Data-Driven Inventory Optimization - part2: dealing with Excessive Inventory

When it comes to inventory, companies are often exposed to a double negative effect, with shortages on the best-selling products and excess on the bad ones. In the previous part of the article, we have seen the optimizations related to out-of-stock; we will now look at the optimizations for limiting excessive inventory.

Excessive inventory accumulation is a major problem, it unnecessarily occupies extra storage space, blocks the company’s capital, and translates into money loss if it exceeds demand or becomes obsolete after a certain period.

High rates of slow-moving and obsolete inventory can ultimately affect profitability as inventory represents a huge investment of money that is not returned until the goods are sold. According to Chapman and Al, inventory represents 20 to 60 percent of all assets on an organization’s books. Not wasting money on excess and obsolete inventory frees up the cash the company can invest in other areas and limit the risk of unsold.

In the following pages, we will approach how to analyze excess inventory, identify the root causes and the actions to get rid of it. We will focus on quantitative analysis and optimizations, but for sustainable success, this must be supported by technology, processes and people.

1) How to analyze excessive inventory?

The KPIs for excessive inventory are Days of Coverage and SLOB.

a) Days of coverage (a.k.a Days on hand or Days of sales or Days of Supply) are the number of days the company holds its inventory before selling it. Basically, it is the inventory level expressed in days of demand rather than units in stock.

To accurately reflect the operational reality, a few calculation principles must be observed:

- Days of coverage should be calculated at the demand unit level (where SKUs are interchangeable) and then possibly aggregated to higher levels (with weighted average) to avoid compensation effect between low coverage and high coverage items that are not interchangeable

- Calculation should use forward-looking demand instead of historical sales, especially for non-steady environments (growing, declining, seasonal…)

b) SLOB is the Slow-Moving and Obsolete Inventory. Every business and sector has its own dynamics that influence the thresholds for classifying inventory as SLOB. But the main segmentation drivers are the days of coverage and the age of the inventory.

In the case of products with shelf life, Slow-Moving and Excess inventory thresholds must be set accordingly as a stock coverage higher than the shelf life, assuming a FIFO (First In, First Out) inventory means the product would expire before being sold.

SLOBs can be assessed as a percentage of total inventory. Below is a benchmark for Consumer Markets on excess and obsolete inventory.

2) The main causes of Excess and Obsolete Inventory

There are several common issues that drive excess and obsolete inventory.

a) Biased demand forecasting

Poor forecasting and especially over-forecasting are among the biggest drivers of excessive inventory. When analyzing forecast quality, it is essential to look at two KPIs:

• Bias that gives the overall direction of the error. Forecasts are, on average, too high (over-forecast) or too low (under-forecast)?
• Precision that gives the magnitude (variance) of the errors.

We want to have a forecast that is both precise and unbiased.

There are many reasons why bias exists. Some of them affect machine forecast and others judgmental forecast.

Machine forecast related bias:

• Wrong forecasting model: for example, if the data serie includes a trend but the model does not, the forecast will be too low or too high.
• Wrong accuracy measure for model optimization: for instance, the MAE is minimized in expectation by the median, so if the future demand is asymmetrically distributed and the model was optimized for the MAE, it will end up with biased forecasts. This is most problematic for low-volume data or intermittent time series, since they are asymmetrically distributed.

Judgmental forecast related bias:

Sales teams insights can be of great value for demand forecasting, especially for responding to sudden demand shifts where statistical models are slower to respond. However, this can introduce the following biases:

• Confidence bias: usually happens with sales teams who are by nature prone to resilience in the face of rejection, which is a great quality given how often a salesperson is told “no”. However, this optimism may cause salespeople to overestimate their close rate.
• Self-serving bias: happens when sales team intentionally increases forecasts to ensure sufficient production/inventory to protect themselves against poor service level and angry calls from customers complaining about delayed orders and out-of-stock.
• Sandbagging bias: happens with systems where rewards and bonuses are based on exceeding the forecasts, inciting people to intentionally lower forecasts.

Multiple methods exist for measuring bias and precision, each with its advantages and disadvantages. As we have seen, optimizing the wrong accuracy measure can introduce bias as well as judgmental forecast adjustments can do.

The Forecast Value Added Analysis (FVA) helps provide the full picture of the forecasting process performance and understand how each step improves or deteriorates the forecast accuracy and bias.

