Posts Tagged ‘size optimization’

Fashion Forward – Part 3: Optimizing Size and Pack (2 of 2)

by Ziad Nejmeldeen • June 22nd, 2011 • 2 Comments »

By Ziad Nejmeldeen, VP of Science, Quantum Retail

Pack Optimization

As previously stated, Pack Optimization refers to finding the optimal configuration of one or more packs that should be utilized for a style-color. This often seeks to balance the pack’s ability to meet store/item demand with the increased handling costs of having too many pack quantities or large quantities of “eaches”, which we will refer to as Loose.

Several U.S. and Canadian retailers have estimated that the handling costs from vendor to DC to store is on the order of $0.25 – $0.50 per pack; this is true whether the pack contains 1 unit or 10. So, we want packs to contain as many units as possible. However, large packs introduce allocation inefficiencies as we are now forced to put a pack into a store to meet the demand for some sizes, even if it means putting in too much inventory of another size. So, we want the packs to contain as few units as possible. By creating a profit function that includes both of these effects, we can find the right pack size that maximizes overall profit.

There are various approaches used to accomplish Pack Optimization; we describe them here in order of complexity.

Single Pack (Bulk): Some retailers choose not to use packs that contain multiple sizes (Ratio Packs), instead relying on Bulk Packs that contain quantities of the same size. In this case, the shape component of the pack is a moot point. However, we still need to determine the volume. In the simplest case, this is a given value (or tight range of values) based on vendor or operational constraints. A more sophisticated approach relies on a data driven exercise to balance the pack’s ability to meet store/sku demand with the increased handling costs of having too many pack quantities.

Single Pack (Ratio): The shape is driven by the optimal size profile associated with the size-range and merchandise area of the item. Identifying the volume is similar to what we described above in the bulk case. The one additional complexity here is that the recommended ratio pack is also dependent on whether bulk or loose will also be utilized. For example, if we do not have bulk or loose, we are more inclined to choose a pack volume that leads to minimal shape distortion; this distortion can occur after the pack volume is multiplied by the size profile and the quantity associated with each size in the pack is rounded to the nearest whole number.

We now describe several different cases that are usually all referred to as “Multiple Packs”.

Multiple Packs (Single Pack in Store Group): The case described here allows a style-color to be ordered in multiple pack configurations, but each configuration is earmarked for a group of stores. In essence, each store group is dealing with the Single Pack case for the style-color, but the style-color can have the pack configuration vary across store groups. This problem is tackled by clustering the store size profiles to create store groups. Each group has an average size profile, and we proceed for each store group using the Single Pack case.

Multiple Packs (Launch vs. Replenishment): An item may have two types of packs that can both go to the same store, but not at the same time. This is often the case when we have one type of pack allocated at the beginning of the season (Launch or Initial pack) and a second, usually smaller, replenishment pack allocated after the item has begun to sell. We may again have operational constraints dictate the sizes of the two packs. Alternatively, we can employ a similar data-driven logic as described in the Single Pack case, but with the Initial (e.g. first 6 weeks) and Replenishment time periods broken out.

Multiple Packs (True Multiple): The most complex scenario allows for multiple pack configurations to exist for an item, with the multiple configurations eligible for allocation to the same store at the same time. Given a range of stores with different volumes and with different size profiles, it is unclear whether we should use packs that a) have the same shapes but different volumes, b) have the same volume but different shapes, or c) something in between. Answering this problem requires running a simulation that assesses packs in combinations. Each combination is measured on both its ability to meet store demand and the handling costs involved. Even the simplest two-pack case can be computationally intensive depending on the number of stores analyzed.

Order Planning/Warehouse Replenishment

Suppose we have completed Size and Pack Optimization, so that we have an entire library of size profiles defined for different size ranges, merchandise areas, and locations. Additionally, we have specified the different pack types (ratio or bulk), have settled on their configurations (shape and volume), and have made a decision on whether or not to allow loose stock. Now comes the problem of cutting a purchase order: How many units of each type of pack should be ordered? How much in loose?

