The Future of Commercial Real Estate Will Belong to Better Predictions, Not Better Questions

Imagine a leasing team discussing a vacant retail unit in a successful shopping mall.

The space is attractive, its location within the property is strong, visitor traffic is healthy and several recognised brands have already expressed interest. From a traditional leasing perspective, this is the kind of vacancy most shopping centres would be pleased to have. There is demand, there are alternatives and there is an opportunity to negotiate favourable commercial terms.

The discussion eventually arrives at what appears to be a straightforward question: which brand should occupy the space?

For decades, commercial real estate professionals have answered questions like this by combining market knowledge, historical performance, leasing strategy and experience. They compare rents, evaluate the strength of each brand, consider the existing tenant mix and negotiate the best possible agreement.

But the real decision is considerably more complicated than choosing the tenant willing to pay the highest rent.

A new brand changes the environment around it. It can influence visitor circulation within the property, affect the performance of neighbouring stores, strengthen or weaken a particular retail category and alter the overall attractiveness of the tenant mix. Its turnover determines the economics of turnover-based rent, while its position inside the mall may influence how effectively it contributes to the property as a whole.

Two apparently similar leasing alternatives can therefore produce very different outcomes over the lifetime of a contract.

This is not simply a leasing decision.

It is an economic decision about the future performance of the asset.

And that distinction matters when we begin talking about artificial intelligence.

AI Has Become Very Good at Answering Questions

The recent development of generative AI has fundamentally changed the way people interact with technology. Instead of navigating complicated menus or searching through reports, users can increasingly ask a question in natural language and receive a useful answer within seconds.

That is a significant achievement.

A manager can ask for a summary of a monthly report. A leasing team can extract information from contracts. A technical manager can search maintenance records conversationally. An executive can request an explanation of unusual performance indicators without manually comparing multiple spreadsheets.

All of these applications make information easier to access.

But accessibility and foresight are not the same thing.

Most conversational AI begins working after someone decides what to ask. A person notices something, forms a question and requests an explanation. The system then searches the information available to it and produces an answer.

In other words, the interaction begins with human awareness.

That model works extremely well when the objective is to understand information. It becomes much less powerful when the objective is to anticipate what has not happened yet.

Consider a simple example. If a shopping mall manager asks why footfall declined last month, an AI system may analyse historical information and provide a very convincing explanation. It might identify weather conditions, calendar effects, competing events, marketing activity or changes in visitor behaviour.

The explanation may be completely correct.

But the month is already over.

Visitors have already decided where to spend their time. Retailers have already recorded their turnover. Variable rents have already been affected. Marketing budgets have already been spent.

Understanding what happened is useful.

Knowing what is likely to happen before it happens is something entirely different.

For commercial real estate, we believe that difference will become increasingly important.

The Most Valuable Questions Are About What Happens Next

Commercial real estate has spent decades becoming better at recording its past.

We know how many visitors entered a shopping mall yesterday. We know how much tenants sold last month. We know what rental income was generated, how much energy was consumed, which work orders were completed and how marketing campaigns performed.

Modern systems can collect these records with extraordinary precision.

Yet very few of the decisions that determine the future value of a property are actually about yesterday.

A leasing director deciding between two brands wants to understand which alternative is more likely to strengthen the tenant mix.

A shopping mall manager planning the next quarter wants to know what visitor traffic is likely to look like before allocating marketing and operational resources.

An asset manager wants to understand how tenant turnover may develop before it begins affecting rental income.

An investor is ultimately interested in what the combination of these variables means for the future financial performance of the asset.

These are fundamentally different questions because they cannot be answered simply by retrieving information.

They require prediction.

And prediction is rarely about a single number.

Every Prediction Exists Inside a System

Footfall is a good example.

Predicting how many people will visit a shopping mall may appear to be a relatively isolated forecasting problem. From an engineering perspective, however, visitor traffic is influenced by a combination of factors that interact continuously: historical patterns, seasonality, weekdays and weekends, public holidays, weather, campaigns, events, the characteristics of the city, the positioning of the property and many other variables.

Even an accurate footfall prediction, however, is only the beginning.

Visitors do not create financial performance simply by entering a building.

They move through it.

They encounter different brands.

They respond to the tenant mix.

They spend differently depending on category, location and occasion.

A store positioned beside complementary brands may perform differently from the same store positioned elsewhere in the same property. Two retailers belonging to the same category may react differently to identical visitor volumes. The performance of a brand can be influenced by its neighbours, its floor, its visibility, the characteristics of the city and the positioning of the shopping mall itself.

This means that predicting tenant turnover requires more than knowing how many visitors will arrive.

