AI Applications in New Product Forecasting

Machine Learning (ML) – ML systems learn patterns from data without explicit programming. Techniques like Random Forest, Support Vector Machines (SVM), and LSTM (Long Short-Term Memory networks) have outperformed classical models in NPF by identifying nonlinear patterns and capturing sparse or noisy datasets. ML also enables real-time sentiment analysis, using internet search and social media data to gauge consumer interest pre- and post-launch.

Fuzzy Logic and Neuro-Fuzzy Systems – Fuzzy logic handles uncertainty and imprecision by working with qualitative terms (e.g., high, medium, low) instead of fixed values. Adaptive neuro-fuzzy inference systems (ANFIS) merge ML and fuzzy logic to address forecasting challenges in volatile sectors like apparel.

Ensemble Learning – By combining predictions from multiple ML models, ensemble methods reduce individual model biases. Homogeneity-based clustering (e.g., using K-means) groups similar products by life cycle stages, enabling accurate segment-level forecasts. This reduces the need for product-specific attributes and aligns forecasting with business strategies.

Virtual Reality (VR) and Macro-Flow Models – VR simulations create immersive shopping experiences to collect behavioural data before launch. Macro-flow models track consumers’ movement through purchase journeys (awareness → interest → decision), providing early insights into market response. Studies show VR-enhanced models can predict first-year sales with errors as low as 1.9%.

Agent-Based Modelling (ABM) – ABM simulates individual consumer behaviours in a virtual marketplace. When combined with AI, it helps firms anticipate peer influence, tipping points, and regional adoption trends. These models are particularly effective in capturing network effects and heterogeneity in adoption patterns.

Dynamic Time Warping (DTW) – DTW enables comparison of sales trends between dissimilar time series, helping forecast new product sales by aligning them with historical clusters. This allows for more accurate forecasting by analogy even when product cycles are misaligned.

Hybrid Models – Hybrid models balance interpretability and performance. For example, Bayesian updating, when enhanced by AI, allows forecast models like the Bass diffusion model to refine parameters over time as sales data accumulates. This combination improves accuracy while preserving explanatory power.