We begin with a pre-season forecast of demand at style-color/store. This forecast can stem from a number of sources; e.g. like-item(s), or a spread-down of a buyer/planner’s chain forecast to store using store weights / volume groups. We can now use our size-profiles to spread the forecast further and arrive at a pre-season estimate of size/store demand.

We can now simulate an allocation to find the quantities of packs and loose that balance the value gained from fulfilling size/store demand (100% loose is best) with the cost of pack handling (100% of largest pack is best). The simulation returns the pack quantities that we wish we had if it was time to allocate today; these quantities constitute our order.

Note that this simulation is a bit more involved if we have both launch and replenishment packs since we will need to separate size/store demand into a launch and replenishment period and evaluate the two separately.

Allocation

At the time of allocation, the pack quantities have been received in the warehouse and are ready for distribution. We can now repeat the same process employed in cutting the purchase order to arrive at an updated size/store demand estimate.

If the initial allocation quantity is pre-determined (e.g. all launch packs, or fixed % of inventory is to be allocated), the allocation problem is a simplified application of the same logic employed when the purchase order was determined; we know what we want to push, we know the size/store demand, and we try to marry the two up as well as we can. On the other hand, if we are also optimizing the initial allocation quantity, the problem is a bit more complex; we must determine the appropriate amount of inventory required in each store to meet expected demand, and the right amount of buffer inventory to meet unexpected demand.

As we continue from initial allocation (before we have observed the item sell) to replenishment (after it begins to sell), we gain some important pieces of information. First, we have observed size/store demand, which we can use to improve our forecast of size/store demand; note that it is at this point that we cease to rely on the size profiles that were so critical in determining the pre-season buy and initial allocation. Second, we have observed the variability (spikiness) of demand within each location; this goes a long way in determining how much buffer inventory we need to capture unexpected demand – the more variable the demand, the more buffer inventory required.

At this point, the remaining packs can be evaluated for the profit we expect them to generate in each store based on the store’s size demand and underlying variability.  If we have a combination of both packs and loose available, we can factor in the pack handling costs into the profit calculation (this makes packs more attractive than loose), or we can rely on a pre-set preference for what gets replenished to stores first (e.g. packs, then loose).

Conclusions

It is our sincere hope that the concepts in this blog help to inform and drive improvements in the management of sizes within fashion retail. We conclude by revisiting what we believe to be the three most important ideas related to this topic:

  1. Size Ranges: Grouping products together with disparate size needs under the same size run will lead to a size profile that does not fit any item well, regardless of what care is taken in estimating demand. Use product attributes wisely to create meaningful size ranges and you can avoid this problem.
  2. Multiple Packs: There is a disconnect if you are looking for a pack solution that gives you multiple pack configurations, but you do not possess an ordering/allocation solution that can utilize multiple packs. In this case, consider the feasible alternative of having multiple packs, but with a single pack per store group.
  3. Demand Variability: When replenishing packs in season, it is important that you not only consider the observed demand in each location, but the variability of demand as well. This can change the amount of inventory that is desirable in the store, and you may now prefer replenishing a larger pack to a store in place of either a smaller one or loose stock.

Look out for next week’s blog on the benefit of multiple allocations and hold-back stock next week.

Sign up to receive updates throughout this series.

2 people like this post.
Unlike

Fashion Forward – Part 2: Optimizing Size and Pack (1 of 2)

by Ziad Nejmeldeen • June 8th, 2011 • 1 Comment »

By Ziad Nejmeldeen, VP of Science, Quantum Retail

Size and Pack Optimization (SPO) is particularly important to the fashion industry. Getting size and pack correct is the key to serving your customers well, and ensures that you get the most dollars out of your inventory investment.

Size Optimization finds the correct ratio of sizes to carry in each store, based on observed size-store demand. Within fashion, the ratio of sizes is defined across a style-color. If “Store” is too granular given existing technology and processes, stores can be clustered as part of the Size Optimization work.

Pack Optimization is often used as a general term to refer to one of three very different processes:

  1. Pack Configuration: this finds the right pack configuration(s) to be used in ordering from a vendor.
  2. Pack Ordering: this finds the optimal amount of each (pre-configured) pack (and possibly loose/bulk) that should be ordered.
  3. Pack Allocation: this optimally allocates packs (and possibly loose/bulk) to stores given known quantities at the DC.

