Business-Blog | adesso insurance solutions

For an AI project, data quality is crucial

Written by Wilm Tennagel | 24.05.2022

Insufficient data quality may not necessarily be immediately noticeable, but it can have serious consequences for a company. High data quality is the basis for many digitization projects and the development of new business models. This should be incentive enough for insurers to address the issue.

Small cause, big effect: The "Mars Climate Orbiter" was supposed to provide NASA with new insights into the atmosphere of Mars. But unfortunately, a unit error caused the orbiter to take a wrong course and burn up in space. An admittedly spectacular example of what insufficient data quality can lead to. However, when it comes to AI, insufficient data quality can have similarly serious consequences for insurers.

Al is not a marvel

The expectations placed in AI methods are high. Information from the IoT forms the basis for telematics tariffs or parametric insurance. Using artificial intelligence methods such as machine learning and predictive analytics, insurers want to find out more about policyholder behavior. AI analyses should provide important sources of inspiration for developing new products and calculating their profitability and risks.

For many decision-makers in companies, however, the term AI remains abstract. It seems to be a marvel of state-of-the-art IT, but it is easy to forget that AI is developed by humans. And they can continue to make mistakes, or they can incorporate their own views and values into the development of algorithms.

AI relies on data. It must be available in large quantities and provide correct information. And its not always as correct as it should be.

Bad data leads to bad assumptions

It is precisely because people make mistakes that information is repeatedly collected incorrectly. Transposed digits in an input mask or an incorrect selection in a form leads, for example, to incorrect addresses or salutations in the customer master data. An incorrect salutation in a mailing simply looks unprofessional and can be quickly rectified.

However, incorrect information is more serious in subsequent processes, such as when business-critical processes are based on insufficient data (such as risk analysis and risk management) or when it is incorporated into AI analyses.

In these cases, AI is no different from standard software, such as Photoshop. Even the best algorithm can't create a masterpiece from a bad photo. And the best AI analytics technology will lead the company down the wrong path if the data quality isn't right.

Many error sources: measuring data quality
 
Data quality can be measured using a number of parameters. These include, for example, completeness, correctness, up-to-dateness, unambiguity, and consistency. There are many error sources that lead to poor data quality in the everyday lives of insurers. There is the human error in collecting or processing information. Data silos also have a significant negative impact that almost inevitably leads to problems. This is because information can be redundant and sometimes contradictory.

And threats to data quality also lurk deep in the IT engine room. If headquarters and subsidiaries are not working on the same version level of their core solutions, changes to the data models can lead to discrepancies and contradictions in the data.

If the sources of error mentioned above as examples affect the entire data warehouse, the errors increase. And the overall quality of the data declines.

Insurers need data governance
 
Based on the criteria mentioned above, one can determine the overall quality of the available data. To improve it, insurers must adopt a two-pronged strategy. On the one hand, it is necessary to improve the data that is already available. This includes deleting redundant data after the information has been consolidated, deriving data from other information, or completing it based on alternative sources.
 
For future data, data governance must be established. It consists of various elements, such as concrete instructions for the correct collection of data, but also software support during collecting and processing in order to ensure that error sources are reduced.
 
However, data governance often also requires a change in mindset. In the age of data-centric work, data owners must also be taught how data quality is important to the entire organization. This enables them to view the collection of (additional) information that from their view has no additional use, as a burden. One needs the motivation to also see data quality as a corporate asset. This is because it forms the basis for reliable reporting and AI projects.

Find out why insurance companies are choosing to use AI and how it is done successfully in our free whitepaper (German only).