The first months of 2025 brought contradictory signals for the Russian commercial real estate market. On the one hand, analysts are recording an increase in the share of online sales and predicting a decrease in traffic to shopping centers by up to 5%. On the other hand, shopping centers continue to transform into multifunctional spaces, which may stimulate visitor interest. Developers and management companies have to make increasingly complex decisions. In the context of changing consumer behavior and general market changes, it is no longer enough to simply see the numbers, says data analysis and management process automation expert Philipp Papayanov. It is important to understand how they will change over time and respond to them promptly. This year, Philipp Papayanov was included in the "Top-40 Digital Experts," a national annual award that identifies those who set the vector for digital development in Russia, thanks to the development of an analytical platform that helps property owners make optimal decisions and increase planning accuracy. The jury members called the combination of OLTP (Online Transaction Processing), financial modeling, and BI (Business Intelligence) elements in the platform a unique solution. The system has proven its effectiveness in TPS Real Estate, a commercial real estate operator that is among the Top 20 in Russia, according to Forbes.
In the interview, Philipp Papayanov explained why classical financial modeling is not suitable for managing shopping centers, when it makes sense to create a hybrid analytical architecture, and why it is premature to entrust the work of an analyst to artificial intelligence.
Philipp, creating an accurate and technological lease management system requires a deep understanding of business processes and market dynamics. The "Top-40 Digital Experts" award recognized your development as one of the best projects for decision-making in the real estate sector. What was the starting point for launching the project?
In fact, it all started with the need to systematize management decisions in conditions of uncertainty. Today, it is especially important to form the most thoughtful lease terms based on data. We see how consumer behavior is changing, how traffic in shopping centers is transforming, and this requires managers to be more accurate and responsive in their decision-making. After all, in shopping and entertainment centers, rent is tied not only to a fixed rate, but also to the tenant's turnover. In some cases, the terms may also take into account additional parameters, such as the percentage of vacant space in the shopping center. But if we talk about our system, its history began long before the current market changes. The idea of creating a comprehensive analytical platform came to me back in 2014, and since 2016 it has been used in the work. Over the past nine years, the system has developed significantly, covered all key business processes, and has become the basis for management decision-making. Therefore, we can say that at the start it was a response to internal needs and a desire for consistency. And today, it is already a logical continuation of the course towards digital transformation and strategic business development.
What is the difference between forecasting for tenants and classical financial modeling?
Classical modeling of the income of an existing business is usually based on the processing of actual income. If we earned X this year, then next year we will earn k multiplied by X, and modeling is reduced to selecting and justifying the coefficient k. Of course, this is a simplification; in reality, everything is much more complicated: there can be many factors, the business can scale, but the basis is still the fact. This approach is not applicable to shopping and entertainment centers, since the terms of contracts are often complex and may provide, for example, for a double increase in the rate after the first year of operation. Therefore, the basis for modeling should not be the fact, but the commercial terms.
Your system allowed the company to abandon disparate solutions and reduce the time to prepare calculations, and it has a lot of original content. For example, turnover forecasting algorithms. What data is most "sensitive" to you in the forecast, and what is important to consider in order not to make a mistake?
To make an accurate forecast, the task must be divided into two areas: for existing tenants and for planned tenants. The algorithms in these cases differ significantly. If a tenant has already worked in a shopping and entertainment center for at least a few months, you can build a fairly accurate forecast based on their actual turnover. At the same time, the key condition is the availability of reliable information: it is important to take into account not only the turnover itself, without its distortion by the tenant, but also the date of opening, possible interruptions in commercial activities, as well as the tenant's product profile, which is critical for accounting for seasonality. Thus, the most sensitive factor here is the accuracy of the actual data. A different approach is used for planned tenants; forecasting is based on the selection of comparable analogues. This is a more complex and creative process that requires industry awareness, and it cannot be fully automated.
The accuracy of turnover planning for 2024 was 98.5%. As noted by the jury, this is a very high figure for such a complex and variable area as commercial real estate. Do you think this is the limit, or can the accuracy be further improved?
This is indeed an excellent result, but forecast accuracy is not an end in itself, but a management tool. The figure of 98.5% is the average accuracy for the entire shopping center portfolio, and it confirms that we are moving in the right direction. At the same time, the accuracy naturally varied at the level of individual objects. That is why our immediate task is to minimize this spread and increase the stability of the forecast at the level of each specific shopping center. We have a clear understanding of the factors that can do this, and we see opportunities for further improving accuracy.
Your system combines elements of OLTP, financial modeling, and BI, which is a rather unusual approach that allows you to cover the entire cycle of working with a tenant in one platform. But, as a rule, such integrations require significant resources. In your opinion, when is it really justified to combine these components into a single system?
Yes, this combination really requires significant effort both at the development stage and during further system maintenance. But for managing commercial real estate facilities, especially on the scale of shopping and entertainment centers, this combination is not just justified, it is necessary. Shopping centers have a very high complexity of contractual relations, and each contract may have its own logic for changing rates: with reference to the tenant's turnover, with phased growth, taking into account seasonality or vacancy levels. It is impossible to model this outside the OLTP environment; you need to work with current, time-varying data. At the same time, it is necessary to take into account the financial consequences of these conditions, that is, to build models, not just record facts. BI in this case becomes a tool for analysis and decision-making: it allows you to see not only what is happening, but also why it is happening. This approach makes sense when business processes are tied to many variables and require constant recalculation and adaptation. If we are talking about a stable, linearly predictable business where conditions hardly change, it may be more effective to use a simpler and more specialized architecture. But in the case of shopping centers, disparate solutions simply will not allow you to manage the lease accurately and efficiently.
You are a member of the International Association of IT Professionals GrowCluster and ACM, a professional association of specialists in the field of computing and information technology, participation in which involves constant monitoring of new trends. Artificial intelligence and machine learning are now increasingly being integrated into analytical systems. How do you feel about connecting AI to BI systems?
In some aspects, artificial intelligence can indeed be very useful. The first thing that comes to mind is automatically generated prompts, recommendations, and conclusions directly in the interface of BI systems. The ability to formulate queries in natural language is also of great importance; this eliminates the need for users to even minimally configure filters manually. Integration with chatbots, when statistics, graphs, and analytical conclusions can be obtained directly in messengers, looks promising. Plus, classification, clustering, identification of anomalies, and potential errors in the data. But I think it is premature to entrust AI with the tasks of calculation and forecasting. The task of an analyst is not just to produce a result, but to explain it, justify the conclusions, and propose a management solution. Today, there is no AI that is capable of doing this at the level of an experienced analyst. Even if such an AI appears, its calculations will remain a "black box," and this will become a serious problem, especially if a factor analysis or a detailed hypothesis check is needed in the future.