It requires understanding the commercial ecosystem those visitors will enter.

Turnover then becomes part of another relationship.

Many commercial leases combine fixed rent, turnover-based components, foreign exchange parameters, indexation rules, thresholds, dates and other contractual conditions. Once these parameters are connected with expected tenant performance, forecasting begins to move beyond retail performance and towards the economics of the property itself.

The sequence becomes important.

Footfall influences commercial opportunity.

Brand mix and location influence how that opportunity is distributed.

Those relationships influence tenant turnover.

Turnover interacts with lease conditions.

Lease conditions ultimately influence rental income.

Each prediction creates additional context for the next one.

This is why, from an engineering perspective, we do not see predictive intelligence as a collection of independent forecasting models. The more interesting challenge is understanding how individual predictions connect to one another and how those relationships can support a real commercial decision.

A Vacant Unit Is Not an Empty Rectangle

Return to the vacant retail unit we started with.

Traditional analysis can provide a considerable amount of useful information about it. We can examine its size, previous tenants, historical rent, location, nearby stores and past performance. We can compare offers from prospective tenants and calculate the expected contractual income from each alternative.

But imagine being able to evaluate a different set of questions before signing the lease.

What type of retail activity would complement the existing brand mix?

How might each candidate perform in that specific unit rather than across the mall in general?

Does the performance history of neighbouring stores suggest that one category creates stronger commercial relationships than another?

Could placing another fashion retailer in the area strengthen the cluster, or has the category already reached a point where additional supply simply redistributes existing turnover?

Would a different sector create a healthier balance and improve the performance of the surrounding ecosystem?

And what could each scenario mean for rental income over time?

Suddenly, the objective is no longer simply filling a vacancy.

It is understanding the likely consequences of filling it in different ways.

That is a very different engineering problem.

It requires historical information, but it cannot stop at historical reporting. It requires contractual information, but reading the contract is not enough. It requires understanding brands, yet brand performance alone does not explain the ecosystem. It requires location information, but location only becomes meaningful when combined with visitor behaviour and the performance of surrounding tenants.

The value lies in the relationships.

This is also why we believe the next meaningful step for AI in commercial real estate will not come from making it easier to ask questions about existing information.

It will come from recognising important relationships early enough that some questions no longer need to be asked in the first place.

Why Should the System Wait for the Question?

There is an important assumption hidden inside most business intelligence systems.

The user knows what they are looking for.

Dashboards assume that someone knows which KPI deserves attention. Reports assume that someone knows which page to read. Conversational AI improves the interface dramatically, but it often preserves the same underlying assumption: someone still needs to recognise the issue and ask the right question.

For an engineering team, that creates an interesting challenge.

What if the system could recognise a meaningful change before the user went looking for it?

Instead of asking why visitor traffic changed, what if the organisation already knew that a change was likely?

Instead of discovering a tenant performance problem after monthly turnover declarations arrived, what if an unusual pattern could be identified earlier?

Instead of asking which sector historically performed well in a vacant unit, what if different leasing scenarios could already be evaluated against expected turnover and rental outcomes?

The distinction may sound small, but it changes the role of technology.

The system is no longer waiting to explain the operation.

It is beginning to anticipate it.

And this leads to a question we believe will become increasingly important as artificial intelligence matures inside commercial real estate:

Why should we always have to ask the question if the system can identify what we are likely to need to know?

For us, that is where predictive intelligence begins.

Predictive intelligence becomes useful when it changes the timing of a decision.

There is an enormous difference between being told that something happened, being warned that something is likely to happen and understanding what can still be done about it. Traditional reporting is excellent at the first. Forecasting begins to address the second. The real opportunity, however, lies in connecting prediction with the decisions that remain available.

This is where accuracy matters, but accuracy alone is not enough.

A model can produce an impressive forecast and still have very little impact on the way a property is managed. If its output remains isolated from the commercial and operational context in which decisions are made, it simply becomes another number on another dashboard.

The engineering challenge is therefore not only to predict an outcome. It is to make that prediction meaningful within the wider economics of the asset.

Prediction Is Valuable Only When It Creates Confidence

Consider footfall again.

Knowing that visitor traffic is likely to increase next month has some value. Knowing why it is expected to increase, where that increase is likely to occur and how it may influence different parts of the property has considerably more.

The same principle applies to tenant turnover.

A forecast suggesting that a retailer will grow by 7% may look encouraging, but a leasing director needs more context before acting on it. Is the growth being driven by stronger visitor traffic across the entire property? Is the brand outperforming its category? Are neighbouring stores experiencing similar improvements? Is the location itself contributing to performance? Would the same brand be expected to perform differently if it occupied another unit?

