Byte-sized Brilliance: AI in Location Planning Consultancy

AI in Location Planning Consultancy

AI has been a key topic of conversation across most industries, and location planning is no exception! In this blog we’ve explored the role of AI can play in Location Planning, and how this is combined with our GMAP consultancy to ensure we still provide our clients with expert insights for their decision making!

Developing the Scene: What is Ai?

Machine learning (ML) models and artificial intelligence (AI) are branches of statistical modelling, whereby models 'learn' patterns within datasets to generalize on unseen data. ML algorithms first emerged in the mid to late 20th century, and they have evolved into incredibly popular and successful predictive modelling algorithms such as Random Forests and Extreme Gradient Boosting machines. AI techniques were initially simple 'perceptron' networks which achieved limited success in early testing.

However, breakthroughs in the last 20 years have made AI techniques increasingly successful in complex tasks and pushed them into the limelight – the most well-known being large language models such as Chat-GPT. ML and AI play an increasingly dominant role in shaping our lives today – from informing supermarkets on what foods to stock and where; to the music recommended to you by Spotify for your daily commute.

At GMAP we are always excited to explore and implement innovative technologies and methods, and also to combine science, art and technology as we have done for the past thirty years in our location planning consultancy and geodemographic analysis. 

The Role of AI in Location Planning

ML and AI are also playing an increasingly significant role in location planning, as big datasets and processing power increase simultaneously. Unsupervised algorithms such as K-Means have become increasingly popular for customer segmentation, automatically grouping similar types of customers together. Supervised algorithms such as Random Forest are increasingly favoured in tasks such as forecasting sales, recommending products to specific customers,  and customer retention.

Spatial regression models, for instance, have been utilized in location planning for decades; however, they have shortcomings that can be addressed using ML methods. Spatial regression models typically assume linear relationships between predictive and target variables, which is rarely the reality. Whilst ML models represent nonlinear relationships, which are abundant in real scenarios, resulting in improved accuracy. When applied to location planning, these models have captured relationships in consumer behaviour and spending patterns previously unknown to clients. Furthermore, ML models offer the advantage of adaptability over time, provided they are re-trained with new data reflecting changing patterns and preferences, ensuring ongoing relevance and effectiveness in location planning strategies.

The Challenges of AI Adoption

ML and AI models are undoubtedly powerful, but they can also be impractical, requiring copious amounts of high-quality data, which can be challenging to access and clean sufficiently. Real-world datasets are noisy, containing missing or erroneous values, and are often smaller than ideal, creating challenges for using ML algorithms. Crucially, even when highly accurate outputs are generated, it may be unclear how they were generated – ML and AI models are black boxes.

While research is ongoing to create 'explainable ML' algorithms, their assistance in decision-making may be more limited, as it may be unknown which factors and relationships are most important for the client. Furthermore, there is the phenomenon known as data drift, whereby changing trends in reality are not represented by the model, which may not have been exposed to these trends at the time of training, resulting in vastly reduced accuracy over time. These behaviours are especially common in datasets based on human behaviour, such as retail spending. Consequently, ML models require frequent retraining with new data, which comes with the challenges previously described.

ML and AI systems also come with ethical and environmental concerns due to their data and resource requirements and reliance. Bias inherent in datasets have been known to be learned and reiterated by ML model outputs, potentially resulting in decisions that further exacerbate these biases – for instance, crime prediction models resulting in a higher police presence and arrest rates in certain areas. Due to the enormous amounts of data and processing power used by these systems, a significant and growing percentage of energy is directed towards powering data centres that support these models – in 2022, Google reported approximately 15% of its total energy use was for training ML models.

The Power of Consultancy in Location Planning

Consultancy at GMAP operates with a white-box approach – we make our data sourcing, processing, and modelling predictions clear to our clients through frequent touchpoints with our expert location planners. This transparency maximizes trust-building and improves conflict resolution both within and outside the team, resulting in more satisfactory project outcomes. Discussion plays a major role in our iterative process, from deciding on features to use in models, to explaining the reasons for our modelling results for actual decision-making. 

GMAP’s Location Planning Approach: Mixing Science, Art and Technology

At GMAP, we combine providing training and guidance for using our software and interrogating our modelling results, while keeping them sufficiently simple and explainable for clients to utilize independently. Our experience following this approach maximizes client satisfaction, evident in our strong client retention rates and numerous partnerships spanning decades. In contrast, AI systems are black boxes, sacrificing transparency, explainability, and accountability for moderate improvements in performance. In cases where an ML-generated output produces poor results, this can result in degraded trust.

Despite their differences, AI and ML have solid potential to complement our location planning consultancy in ways that add value to our clients, particularly in projects using very large geodemographic datasets or DVLA data. With the growth of open-source ML and AI solutions such as scikit-Learn and TensorFlow, we view tasks where maximizing accuracy is critical over explainability as having the greatest potential for integration. Furthermore, as explainable AI modelling strategies are further researched and developed, these could be effectively incorporated into our Ideal Network Plans (INPs) for more powerful insights, particularly as our location planning data assets continue to grow. 

Conclusion