Article

Navigating the Future of Demand Forecasting: Balancing AI/ML and Traditional Statistical Methods for UK Retailers

Suddha Ray | Senior Consultant | Retail Reply, London, UK
Version 1.1 | March 2025

Introduction

Demand forecasting has always been a critical component of retail strategy. Accurate forecasts allow retailers to optimise inventory, reduce wastage, improve customer satisfaction, and ultimately protect margins. Historically, traditional statistical models like Bayesian inference, exponential smoothing, and time-series analysis have been trusted and stood the test of time. However, the rise of Artificial Intelligence (AI) and Machine Learning (ML) presents new and interesting opportunities for a more dynamic and data-driven forecasting approach.

For UK based retailers, particularly in a volatile market shaped by Brexit, inflation, political turmoil and shifting consumer behaviour, the question is not just about which approach to use, but how to best integrate emerging AI/ML solutions with established practices. Crucially, retailers must also recognise that AI is not a panacea; the foundation of good forecasting starts with a deep understanding of their own data. Without this, even advanced AI models may fail to deliver the expected results.

A screenshot of a computer

Description automatically generated

Figure 1: Comparison of Demand Forecasting Methods (Green boxes indicate strengths, red boxes indicate downsides)
 

Traditional Forecasting Approaches: Strengths and Limitations

1. Established Reliability

Statistical methods like exponential smoothing or Bayesian models rely on historical data patterns to predict future demand. These methods are transparent and interpretable, ideal for stable, consistent patterns in demand. They are also easier to maintain, requiring less computational power than AI models.

2. Limitations

  • Lack of Adaptability: These methods often struggle to capture abrupt changes in consumer behaviour or external disruptions like market shifts, supply chain disruptions, or pandemics.
  • Limited Data Inputs: Traditional models typically factor in fewer variables, ignoring the potential insights from external datasets such as social media trends, weather patterns, or macroeconomic indicators.
  • High Maintenance: Bayesian models, in particular, can be labour-intensive in data collection and parameter tuning, especially for larger retailers with diverse product ranges.

A screen shot of a chart

Description automatically generated

Figure 2: Strengths and Limitations of Traditional Forecasting Methods

Sources: 

AI and ML for Demand Forecasting: Opportunities and Challenges 

  1. Data-Driven Flexibility: AI/ML models, especially those using deep learning or reinforcement learning, allow for:

    Real-Time Forecasting: These models continuously learn from new data inputs, adjusting predictions dynamically. At the same time, they are capable of complex Input handlingIn other wordsAI/ML can typically process vast amounts of data, including real-time sales data, consumer sentiment from social media, economic indicators, and even weather patterns and geopolitical situations

    Pattern Recognition: Machine learning models can detect patterns in data that traditional methods might overlook, helping to optimise inventory and reduce stockouts. 
     
  2. Challenges for UK Retailers:

    Data Quality and Understanding: While AI/ML can process large datasets, the success of these models depends heavily on how well retailers know their own data. Retailers must have a strong grasp of their own demand patterns, seasonality, and category-specific trends. AI/ML is not a remedy for lack of data governance practices or not having cognisance of past trends and their anomalies down to the subcategory level, if not at the SKU level. An AI model trained to forecast future demand, however advanced, cannot compensate for ignorance of key demand factors. If a retailer does not understand the historical trends and root causes of mis-forecasting, the AI model may only end up amplifying these errors and make things worse.

    High Initial Investment: AI/ML models require significant investment in infrastructure, talent, and time. For many retailers, the ROI is only visible after significant investment.

    Model Transparency: AI models can be "black boxes," making it difficult for decision-makers to understand how predictions are derived, unlike traditional statistical models which are more interpretable. 
     

Informed Decision-Making: A Hybrid Approach

For UK retailers to make an informed decision on AI/ML implementation, they need to balance innovation with proven techniques. Here’s a framework for how they can approach this:

  1. Evaluate the Complexity of Demand Patterns:
    Retailers should assess the nature of their demand. If their demand is stable and predictable, statistical methods may still suffice, and AI models can help smooth the minor noise from the overall forecast. However, for more volatile product categories like fashion or technology, AI/ML may offer greater accuracy, or at least a starting point for further analysis. AI/ML models offer significant advantages for volatile product categories like fashion and technology. Fashion and technology products often have shorter lifecycles with rapid obsolescence. The model’s ability to incorporate real-time data, recognise complex patterns, and continuously adapt to new trends makes them invaluable in predicting demand for products where rapid changes in consumer preferences (i.e. declining interest) or market conditions are to be expected. Traditional statistical methods, which rely heavily on historical data, struggle with such short-lived patterns. For instance, in fashion, AI/ML models can analyse social media sentiment or influencer campaigns to anticipate demand for specific colours or styles. Similarly, in technology, they can account for product announcements, economic conditions, and even geopolitical events impacting global supply chains. This is particularly useful if historical demand forecasting has mostly been missing the mark more often than not leading to out of stock scenarios or overstocking. AI/ML offers a powerful advantage in managing high-volatility product categories, not just because it reacts to known trends, but because it can discern complex social patterns within a broader range of data.  Traditional methods often struggle with unforeseen events like the COVID-19 pandemic, which disrupt established trends.  While effective training is crucial, the real strength of AI/ML lies in its ability to analyse diverse datasets – encompassing historical sales, market trends, social media sentiment, weather patterns, and even supply chain data – to identify subtle correlations and anomalies.  This capability allows it to anticipate shifts in demand and adapt to volatile market conditions, even when the underlying causes are not immediately apparent.  By uncovering hidden patterns, AI/ML can provide insights that might be missed by human analysts, enabling more proactive and responsive decision-making in the face of uncertainty. 
     
