By Marcelo Cajias, Head of Investment Strategy & Data Intelligence
For decades, the mantra in the property sector has been that ‘location is everything’. It has endured because, on the whole, it is true. However, ‘location’ was for a long time a far less precise term than the confidence with which we use it would suggest.
For much of my professional career, the information available for assessing a location was relatively sparse. We relied on comparative data from estate agents and official statistics, supplemented by the assessments of people who knew the relevant market very well. Both remain important. However, we often had to deduce the characteristics of a street or a neighbourhood from a relatively small number of observations.
Artificial intelligence does not render this expertise obsolete. Its greater benefit lies in the fact that it provides us with a much broader data set against which we can verify such assessments.
Let’s take Munich as an example. For our latest housing market analysis, we evaluated around 188,000 rental listings from 2014 to 2026. To do this, we divided the city into 496 areas, each covering approximately one square kilometre, and took into account differences in the characteristics of individual residential properties in order to make a meaningful comparison of rent trends. The results show just how much can be lost in average figures at city or even district level.
Among the locations with stronger rent growth were also those with low rents in absolute terms. Furthermore, modernisation was associated with an additional increase in the annual rent growth rate, which in our model was more than twice as high as the average difference in rent growth rates between Munich’s city districts. Below the district level, even greater differences were evident between individual micro-locations.
This does not mean that the conventional assessment of the Munich market is incorrect. Rather, the results show that broad market definitions inevitably obscure some of the factors on which the performance of individual properties depends. Buying a property in the ‘right’ city or even in the ‘right’ district may therefore say less about the building’s prospects than we assume.
For investors, this allows for more precise questions to be asked. Instead of simply asking whether a district is attractive, we can examine the relationship between an individual property and its immediate surroundings and assess whether changes to the building or its location could significantly affect its prospects.
We have been refining this approach for several years. PATRIZIA’s Amenities Magnet, for example, uses data on local amenities, leisure facilities and infrastructure in the surrounding area to compare the attractiveness of individual locations. The key advance is that we are increasingly moving beyond simply measuring the current state of affairs and are able to examine what might happen if assumptions were to change.
In Berlin, we used the model to analyse a residential development project in Spandau. Compared with Berlin as a whole, the location scored 40 and did not initially appear particularly attractive. However, when compared with locations within a 15-minute cycle ride, the score rose to 78. This provided us with a different basis for assessing the location and for considering how the amenities planned as part of the project might influence its attractiveness over time.
We applied the same principle in Birmingham. There, by modelling changes to the services and facilities in the vicinity of a planned construction project, the location score rose from 60 to 72. The analysis also revealed which specific measures were key to this increase. The investment team was able to examine these in detail, rather than making a blanket assumption that additional services would improve the location.
In my view, it is precisely this aspect that is often overlooked in the discussion about AI in the property sector. The aim is to be able to assess investment theses more effectively, rather than replacing investment professionals with algorithms. If we believe that a micro-location has better prospects than neighbouring locations, we can critically examine the evidence for this. If a business plan assumes that modernisation will improve rental growth, we can test the scale of the potential effect before committing capital. Sometimes the analysis confirms the original thesis. Equally important: sometimes, however, it also calls the thesis into question.
However, there are clear limits to what such analyses can tell us. A model can identify a location with attractive rent-related characteristics without being able to assess, at the same time, whether a property there can be acquired at the right price or whether the business plan is actually feasible. An opportunity presented by a single building may also be difficult to replicate on an institutional scale.
Human judgement therefore remains crucial. Local market knowledge is still necessary, not because the data fails, but because an investment decision ultimately requires an understanding of both the individual property and its market. The model can refine the data basis and make it easier to critically examine assumptions, but it cannot take responsibility for the decision.
What has changed is the quality and level of detail of the information available for such decisions. Ten years ago, an investment committee might have discussed differing assessments of a location largely on the basis of comparable properties and market experience. Today, some of these assumptions can be examined at the level of individual properties or square kilometres before capital is committed.
This may be a less spectacular application of AI than some claims about the technology suggest. For property investors, however, it could prove far more useful in the long run to recognise what conventional definitions of ‘location’ fail to take into account.