Unless otherwise stated, any reference to Pack Optimization in this document implies Pack Configuration Optimization.

We will be making use of several terms, some of which have yet to become entirely commonplace; some definitions should help:

  • Pack (aka Prepack): A combination of several sizes belonging to the style-color; such a pack is sometimes also referred to as a Ratio Pack
  • Bulk (aka Bulk Pack): Multiple units of the same size
  • Loose/Eaches/Bin: Single units of individual sizes
  • Size Run: The “run” of sizes over which we are optimizing. For example, XS – XL vs. S – L, 2-16
  • Size Range: This is often used interchangeably with Size Run. However, we use Size Range to specifically mean a Size Run that has been combined with attributes of interest. For example, XS-XL Slim Fit or 2-16 Light Colors.
  • Size Profile: A percentage associated with each size in the Size Range such that the sum is 100%; the output of Size Optimization are Optimized Size Profiles
  • Demand: What could have sold if inventory was fully available; it is the ratio of demand across sizes within a size range that yields a size profile.

Relevance in Fashion

In this blog, we will focus on how SPO influences decisions pertaining to Order Planning/Warehouse Replenishment and Allocation. However, we note that there are several peripheral uses of SPO outside of these processes. For example, range planning (determining eligible stores within the assortment planning process) can utilize SPO to determine the categories and stores where we can increase/reduce fringe sizes.

Order Planning/Warehouse Replenishment: At the time a purchase order is placed for an item, Size Optimization informs future store/size (relative) demand while Pack Optimization specifies the pack configurations that should be utilized. These are both essential to creating a purchase order, but they are not enough. We additionally need to “simulate” the allocation of these packs to meet store demand in order to determine the specific quantities of each of the packs and/or bulk that should be ordered; more on this next week under “Order Planning”.

Allocation: When we initially allocate prior to observing in-season demand, the results of Size Optimization are used to spread a pre-season demand estimate from style-color/store down to size/store. This is critical in Initial Allocation when stores are given a specific distribution of sizes to meet anticipated demand. While Pack Configuration Optimization is not relevant in this process (the pack configurations are already specified), we will require an intelligent Pack Allocation process; the lack of such a process can severely limit the options (and therefore value) available in Pack Configuration Optimization.

Size Optimization

Demand: A key input into creating Optimized Size Profiles is proper demand estimation. Demand estimation should ideally use historic store/sku/day sales and inventory information to assess the sales that could have been when inventory is stocked out. Basing the size profiles on demand – especially demand prior to markdown – allows us to avoid repeating the mistakes of the past.

Basing size profiles on demand is superior to the traditional approach wherein sales are aggregated over some specified time period. The goal of this approach is to use sales prior to stock-outs; variants of this approach include using:

  • Fixed number of weeks after the item starts
  • All weeks prior to markdown
  • Weeks prior to some % of sizes stocking out

Regardless of which of the above we use, this approach invariably suffers from one or both of the following: i) the product’s life cycle is artificially constricted, with important periods impacting profitability omitted, ii) some sizes may still exhibit pockets of stock-outs over the period evaluated, yielding biased results.

Size-Ranges: Well-defined Size Ranges constitute a second crucial input into Size Optimization. Recall that a Size-Range segments a Size Run by attributes; these attributes can include shape or cut, color, fabric, price-point, season, to name a few. When several attribute options exist, an analysis can be conducted as part of the size optimization service to determine the set of attributes that yields a statistical significance for size differentiation.