Prediction without context produces a number.

Prediction with context begins to create confidence.

That distinction matters because commercial real estate decisions are rarely reversible without cost. Lease agreements may define relationships for years. Refurbishments require capital. Changing the tenant mix can influence the positioning of an asset long after the original decision has been made. Even relatively small operational choices can have consequences that only become visible months later.

Decision-makers therefore do not need artificial intelligence to sound certain.

They need enough evidence to understand uncertainty.

From an engineering perspective, this is a much more interesting objective than simply producing the highest possible forecast accuracy. A useful predictive system should help users understand which variables are influencing an expected outcome, how strongly the system supports that expectation and what alternative scenarios may look like.

The goal is not to remove judgement from the decision.

It is to give judgement a better view of what may happen next.

The Mall Is an Ecosystem, Not a Collection of Stores

This becomes especially important when we think about brand mix.

A shopping mall can be represented in a database as a collection of units, tenants, categories, contracts and financial records. That structure is necessary for managing the property, but it does not fully describe how the property behaves.

In reality, stores influence one another.

A strong anchor may draw visitors into an area that benefits surrounding tenants. A successful food and beverage cluster can extend dwell time and change visitor circulation. Two apparently complementary retailers may strengthen one another, while another combination may compete for the same spending without creating additional value for the property.

Even physical distance can have commercial meaning.

A brand on the ground floor beside a high-performing fashion cluster exists in a different commercial environment from the same brand placed beside a service corridor on an upper floor. The name on the storefront may be identical. The likely outcome is not.

City matters as well.

A brand performing exceptionally well in Istanbul may behave differently in Antalya, Skopje or another market. Local purchasing power, tourism, climate, demographics, competitive supply and shopping habits all influence performance. The same applies to the shopping mall itself: its positioning within the city, catchment area, accessibility and existing brand mix create a context that cannot be understood from the retailer's identity alone.

This is why we are cautious about describing commercial real estate AI as simply “learning from more data.”

More data is not automatically better intelligence.

The relationships within that data are what matter.

If the objective is to predict turnover, the system needs more than historical turnover. If the objective is to recommend a sector for a vacant unit, it needs more than a list of successful tenants. If the objective is eventually to forecast rental income, it needs to understand how expected commercial performance interacts with the actual economic structure of lease agreements.

Every layer changes the meaning of the layer before it.

That is what makes the problem difficult.

It is also what makes it valuable.

From Forecasting Performance to Modelling Economics

For investors and asset managers, the prediction chain ultimately reaches a financial question.

What is this property likely to earn?

Rental income in a shopping mall is not generated by one variable. It emerges from hundreds of commercial relationships governed by different contractual structures. Fixed rents, turnover-based rents, thresholds, indexation mechanisms, currencies, dates and other lease parameters interact with tenant performance over time.

When these contractual parameters exist as structured information rather than documents sitting in separate archives, another possibility emerges.

Expected tenant performance can begin to interact with expected lease economics.

A turnover forecast is no longer only useful to the retailer or leasing team. It becomes an input into rental-income expectations. Changes in visitor behaviour can become relevant to financial planning. A brand-mix decision can be evaluated not only in terms of occupancy but also in terms of its possible contribution to the future income profile of the asset.

This is the point where prediction begins to move beyond operational optimisation.

It starts becoming financial intelligence.

And that is where the distinction between reporting and predictive intelligence becomes much more consequential.

A traditional financial report can tell an investor exactly how much rental income a property generated last quarter.

A predictive system should eventually help answer a harder question:

Given what we know today, what is this asset likely to generate next?

That is not a question we believe should be answered by a single model or a single dataset.

It is the result of a chain.

Expected footfall informs expected commercial opportunity. Commercial context influences expected tenant performance. Tenant performance interacts with lease conditions. Lease economics shape rental income. And those financial outcomes contribute to the decisions made around the asset.

Each layer carries uncertainty.

Each layer also adds information.

Engineering the connections between them is where predictive intelligence becomes considerably more interesting than producing isolated forecasts.

Better Decisions Do Not Require Perfect Predictions

There is an uncomfortable truth about prediction that is sometimes lost in conversations about artificial intelligence.

The future cannot be known perfectly.

A new competitor can enter the market. Weather can behave unexpectedly. Consumer confidence can change. A tenant can launch a successful campaign or make a poor commercial decision. Economic conditions can shift faster than historical patterns suggest.

A system that promises certainty in that environment should probably be treated with caution.

But commercial real estate has never required certainty in order to make decisions.

