.jpg)
In recent discussions with industry leaders and as outlined in our recent thought leadership piece “Evolution of Real Estate Network Planning in the Age of AI”, one of the expected benefits from leveraging AI/ML technologies is that through pure automation, the gains in speed and accuracy of prediction machines reduce the risk making mistakes in the site selection process through more accurate predictions of market potential. In fact, there is an expectation that a shift towards automated AI/ML model updates allows GeoAI teams to stay current and future-proof predictions by constantly improving model testing and validation results to enhance prediction accuracy.
There is a lot to unpack here, so let's dive in!
The overwhelming perception from GIS and Real Estate Executives is that AI/ML technologies will have a profound effect on the speed of producing sales forecasts, the measurement and prediction of cannibalization, and the ability to find white and gray (infill, repurpose) space opportunities in retail networks. According to some practitioners, these things should be running in seconds or minutes, not overnight or for two days. AI/ML technologies should almost make it “real-time”. But speed is not where the true “risk abatement” value emanates from; it’s through automation! The faster speed to run complex model predictions is generated through AI/ML automation and the technological resources of a cloud-based workbench.
The perception is that automated AI/ML processes will help keep the forecasting system “fresh and relevant” in the markets they care about. This specifically means that when core databases get refreshed with new data, the models “automatically” or by schedule, train new parameters, estimates, test and cross-validate for out-of-sample forecast accuracy and load the “fresh models” - sometimes overnight!
“The automation part has real value because now our models can be updated overnight as opposed to weeks or months like third party systems. This allows both speed to decision making and keeps the models accurate and up to date.” (Dave Chipman, Head of Retail Strategy, The Aspen Group)
In addition, it is perceived by some executives we talked to that AI/ML systems should be incorporating new and unique data sources on a regular basis as they are discovered to see if they add any further accuracy to the AI system. For example, store or restaurant customer survey data? We test for things like store operational variables compared to real estate attributes like “box size”. It would be important to know if the real estate attributes are more or less influential than operational aspects like customer service/satisfaction or food quality, store staff efficiency, and managerial competence, to name a few. Or, has a new competitor entered the market that reduced store performance? The models need to take this into account on a regular basis.
It is perceived that AI/ML capabilities ensure that models evolve in real-time, capturing the nuances of ever-changing market dynamics embodied in real-time data updates. In addition, the models can be tested and refreshed in the presence of new sources of data that include new attributes or characteristics of the market not previously known.
But this is only a perception, and very few within the retail and CRE sectors are actually doing it.
The reality is that most good modeling takes time and even with AI/ML systems, there is even more scrutiny on the model development and maintenance process because of the “black box” syndrome. Even in AI High Performing companies, automating the “model refresh” process has been a goal and they are well on their way, but the original build still takes time to include review cycles with executives in order to build trust in the system from the beginning. But progress will be made and the traditional annual or quarterly model update is replaced by a dynamic, adaptive system that consistently and continuously refines itself, while guarding against anomalies and outlier predictions to reduce decision risk and foster executive trust.
To get to that next level of automation, GIS experts are looking for huge value in using Generative AI to generate code and identify model bugs and fixes. Essentially, the role of AI as a modeling assistant will be about empowering analysts with tools that go beyond automation, aiding in complex model vetting decision-making processes. The emphasis on standardization will create a collaborative environment where analysts can trust in the uniformity of methodology and accuracy of the output. AI becomes a collaborator, a partner in the analytical journey, ensuring that the human touch and AI synergy produce insights that are not just accurate but robust, reliable and explainable.
The PiinPoint Way
At PiinPoint, we invested heavily in our GeoAI capabilities and our cloud-based infrastructure that promote speed and accuracy through automation. Over the years, our model updating and refresh capabilities will continue to evolve based largely on the direct feedback from clients, to ensure that the tools that they use for decision-making are as current as possible. We look at things in two ways:
- First time Model Calibration: The first time model build is a slower more deliberate process - models are custom built to fit the client's market situation, store network, and competitor set. We work closely with clients to acquire and validate their data, explore and derive insights from the data to inform feature choice and then execute on a detailed Train-Test-Cross Validation process, all along, informing the client what we are coming up with and why! This process takes time and we trade that off to build trust in the system for the user, and accuracy and consistency of the model predictions.
- Model Update Protocols: PiinPoint updates models in two ways 1) a more automated “refresh” when predictions tend to drift off baseline, and 2) Model Recalibration where we test new variables from new data sources. The refresh process is deliberately scheduled and automation can be a huge benefit to reduce turnaround times. The model recalibration process, although not as long as the original model build, is still a process that requires client collaboration and good “human judgment” as to the value of the new models compared to current models before loading for production.
There are important distinctions. Over time, can we reduce the turnaround times to be near “real-time” and still maintain client trust and maintain stability and accuracy of the forecasts? The answer is likely yes, given the pace of AI development. We know that the integration of AI/Machine Learning to automate prediction and forecasting models will both streamline operations and inject a level of dynamism in the fiercely competitive arena of CRE. It's a strategic play where accuracy and relevancy become the armor against financial pitfalls.
Check out PiinPoint and our full suite of Enterprise offerings using our AI/ML capability. We exist to support our clients' need to increase their speed to market and close deals faster!