  2. Data Readiness and Familiarity:
    A deep understanding of existing data is critical before embarking on AI/ML adoption. Retailers must ensure they are collecting high-quality, clean data and are aware of key patterns and trends in their categories. Retailers who are unsure of their own demand drivers or trends will struggle to see benefits from AI, as it won’t be able to mitigate underlying data issues. Knowing the data ensures that AI models can be fine-tuned to spot and improve real-time deviations.
     
  3. Pilot AI/ML Solutions for High-Impact Products: 
    Piloting AI/ML forecasting with select high-margin, high-volatility products allows retailers to evaluate a specific model's effectiveness within their own context. Since purchased AI/ML solutions don't always guarantee success, this real-world testing is crucial to assess the model's accuracy, identify limitations, and ensure it aligns with the retailer's data and needs before wider deployment.
     
  4. Cost-Benefit Analysis:
    Traditional methods are more cost-effective in the short term but may not provide the flexibility needed in a rapidly changing retail environment. Retailers should consider:
  • Short-term vs. long-term costs.
  • Potential savings from reduced overstocking or stockouts.
  • Improved customer satisfaction through better demand alignment. 

     5. Blending Models:
          A hybrid model combining AI/ML with traditional methods can be an ideal solution if implemented 
          carefully. For instance:

  • AI/ML can be used to predict trends based on external data (social media, weather, competition prices, geopolitical factors etc.).
  • Traditional statistical methods can then adjust for known, predictable seasonal trends or promotions.
     

The UK Market Context: Considerations for Retailers

  1. Post-Brexit Complexity:
    Brexit has increased the unpredictability of supply chains, labour shortages, and price fluctuations in the UK. AI/ML models, which can factor in real-time market and logistical data, offer a unique advantage in this uncertain environment.
     
  2. Changing Consumer Behaviour:
    UK consumers have shifted significantly toward online shopping post-pandemic. AI/ML can help UK retailers tap into real-time e-commerce trends and consumer sentiment analysis, offering a more granular understanding of demand compared to traditional approaches.
     
  3. Sustainability and ESG Goals:
    Retailers pursuing sustainability goals benefit from AI/ML's ability to reduce overproduction and waste via accurate demand forecasting, aligning with UK ESG compliance. While AI/ML models use energy, the resulting environmental benefits from reduced waste (less resource use, transportation, landfill) often outweigh this, especially with increasingly efficient AI and renewable energy use.

Figure 3: Factors impacting AI/ML adoption

Sources:

The above chart underscores the challenges facing AI/ML adoption in UK retail. Market volatility tops the list, highlighting the sector's dynamic nature and the inherent risks associated with investing in modern technologies. Data quality and concerns about return on investment (ROI) also pose significant barriers. Interestingly, talent availability and model transparency are perceived as less impactful relative to others, suggesting that while skilled professionals are important, their availability is not the primary hurdle.

Conclusion: Charting the Future

UK-based retailers must recognise that the future of demand forecasting lies not in choosing between AI/ML and traditional methods, but in understanding how to leverage both effectively. AI and ML bring the potential for nuanced, real-time forecasting and adaptability, while traditional methods offer reliability and interpretability. However, it’s equally important for retailers to deeply understand their own data before adopting any AI/ML model. AI is not a one-size-fits-all solution to data problems; its success hinges on the retailer’s ability to supply it with meaningful, well-understood data inputs. Ignorance of demand trends and patterns will undermine AI’s ability to deliver the promised results.

However, for AI/ML models to deliver meaningful results, retailers must provide accurate and well-prepared data inputs. AI is not a universal fix for data challenges; its effectiveness depends on the clarity and relevance of the information it processes. A lack of insight into demand patterns or inconsistencies in data can significantly limit the potential benefits of these technologies

An informed decision involves a clear understanding of the retailer's demand complexity, data readiness, and resources. As retailers face increasingly unpredictable market dynamics, a hybrid approach blending AI with statistical models may offer the best path forward, enabling both operational efficiency and agility in a competitive landscape.

Making the right decision requires retailers to assess their demand patterns, data quality, and available resources. In the face of volatile market conditions, adopting a hybrid approach that combines the strengths of AI with traditional statistical models can provide a balanced solution. This strategy enhances both operational efficiency and the ability to adapt quickly in a highly competitive environment

Final Thoughts

For UK retailers, the question is no longer whether to adopt AI/ML for demand forecasting, but how to approach it strategically. Striking the right balance between costs and the opportunity to innovate is critical for meeting evolving customer expectations. Equally important is laying a solid foundation by ensuring the data infrastructure is robust, organised, and actionable. Retailers who prepare thoughtfully will be better positioned to unlock the full potential of AI/ML technologies and drive meaningful outcomes from their forecasting efforts.