Why is it so important to define size-ranges well? Consider the following two alternatives:

  • Poor Size Runs: The Size-Range is first and foremost dependent on the Size Run being defined well. Suppose you are a retailer where some items are carried in Small, Medium, and Large while other items are carried in X-Small, Small, Medium, Large, and X-Large. However, rather than assigning the former to Size Run S-L and the latter to XS – XL, all items are assigned to XS – XL. This is a poorly defined Size Run. Any analysis based on this Size Run will invariably lead to a Size Profile wherein the % applied to X-Small and X-Large is too low; it is left to the reader to think carefully why this would be the case.
  • Ignoring Key Attributes: Suppose we have a Ladies Pants category where all items are carried in Size-Run 2-16. However, some items in this category are classified as Petite while others are classified as Casual. If we make use of these attributes, we find that Petite has more demand in the smaller sizes while Casual has more demand in the larger sizes – we can use this information to buy properly pre-season. However, ignoring these attributes leads us to buy a mixture of the two demand profiles for all items in the category; the end result has us markdown Size 16 “Petite” and Size 2 “Casual” at the end of the season.

Optimization Frequency: An important question to consider is how often we should update Size Profiles. The answer depends on several factors:

  • What is the window of time on when the product is ordered before it begins selling?
  • Are the Size Profiles season-based? For example, did we find that there was a statistical significance in having them vary by quarter, or are they the same for the entire year?
  • Do we believe the demographics of the customer base has been changing drastically?
  • Has a merchandise reclassification recently occurred?
  • If the retailer is not private label, have new labels been carried in the store recently where the size fits may not be the consistent with other labels?
  • Is Size Optimization offered as a service or product? What is the cost of rerunning?

Let us consider the following case study to illustrate how to best determine frequency. We have a retailer who has run Size Optimization at the end of Year 1 based on data in Year 1. The retailer decided that, due an observed increase in demand for larger sizes during Christmas, Size Profile should vary by season; quarter was used as part of the Size-Range definition. Additionally, this retailer must order items for each season 16-weeks in advance (which is relatively short). Suppose we are now in Year 2, and Q1 has just ended; the retailer is assessing whether they want to re-run Size Optimization based on the last 13 weeks of data.

If Size Optimization is run, it will not update Size Profiles for Q2 through Q4 of Year 2.Those size profiles are based on corresponding quarters from Year 1 and will remain fixed; the rerun will only impact Q1 of Year 3. However, the retailer has the option of waiting until the end of Q2 and re-running Size Optimization on Q1 and Q2 of Year 2. Note that at the end of Q2, the retailer is 26 weeks away from Q1 of Year 3, which gives us 10 weeks to rerun Size Optimization and still make the 16-week order window before Q1 of Year 3 begins. In this case, running Size Optimization twice a year is a reasonable option.

Look out for Part 2 of Ziad’s blog on Optimizing Size and Pack next week.

Sign up to receive updates throughout this series.

2 people like this post.
Unlike

Fashion Innovation Series: Part 1 – Size Optimization & Pack Optimization

by Ziad Nejmeldeen • April 20th, 2010 • Write a comment »

This is the part 1 of a 4 part series on Fashion Innovation and Optimization.

KEY TOPICS IN THIS SERIES:

  1. Size & Pack Optimization
  2. Assortment & Range Planning
  3. In-season Replenishment
  4. Allocation Optimization

You can also follow our 4-part 2010 Retail Outlook series here.

If you’d like to be emailed PDF’s of this series as they come out, make sure to sign up for the email series updates! (we will only send you email for our retail series reports)

In fashion retail, Size and Pack Optimization are key

Local demand changes at every store on a daily basis. Clustering stores together by store size and geography might simplify the process, but is inefficient and does not take into consideration individual store patterns for size, color, style and quantity of local demand and product preference.  Retailers need to monitor the changing demand at every store to align their assortment in the way that is most profitable and aligned to their strategic objectives.

It sounds like a no-brainer, but when supply chains become complex, retailers cannot keep up with store level demand and will send the same amounts of every product to similar store types. However, localization of store level assortments and order plans is proven to increase availability, full price sales and customer satisfaction. It is also proven to reduce overall inventory, wastage and markdowns which all erode margin.

