocks
forward-looking consumer policy. The report focuses for legal provisions regulating scoring in general. Part E
on the dynamic area of novel forms of scoring. concludes with reflections on enforceability and oversight.
Following an outline of the current state of research in
Part B, the SVRV presents a comprehensive study of the
market segments of credit checking and mobility in Part
C. In the first of these market segments, scoring meth-
ods are already long-established, while in the other they Objective 3: Suggest rules for
have been gaining an increasingly firm foothold in re-
cent years. This study also examines the extent to which
consumer-friendly scoring
scoring methods have already taken root in the sphere of
statutory and private health insurance. As this has only In particular, the SVRV wishes to use this report to
happened to a limited extent, we could use the term propose criteria for consumer-friendly scoring as a
‘proto-scoring’ in this context. A total of three question- basis for discussion. In the view of the SVRV, scoring is
naires were devised for the purposes of this study and consumer-friendly if scores are presented comprehen-
were sent to all identified credit reference agencies, mo- sibly to consumers, if awareness of scoring and scor-
tor insurers, statutory health-insurance funds and private ing skills are sufficiently available, if discriminatory
health insurers. The questionnaires, which were complet- elements are probed and revealed, if telematics-free
ed in written form, were digitised, and the replies were options are available and will remain so in the future
coded for comparability and aggregation and analysed. without significant disadvantages, if the quality of
The main purposes of this study were to examine the scores and data is guaranteed, if supervision of scor-
penetration of these three market segments by scoring ing is significantly improved and if the use of super
practices and also to investigate which consumer char- scores is effectively prevented.
acteristics were normally recorded and used to calculate
the individual scores and which quality criteria were met
by the algorithm behind the scoring. To add more depth
to the discussion, background talks were conducted with
individual companies and experts from the three sectors.
In addition, the SVRV devised a public survey and had
it conducted to find out more about awareness and ac-
ceptance among the German population of established
scoring practices as well as those that are technically
feasible in principle but not yet established; this is de-
scribed in Part D. In cooperation with a social and market
research company, the infas Institute for Applied Social
Sciences, a representative survey was conducted by
means of the CATI (computer-assisted telephone inter-
view) method. The sample comprised 2,215 respond-
22 About this report
The history of scoring
In historical terms, the desire to assess individuals’ the most exciting possible spectacle and the fairest pos-
characteristics, behaviour and preferences as precisely sible match. In boxing and weightlifting, for instance,
as possible and to infer future developments from that competitors are assigned to particular weight divisions,
information is nothing new. Even in the analogue world, golfers receive handicaps, and tennis players, for exam-
conclusions were – and still are – drawn about individu- ple, are seeded. Lastly, sportsmen and sportswomen
als from certain characteristics and modes of behaviour, are characterised time and again by their scores. One
and in some cases numerical values are assigned to peo- example is Armin Hary17, the first athlete to run 100 me-
ple on that basis. Digitisation, involving complex algo- tres in 10.0 seconds. The inofficial electronic measure
rithms and broad information bases, has merely inject- of his sprint – which was more precise than the official
ed fresh impetus into an old practice, and some believe hand-held stopwatch – showed that he had taken about
that artificial intelligence (AI), with its auto-adaptive 10.2 seconds with a borderline tailwind, and his case
algorithms, has added another new dimension. may therefore be used to illustrate the measurement
problems that can arise with all sorts of scores.
The assessment of performance and abilities by means
of numerical values or standardised verbal formulas has Particularly in business deals, in which contracts are very
a long tradition. In German schools, for example, marks frequently concluded with hitherto unknown partners
have been awarded since the 16th century (Lintorf, 2012) and a certain leap of faith must be made, risk minimisa-
and not only for learning achievements but also – as is tion plays a major role, which is why great importance
now planned for all citizens in China (Kostka, 2018) – for has attached to scoring in this domain for many decades.
social behaviour. Today, assessments of individuals’ at- Businesses need to inform themselves about the reliabili-
tainments and learning successes still have consequenc- ty and creditworthiness of customers, and in the 19th cen-
es such as admission to a higher class or a particular type tury this need led to the emergence of the first credit ref-
of secondary school or the award of a final certificate, erence agencies. Among the first such agencies in Europe
such as the German Abitur. A pupil’s mark in that final were Wys Muller, founded in 1861, Schimmelpfeng, found-
examination is one of the key criteria for university ad- ed in 1872, and Creditreform, founded in 1879. These
mission. The Abitur grade is calculated as an average of agencies collected economically relevant information on
the pupil’s grades in all subjects, although some grades, individuals and companies and sold it to businesses and
for example those obtained in the pupil’s main subjects, banks. Since then, credit reference agencies have been a
are weighted more heavily than others. In the admission cornerstone of every functioning credit system.
procedure, the applicants with the best Abitur grades –
usually in combination with their previous waiting time The first attempts to quantify people’s default risk and
for a university place – are accepted until the number present it as a numerical value were made in the 1940s.
