21st Real Estate (21st) is a company that offers RELAS (Real Estate Location Analytics Software), a market price and location analysis software for professional users. 21st's target groups include project developers, asset managers, banks and valuers. Data provides the foundation of the business model. 21st's data division aggregates, validates and cleanses data. 21st provides this data nationwide for Germany in RELAS. The database covers almost all types of use, such as residential and commercial real estate, senior living, student living and logistics.

For 21st, data related to real estate markets are of primary relevance. In addition to supply and transaction prices in the rental and purchase segment, this primarily includes data that serve the valuation of a property: economic indicators such as income data or employment structure, demographic data e.g. on age groups and population forecasts, or ESG indicators such as energy consumption or climatic changes. The data basis of the situation indicators for the macro and micro situation includes rates of change to show positive and negative developments. 30-, 60-, and 90-minute catchment areas are included to achieve synergy effects from neighboring communities. In addition, digitized rent indexes and maps of standard land values or cadastral data provide further aids for a well-founded location analysis.

21st researches and categorizes data through strategic partnerships with recognized and long-established data suppliers. These include Immoscout and the Gesellschaft für Konsumforschung (GfK). In addition, 21st uses publicly available data from state statistical offices or industry directories. 21st invests a 6-digit amount in data procurement every year.

The data in RELAS is available for every municipality in Germany. This enables a comparison of locations within a city or a comparison across municipalities in the microlocation. The data is processed according to a strict control principle. In order to provide data in RELAS, enough data points must be available to determine representative values. Data are validated and adjusted for outliers to smooth out non-relevant fluctuations. The data is updated regularly, although the respective update cycles vary. Google's points of interest are always updated on a daily basis, comparables are updated monthly, and the "green spaces" location indicator, for example, is adjusted once a year, since few changes are observed here.

RELAS scoring allows comparative values on a micro and macro level. In the microscale, scores are evaluated within a tile system of 200 x 200 meters, while the nearly 11,000 communities in Germany represent the macroscale. The scores range from 0 to 100, covering the entire range of values in the raw data. All values in between are equally distributed within the range. A tile score of 23 - for example, for the "supermarket" indicator in Potsdam - indicates that the location in this tile has comparatively low access to supermarkets. Conversely, there are 77 percent of tiles in Potsdam that have better accessibility to supermarkets. A scoring value of 91 for annual housing completions in Bonn within a 90-minute drive says that only nine percent of the other communities within a 90-minute drive have a higher number of housing completions.

21st's scoring procedure allows this figure to be compared both within the city of Leipzig and in a regional or national comparison. Looking at Leipzig within Saxony, the state's largest city performs better than other cities, receiving a score of 68 for purchasing power per capita. Comparing Leipzig's purchasing power per capita with all other municipalities in Germany, Leipzig receives a score of 26. Looking at Leipzig in the context of the other 13 B-cities, Leipzig receives a score of 8, putting it in second-to-last place among the B-cities. Further comparative possibilities for the score distribution in the macro location result from the definition of the context. It is therefore crucial for a well-founded market price and location analysis to always consider the data in the appropriate context in order to be able to make well-founded decisions.

Machine learning algorithms are particularly superior when the supply density is low and the characteristics of properties (year of construction, area, condition) exhibit strong regional heterogeneity. Compared to the standard comparable value methods, this is especially the case for retail and office rents. For retail rents, an average improvement in price determination of 32 percent and for office rents even 46 percent could be demonstrated. Rents for apartments and houses can be calculated more accurately by 23 percent, as can house purchase prices. For the purchase prices of condominiums, the improvement is even 26 percent. We have been experts in machine learning for 4 years and have regularly tested our results together in the market and with customers. In the end, all users were always convinced of the reliability of machine learning prices, from transaction managers to appraisers.