Understanding how a product will sell through its entire life on a location by location basis – is essential for:

  1. Meeting sku/store demand: i.e. avoiding missed sales opportunities
  2. Reducing sku/store over-allocations: which would otherwise be dealt with through markdowns
  3. Minimizing handling costs: as the inventory makes its way from vendor to warehouse (where applicable) to store
  4. Reducing Markdowns: by having the appropriate level of inventory and the best assortment possible

The initial assumption of the product assortment is an important part of the process. Retailers need to know what is selling where and why, they need a strategy and goal for why that product is in their assortment and they need to make sure they can  continuously re-evaluate how they expect the product to sell – in real time. This enables retailers to understand which stores will offer the greatest potential for full price sales – and appropriately decide what inventory is best and where.

When they can pinpoint the demand at their stores – they will cut distribution costs and decrease lost sales. With the ability to assign specific pack sizes will also help retailers get the exact amount of inventory to every store and reduce markdowns.

Get the right product in the right place and fulfill based on product performance //

The objective is clear: get the right product in the right place to start with – then fulfill based on how products are really performing at each store – giving the product the best chance to sell at full price and identify when and where markdowns are truly necessary.

Size, pack and prepack innovation for progressive retailers

Size Optimization uses historical sales and inventory data at the size/store level to infer historical demand, and then aggregates demand across groups of items and/or locations. Items are grouped according to the size run, attributes of interest, or merchandise classification that they share. This aggregated demand, when normalized across the sizes that compose a size-run, yields a Size Profile. These size profiles can be used pre-season to impact the size buy for the chain, or in-season to impact store-specific size allocations.

Prepack Optimization refers to the pre-determination of  prepacks that contain fixed quantities of each size in the size run. Like size profiles, prepacks can be defined for groups of products where the grouping is defined by size run, specific attributes, or a common merchandise classification. Unlike size profiles, prepacks are not store-specific – a given pre-pack can be allocated to several stores, if not the entire chain.

In the most trivial cases, Prepack Optimization can be considered a by-product of Size Optimization. Suppose that we want an n-pack solution, have designated that each store should only receive one type of pack, and have pre-determined the approximate number of units in a pack. Then, we can cluster store-level size profiles into n clusters, and use each cluster size profile to determine the optimal cluster pack by multiplying the size profile by the pre-determined number of units and rounding the resulting size units to the nearest whole number.

However, pack optimization becomes more interesting when each pack in a solution can go to all stores, or when the pack quantity range is broad, thereby requiring optimization of the units in the pack. In these cases, you need more sophisticated approaches to obtaining the optimized packs – approaches that utilize historical store/size demand, allocation quantities, and pack handling costs.

Localizing sizes and packs and rationalizing SKUs:

In order to optimize sizes and rationalize SKUs at a local store level you need an acute awareness of product behavior. There are dozens of product behaviors unique to every store. In order to analyze these behaviors, retailers should optimize by style, color, brand, promotion, price, and seasonality at each store.

The concept of localization works on two levels:

  1. Retailers can look at the unique behaviors of every product – to determine each stores’ selling patterns for size, color, style, quantity, brand, season, etc. With this understanding, a retailer can plan orders on a store-by-store basis to deliver the right amount of the products that customers are buying at each location, allowing the retailer to achieve the highest turn rates, reduce inventory to the appropriate levels, reduce over stocking and stock outs and ultimately increase margin.
  2. The second concept of localization comes from localizing distribution, optimizing routes, re-locating product in the most optimal way, or utilizing vendors that are in a short vicinity of each store.

Size Optimization Overview:

Size Optimization refers to finding the optimal ratio of sizes to carry for given product in a given store. After segmenting products by Size Run (e.g. XS – XL vs. 2-16) and attributes of interest (e.g. Shape, Color, Fabric), the optimal ratio is found by looking at historic demand, which incorporates actual and lost sales. Size profiles for each group of products are computed at the store level, where enough data exists. A number of Quality Assurance steps are applied to the final output to capture and correct for any exceptions. The client can use the size profiles to both impact the size buy pre-season and the store-level allocation in-season.

Pack Optimization Overview:

Pre-Pack Configuration Optimization refers to finding the optimal configurations and sizes for a combination of packs. Optimality is defined in terms of maximizing an objective function that includes handling costs, lost sales, and markdowns (or wastage).  Pack Optimization involves choosing packs such that the increased profit from sales increase and waste reduction more than offsets any increase to handling costs.