of available places is exhausted (the system of numer- Until then, rudimentary scoring systems, operated by
us clausus). In this context it becomes particularly clear mail-order firms for example, comprised a catalogue of
that the Abitur grade not only represents an assessment criteria with the aid of which sellers would verify fulfil-
of past performance but is also credited with predictive ment of a number of conditions and tally the number of
power as an indicator of future attainment. A good Abi- ticked boxes (Thomas, Crook and Edelman, 2017). In a
tur grade is supposed to show that the student may be research project in 1941, mathematician David Durand
expected to perform well and is likely to obtain a degree. became the first to use discriminant analysis to deter-
mine the default risk of loans (Durant, 1941). He ana-
Another area that traditionally depends on the meas- lysed data sets on previously granted loans to identify
urement of human performance is competitive sport. the decisive factors that had led to smooth repayment
Not only performances are measured, however; in and those that had been responsible for repayment dif-
many disciplines people themselves are measured too, ficulties and developed a credit score. The first firm to
so that they can be categorised for the sake of creating develop statistical models for granting credit on a com-
17 Abgerufen am 1. Oktober 2018 von URL https://de.wikipedia.org/wiki/Armin_Hary.
About this report 23
mercial basis was Fair, Isaac and Company, now known es. When a particular number of points is amassed, the
as FICO, in California. From the 1950s it sold scoring Authority issues warnings, orders drivers to attend driver
products to financial institutions, retailers and mail-or- fitness seminars or withdraws their driving licence (Kraft-
der firms (Dixon and Gellmann, 2014). fahrt-Bundesamt, 2017).
In subsequent decades, mathematical advances were The forms of scoring described above have a long history
accompanied by innovations in electronic data pro- and already existed, to be sure, in the analogue world.
cessing, which ultimately paved the way for largely au- Yet it is also undeniable that the way in which scoring is
tomated credit scoring. The combination of computers carried out has changed radically with the technological
and algorithms as well as the experience of businesses developments of the digital age. In France, for instance,
that had seen a sharp reduction in default rates for their the allocation of university places has been regulated
loans and in frauds led to the scoring products of credit since 2018 by a scoring algorithm known as Parcoursup,
reference agencies with which we are familiar today. which analyses the applicant’s fulfilment of the entrance
requirements, place of residence and preferences (Jo-
Another area in which such forms of risk assessment eres, 2018). In the realm of e-commerce, a consumer’s
have long been established is that of insurance, where creditworthiness can be calculated automatically in a
the primary purpose of scoring was, and still is, to cal- matter of seconds and appropriate payment options of-
culate sums assured and contribution rates for each fered. And in motor insurance, we have seen the advent
individual customer. Back in the 1920s and 1930s, of telematics-based tariffs, in which driving behaviour
German health insurers became interested in putting is constantly evaluated and scored and premiums are
the calculation of health insurance contributions on a adapted accordingly.
sound mathematical and statistical footing. Using what
are called morbidity tables, insurers found that medical Algorithm-based scoring, moreover, is being used in-
costs could be expected to vary depending on a person’s creasingly in many new areas, assessing consumers and
sex, age and occupation (Wagner-Braun, 2002). Even to- groups of consumers in the widest variety of ways and
day, premiums for private health policies are calculated with the most diverse consequences (Dixon and Gell-
individually on enrolment. The same applies to term life mann, 2014). There are scores that predict households’
insurance and to occupational disability insurance. Con- purchasing power or propensity to spend (Equifax, 2018,
sumers are categorised on the basis of a combination of and Blackbaud, 2014), scores that indicate whether
individual characteristics, such as age and medical his- consumers will switch their custom to other companies
tory, the risk to the insurer is assessed on that basis, and (Versium Analytics, Inc., 2018), scores designed to detect
the premiums are calculated accordingly. pregnancies (Duhigg, 2012) and scores that measure en-
ergy consumption behaviour (Trove, 2018). Dating ser-
The calculation of premiums is normally particularly vices are based on scores which quantify how closely
complex in the case of motor insurance, where tariffs are personal profiles are matched (Carr, 2016).
tailored to individual customers on the basis of numer-
ous criteria. Among the key factors are the vehicle mod- A culture of assessment and quantification is devel-
el type, regional weighting and the driver’s no-claims oping (Mau, 2017). Whether it is Likes on Facebook,
history as well as characteristics such as the number of the number of followers on Twitter or stars on Airbnb,
drivers, the drivers’ ages, the age of the vehicle, its mile- scoring is no longer the preserve of companies who as-
age and where it is kept (Gesamtverband der Deutschen sess consumers and assign them numerical values – it
Versicherungs-wirtschaft e. V., 2016). And there is yet an- has become an everyday activity.
other scoring system for drivers, namely the driver fitness
assessment system administered by the Federal Motor
Transport Authority (Kraftfahrt-Bundesamt) in Flensburg,
commonly known as the Flensburg points system. Since
1974, the Authority has been entering penalty points in
a register for administrative and criminal traffic offenc-
24 About this report
Areas for action: the state of research 25
B
Areas for action:
the state of research