Investors model scenarios. Leasing teams compare alternatives. Asset managers work with budgets and forecasts. Management teams make decisions every day using incomplete information because waiting for certainty would often mean waiting until the opportunity has disappeared.

Predictive intelligence should not pretend to eliminate that uncertainty.

Its role is to make uncertainty more manageable.

If one leasing scenario appears materially stronger than another across several relevant variables, that information matters even if the future cannot be guaranteed. If a turnover pattern begins moving outside its expected range, recognising that change early can be useful before anyone knows exactly why it is happening. If rental income is developing differently from the current budget, identifying the likely direction early gives management more time to respond.

The objective is therefore not perfect foresight.

It is better-informed action.

That distinction is important to us because it also defines how we think AI should interact with experienced professionals. A prediction should not become a decision simply because a model produced it. Commercial context, negotiation, strategy and human judgement remain essential.

AI can expand the field of view.

People still decide where to go.

The Next Interface May Not Be a Chat Window

One of the most visible changes introduced by generative AI is the chat interface.

It feels natural because people already understand conversation. Ask a question, receive an answer. Refine the question, receive a better answer.

We expect conversational interfaces to remain useful in commercial real estate as well. There will always be situations where a user wants to explore information, investigate an anomaly or understand why something happened.

But we do not believe chat should be the final interface for operational AI.

If a system is capable of identifying patterns, forecasting outcomes and recognising meaningful deviations, forcing the user to discover everything through questions creates an unnecessary limitation.

The more interesting interface may begin with the system itself.

A manager opens the platform in the morning and does not see another collection of charts demanding interpretation. Instead, the system identifies a small number of developments that deserve attention.

Visitor traffic for an upcoming period is moving outside the expected range.

A tenant's projected turnover is beginning to diverge from its recent performance.

A vacancy appears better suited to a different retail category when neighbouring performance and brand mix are considered together.

Expected rental income is moving away from the current financial plan.

The examples will differ by property, and the quality of each prediction will depend on the information available. The principle, however, remains the same.

The user should not always have to know which question to ask.

Technology should become capable of recognising when there is something worth asking about.

From Predictive Intelligence to PredictAI

This engineering philosophy is what led us to PredictAI.

PredictAI is not intended to be a chatbot attached to a property-management platform. Its purpose is to develop a predictive intelligence layer on top of connected commercial real estate information.

We began with footfall because visitor traffic is one of the fundamental signals within a shopping mall. Understanding likely visitor volumes creates a foundation for analysing what may happen commercially inside the property.

The next layer is tenant turnover.

Here, the problem becomes considerably richer. Expected visitor traffic is only one input. Brand, category, unit position, neighbouring tenants, historical performance, city characteristics and the positioning of the shopping mall itself can all contribute to the expected outcome.

From there, our objective extends towards rental income.

Because lease parameters are already part of the operational data environment, expected tenant performance can eventually be evaluated together with the contractual structures that convert commercial performance into property income.

And beyond forecasting individual metrics lies a more ambitious objective: helping management teams evaluate possible decisions before those decisions become historical data.

Which sector is likely to create the strongest outcome for a vacant unit?

How could changing part of the brand mix affect the wider property?

Which tenants appear likely to outperform or underperform expectations?

How might those changes influence rental income?

These are the questions that shape the direction of PredictAI.

Not because we believe an algorithm should make leasing or investment decisions on behalf of professionals.

Because we believe professionals should have a clearer view of the likely consequences before making them.

The Future Will Arrive Before the Question

Commercial real estate has built an extraordinary ability to document what has already happened.

That capability will remain essential. Reliable records, reporting and operational data are the foundation on which everything else depends.

But the next stage will be judged by a different standard.

Not only by how accurately technology can describe yesterday, but by how effectively it can help organisations prepare for tomorrow.

For decades, the industry's systems have waited patiently for people to open reports, interpret dashboards, identify problems and decide which questions to ask.

Artificial intelligence gives us the opportunity to rethink that relationship.

The future of commercial real estate will not belong to the organisations that simply ask better questions of their data.

It will belong to those that recognise the right signals early enough to make better decisions.

Sometimes, before anyone needs to ask at all.

EPPSO Perspective

At EPPSO, we believe the value of artificial intelligence in commercial real estate will ultimately be measured by its impact on decisions rather than the sophistication of its answers.

Connected operational data provides the foundation. Experience provides context. Prediction provides a view of what may come next.

The purpose of PredictAI is to bring those elements together so that shopping mall managers, leasing teams, asset managers and investors can evaluate the future with more information than experience alone could ever process.

We are not trying to remove uncertainty from commercial real estate.

We are trying to see more of it before it becomes history.