Implications of changing pack size:

As the pack size decreases:

  • Handling Costs Increase: we are ordering roughly the same quantity as before, but doing so with more packs. Assuming a given cost per pack (typically 30p), we can compute the increased cost.
  • Sales Increase: greater sales are achieved by allocating more units to a store where the pack size restriction was previously a barrier.

Ultimately, you can arrive at combinations of packs that work well together to meet store/size demand and minimize handling costs without excessive over-allocation of sizes.

Get back in the game

Are you ready this year to know exactly what your customers are asking for at every location and to have the ability to react as their wants change? If you are looking for a solution that can drive momentum for your business this year, check out the solutions offered by Quantum Retail. Our customers see valuable results in 8 to 12 weeks, and our implementation approach gives your team access to the system from the beginning, so you can manage changes to your processes with ease. Quantum Retail continues to help all of its clients drive positive business value more rapidly than anything seen in retail.

For more information, visit: http://quantumretail.com/services/size-pack-optimization

Download this blog as a PDF.

2 people like this post.
Unlike

The truth about size optimization, SKU rationalization and localization

by Quantum Retail • February 4th, 2010 • Write a comment »

Getting the right sizes, colors, styles and quantities to the right location

Local demand changes at every store on a daily basis. Clustering stores together by store size and geography might simplify the process, but is inefficient and does not take into consideration individual store patterns for size, color, style and quantity of local demand and product preference.  Retailers need to monitor the changing demand at every store to align their assortment in the way that is most profitable and aligned to their strategic objectives.

It sounds like a no-brainer, but when supply chains become complex, retailers cannot keep up with store level demand and will send the same amounts of every product to similar store types. However, localization of store level assortments and order plans is proven to increase availability, full price sales and customer satisfaction. It is also proven to reduce overall inventory, wastage and markdowns which all erode margin.

Optimizing sizes and rationalizing SKUs

In order to optimize sizes and rationalize SKUs at a store level you need an acute awareness of product behavior. It does not make sense to only optimize on size – if a retailer is going to take the time to assess sales and demand at a store-by-store level, they should rationalize SKUs at a local level by using store data.

There are dozens of product behaviors unique to every store. In order to analyze these behaviors and make the most of their efforts, retailers should optimize by style, color, brand, promotion, price, and seasonality at each store. But the complexity of this exercise can become time consuming when a supply chain is vast.

There are however, technologies available that can simplify this process and make it ongoing, creating a strategy for these attributes and applying it to all levels of inventory management, from order planning, allocation, replenishment, forecasting and distribution.

The concept of localization works on two levels:

  1. Retailers can look at the unique behaviors of every product – to determine each stores’ selling patterns for size, color, style, quantity, brand, season, etc. With this understanding, a retailer can plan orders on a store by store basis to deliver the right amount of the products that customers are buying at each location, allowing the retailer to achieve the highest turn rates, reduce inventory to the appropriate levels, reduce over stocking and stock outs and ultimately increase margin. The second concept of localization comes from localizing distribution and utilizing vendors that produce products in the vicinity of each store. This type of localization is most easily applied to fresh foods and markets – where customers prefer to support their local farmers and local brands.
  2. When retailers realize that they cannot optimize sizes and packs unless they have an awareness of store demand, stock outs, and customer behavior at the local level, they quickly become more profitable and able to compete in today’s retail market.

Q – the quickest and most profitable solution for size optimization, SKU rationalization and localization

The Q system continuously monitors business strategies, profitability, service levels, stock levels and 35 different aspects of behavior for every product in every store. Q is so intelligent that it learns from data like stock outs, lost sales, slow movers, lumpy sellers, packs, sizes, colors and styles. It takes the most recent data for each item and automatically recommends inventory movement decisions driving toward your corporate objectives. Plus, it optimizes the way you phase in a new product and phase out another – ensuring that you are always reaching your optimal performance, sales and service levels, giving you the highest return on the inventory you are buying.

CLICK HERE for more information on Q

Get resources on how to adapt to the challenges of today’s retail market HERE »

Download this blog as a PDF

Be the first to like.
Unlike