Retailers make hundreds of thousands of inventory decisions every day. Where do I fulfill that online order from? Where do I place the inventory I bought? What's trending and how does that impact my merchandise strategy? What and how much should I buy? The list goes on. Given all the complex questions retailers must answer across the inventory lifecycle, the opportunity to make smarter decisions using data and science is paramount.
In the past, science was used with limited success. Traditional forecasting methods failed to address many of the data challenges retailers face today, primarily due to the nature of their data—which is often volatile, sparse and noisy. Additionally, new pressures to meet new consumer expectations in today's complex retail environment (where customers can buy, receive and return anywhere) make it difficult to accurately forecast at the granular level necessary to improve inventory decisions. In the end, outdated forecasting methods paired with rules-based decision-making have led to suboptimal business results time and time again.
Artificial intelligence and machine learning are poised to revolutionize inventory decision-making, especially when the next generation of demand prediction is combined with the next generation of optimization technology. The application of artificial intelligence unearths insights and trends from multiple internal and external data sets, and identifies opportunities missed by business users or traditional forecasts.
GET BEHIND THE AI HYPE
Artificial intelligence (AI) is the simulation of human cognitive functions by machines, such as the ability to perceive, reason, learn, and problem solve. AI technology is part of a broader concept of machines with an ability to carry out tasks in a way people would consider “smart.”
Machine learning (ML), a subset of AI, is based around the idea of giving machines access to data and allowing them learn for themselves. Rather than teaching the machine to perform an action, step-by-step, with ML, machines can learn to work by observing, classifying, and learning from their mistakes—just as humans do.
Deep learning (also known as deep structured learning) is part of the broader family of ML methods based on artificial neural networks. While traditional programs build analysis with data in a linear way, the hierarchical function of deep learning systems enables machines to process data with a nonlinear approach.
Celect takes a two-pronged approach to address inventory decisions: By combining a patented context-aware AI-based prediction with opportunity-aware optimization, Celect technology can consider both leading indicators from structured and unstructured data to address retail's most pressing inventory challenges.
Our science is all made available via the Celect Inventory Optimization Platform which is a repeatable, cloud-based offering that delivers the prediction and optimization building blocks needed to support the Celect Inventory Optimization Suite (Plan & Buy Optimization, Allocation Optimization and Fulfillment Optimization), as well as the ability to incorporate these capabilities into other SaaS offerings.
How would knowledge of a future outcome impact your decisions today? Celect uses a combination of standard machine learning, standard Deep Learning and Celect patented approaches to solve the retail prediction challenge at hand. In fact, Celect customers see a 15% to 79% prediction accuracy improvements. There are four key elements that make Celect prediction unique:
1. Models - Celect uses AI-based models (industry standard ML, DL, or Celect patented models) to power prediction. These models are flexible enough to accept new data sets, identify relevant data sets and features, and uncover interdependencies between various data sets.
2. Context - Celect uses patented algorithms to identify context, then predict probable outcomes based on such context. For example, Celect uses patented choice modeling algorithms to provide insight into not just what product was sold, but how a product was impacted by the assortment around it.
3. Multi-technique Temporal Prediction - Blended frequency, causal and trending techniques address unique retail data challenges. A combination of spectral domain and time series approaches are used to identify and predict trends.
4. New Scale -Sparse data has always been an issue when predicting demand in retail. To predict demand accurately, retailers need to know the probability of a customer buying a particular product at a particular time and touchpoint. Celect uses patented tensor completion algorithms to make sense of and predict really sparse data.
Prediction provides the context of a likely future outcome, but needs to be combined with optimization to account for realistic business objectives and constraints. Celect’s approach to optimization is unique because it considers four key areas: Opportunity Cost, Multiple Objectives at Scale, Assortment Context, and Robust Optimization.
Consider the tradeoffs for each relevant combination (i.e., customer expectation, constraints, costs, demand prediction) in sub-seconds.
We make the optimization more accurate by considering the risk or uncertainty associated with the predicted demand.
Consider the assortment implications - instead of optimizing for one product at a time. For example: this allows us to, again, investigate the future and tradeoff choosing to carry a product in a store versus sending more of that product to another store and avoiding an out-of-stock.
MULTIPLE OBJECTIVES AT SCALE
Make the best possible decision across multiple competing objectives at the scale of retail. For example: this allows us to investigate the future and tradeoff taking a future markdown versus fulfilling an online order from a store that might have a slightly higher transportation cost.