The Problem Statement
One of the main principles of merchandising is to stock the right product at the right place at the right time. This is much easier said than done when dealing with “one-size” items but when size, colour and gender is added to the mix; it increases in complexity and difficulty in ensuring the right amount of stock is invested. Today, there is still a huge sizing problem in the fashion industry and a huge gap in the market for sizes. There is no universal sizing standard and it differs from retailer to retailer, product category and end-use. Therefore, it is imperative that we analyse and classify the size curve to understand and cater to it better.
There are two types of sizing classifications:
- Dimensional expressed as alphanumeric:
- Product dimensions
- Body dimensions – leg-length, head-girth, etc.
- Zonal – Core, Extended, Extreme
It is a known fact that having the right mix of sizes has a direct impact on customers’ loyalty and is crucial to retailers’ success. The inability to cater to the customer, in this ‘’buy now, wear now’’ culture, will not only result in lost sales, but potentially a lost customer too.
This poses some very important questions that retailers are asking themselves:
- How do we get the size curve right?
- Is the launch size curve the same as the steady state size curve?
- Should the size curve be customised based on the product life-cycle?
- How do I ensure that past sales and stock-outs are not impacting the size curve allocated to stores?
- How do we ensure the correct size curve being available to influence and maintain the sales as well as brand experience and sustain the brand loyalty? This basic question is complicated by the fact the a newly acquired customer costs multiple time the money of retaining an existing customer.
The Components of Size Curve Calibration
Traditionally, Size curve generation is done as a two-step process:
Step 1: Clustering
Clustering plays an important role by providing in-sight into the historical store, product and sales data. The current practice within retailers’ is to assign/develop a size curve based on the store groups or clusters they are a part of. Clusters are determined based on turnover, a grading and the size of the store; hence using a very simplistic approach. More often, experience and intuition become the deciding factor rather than a scientific approach that will more likely yield a better meaningful result.
Step 2: Store-Sales Ratio for Core and Sell-through for Fashion/trend
Store-sales ratio also plays a key part in shaping the model stock for Core products. Stock-to-Sales Ratio (SSR) is the amount of the inventory available compared to the quantity sold. It answers the question: For every unit sold, how many units should be on-hand?
Sell-through percentage represents the number of units sold to the amount of stock available of that product (opening stock plus bought received quantity). It helps to understand the velocity of the stock turn and helps to identify the sizes which are under-potentialised.
The combination of the cluster behaviour along with the store- to-sales ratio, helps determine the ideal model stock each store should have.
To help focus on the gaps and missed opportunities, a new approach would consider the following points:
- Post Clustering, employ and analyse the assortment within each cluster to understand the movement of different sizes and ask questions like: why do certain clusters sell a size at a certain rate and others do not? Is there a stock-out situation creating the gap?
- Deploy Supervised Machine Learning to train the algorithm to understand specific problems and to identify gaps based on constraints. This will help to bring in using machine learning.
- Create a Reference System for the unknowns. Historical sales can only be used as a base to an extent as it reflects sales including stock-outs. As a workaround, a reference system can be developed using the eCommerce back-orders, eCommerce digital conversion rate and traffic flow as well as wholesale order-book, to determine the starting point and follow the journey to track the lost sales.
Despite the efforts in analysing the sheer volume of data, we still experience inconsistencies in sizing affecting the profitability and bottom-line. The main factors we tend to overlook is the ability to capture the impact of binary sales or stock-out situations, customising the approach to the nature of the product and aligning to local demand.
The retailer’s estate could be spanning from a single store to multiple stores in one location to multiple stores in multiple locations. The demography plays an important role in influencing the buying pattern and therefore needs to be factored in while thinking of what sizes to buy and allocate.
The failure to address localised sizing demand will result in retailers not understanding the true demand and running the risk of ordering the incorrect stock. This would ultimately lead to overstocking or sell-out situations and unnecessary markdowns; thereby adversely affecting the margin.
Binary Sales or Stock-Out Situation
Mostly, the sizing analysis will be carried out for a set time period, ideally 6 weeks after the launch to capture the true demand. There may be instances where accounts, customers or styles have a delayed reaction or are slow sellers where sales are recorded as binary numbers. This would skew the true demand if not addressed properly.
The same applies to a stock-out situation. If the stock sells out in 2 weeks, the ability to capture the opportunity as well lost sales while planning for future is important.
Inventory Imbalance across the Store Network
The larger the size curve and the store network, the larger the risk of over-stock (inventory risk) and under-stock (sales risk). Size curves are generated for each cluster based on the collective performance of all stores within the cluster and rarely individual potential of the stores are looked at and addressed while allocating, unless there are known exceptions. Depending on how spread across the stores are within network, consolidating stock incurs a cost and is considered a last resort. This creates an imbalance across the network and drives the markdown spend-up thereby pulling the margin down.
Fractional Buying Problem in large Size Curves
It is an age-old practice within retailers to buy the fringe sizes shallow and core sizes deep and when it comes to allocating the sizes out, only the top, hand-picked stores based on intuition would receive the full size-set by default. The history, therefore, fails to record the actual demand for fringe sizes elsewhere in the store network and by spreading thin creates a fractional buy instantly from the start of the product life-cycle.
Products follow a predictable pattern during its life-cycle and go through a series of stages; beginning with the introduction, followed by growth, maturity or saturation and finally the decline of demand. The time span of stages of these products vary considerably depending on demand characteristics, life cycle stage, demand pattern, volume and planning horizon and can be classified as Basic/Continuity/Never out of Stock (NOOS), Fashion and Seasonal.
Basic/Continuity/Never out of Stock (NOOS) is characterised by a stable and predictable demand with a long-life cycle and do not change quickly with time and normally incurs a lower margin. They will be backed by heavy volume, shorter replenishment lead-time and a longer planning horizon.
Seasonal is characterised by very short life cycles, greater variety and high margins. The Life-cycle reaches saturation quickly, backed by moderate volumes, longer lead-times and shorter planning horizon. E.g.: Coats/Jackets
Fashion is characterised by short life cycles, low variability and high margins. The Life-cycle reaches saturation quickly, backed by moderate volumes, longer lead-times and shorter planning horizon. e.g.: sportswear.
Product lifecycle is an important deciding factor for the question: how much should we buy? It requires a customised approach to ensure we have planned for and supported the journey from entry to exit.
Due to the sheer volume of data to be analysed, it is imperative that it is cleansed to extract a meaningful, accurate dataset. Clear, well-defined data ensures stronger analytics and has less margin for error.
Addressing the gaps discussed earlier, the following steps help to deliver an effective size-curve:
- Create Store Clustering
- Review Each Cluster for Gaps and Sales Opportunities
- Rank Gaps and Sales Opportunities
- Generate Size-curve customised to the product Life-cycle
Effective clustering will not only used to analyse the historical performance of the assortment at Store/SKU level but also establishes an ideal assortment within each cluster. This will create a baseline for performance against which each store in the cluster can be compared. This comparison is then used to identify sales opportunities at the SKU and store level within each cluster.
Cluster performance as well individual stores performance will need be analysed in parallel to identify opportunity and true demand at regional level. This will ensure that the regional/local appetite is catered for and maximum potentialised. The opportunity will then be ranked by probability and logic ensuring the potential is captured and maximised and size-curve generated.
The customisation not only investigates ways to maintain the sales but also helps plan a hassle-free exit towards the end of the life0cycle. For launch products (seasonal and fashion), we would look into bringing in enough to launch, then building up rapidly to match the growth pattern as it saturates quickly. Steady state (NOOS) will be heavily built in the beginning to sustain the demand, with fringe sizes exiting first following by core sizes.