Measuring Forecast Value Added will also improve ownership and accountability of the forecast by the different stakeholders.

The demand consensus meeting and risk & ops review of the Sales & Operations Planning process (S&OP) bring an opportunity to reduce bias by establishing a healthy tension between sales, demand planners and operations.

b) Too high MOQ (Minimum Order Quantity)

Too high MOQs vs. demand increases the inventory level and risk of obsolescence.

Excessive MOQs can be analyzed by looking at their days of coverage. MOQs with more than a threshold value (90 days for example) should trigger a red flag and call for corrective actions.

c) Too much safety stock

In the previous part of the article, we have seen the role and importance of the safety stock to protect against higher demand or lead times than predicted, and that it is calculated from a desired service level. An important fact is that the relationship between service level and safety stock is non-linear; higher service levels require disproportionally higher safety stock. For example, increasing the service level from 90% to 98% would require an increase of safety stock by 1.6 times.

J. Vermorel proposes a formula to calculate the optimal service level given the stock holding and stock-out costs.

This formula is based on the assumption that costs, both storage and out-of-stock, are linear.?

For the sake of simplicity, rather than setting service level targets for individual products, it is recommended to do it on groups of products based on criteria such as strategic importance (bottlenecks, high cost of stock-out, high carrying cost…).

The original King’s safety stock formula uses the standard deviation of demand as the proxy for demand variability; however, if the company has demand planning capabilities, the standard deviation should be replaced by the forecast error, which is a better estimate and more likely lower resulting in fewer safety stock.

Also, in many companies, safety stock covers other causes than demand and supply variabilities like IT or logistics issues. In such cases, instead of increasing safety stock, it is recommended working on root causes.

Probabilistic forecast, a forecasting approach that assignsa probability to different possible outcomes instead of predicting a single (point) expected outcome, brings new ways of optimizing inventory with no need for safety stock (at least to protect against the variability of demand) as the uncertainty becomes part of the forecast itself. To calculate the optimal inventory level, the expected cost for each supply scenario is computed based on the possible demand outcomes weighted by their likelihood.

This approach is made possible at scale thanks to increased computing power.

d) Incorrect Data & Settings?

Replenishment decisions are calculated based on defined algorithms and input parameters like demand forecast, inventory on-hand, lead time, order cycle, MOQ, etc. While some inputs are calculated with a high frequency like demand forecast and inventory on-hand, others are rarely revised like lead times, order cycle or safety stock. Over time, gaps can appear between the initial values set in the system and the actual observed values.

Setting up a routine for parameter revision and update at least once a year will allow sanitizing the systems from wrong parameters that can negatively impact replenishment decisions.

d) Uncontrolled SKU proliferation

While adding new SKUs is an essential part of growth, expanding the product portfolio increase complexity on operations driving hidden extra costs and inventory.

We will further explore the product portfolio optimization in a dedicated article but best practices include:

• Regular portfolio review and rationalization of low volume/margin items for which the costs induced by complexity are higher than the profit value.
• Setting sales forecast thresholds for new SKU launches to foster candidates with high potential and do a No-Go on low potential candidates.

3) What to do with excessive inventory

After identifying Slow-Moving and Obsolete inventory and working on the root causes to prevent further accumulation of unnecessary stock, how to deplete it while maximizing its residual value? Some options are more relevant for companies that sell finished goods while others are more suitable for companies that work with raw materials.

• Return inventory for a refund or credit: if allowed by the supplier, this might be the best option.
• Remarket / re-merchandize: sometimes, the reason a product is not selling lies in how it is marketed and positioned and not in the product itself. If this is the reason, refreshing the marketing and merchandising might boost sales. A handy tip for slow-moving inventory is to double their exposure by displaying them in multiple places in the stores. For online channels, take news photos, use new titles, description and keywords.
• Rebalance inventory: move inventory from retail location where it is in excess to a retail location where that inventory is in demand.
• Bundle products: group slow-moving products with fast-moving ones and sell them for a slightly lower price than if bought separately.
• Offer a discount: if remarket and re-merchandize do not work, lowering price will make the products more attractive and increase sales.
• Re-use the inventory on new products: if the materials or components can be used to make other products.
• Liquidate Items: in most industries, there are liquidators that buy leftover inventory at a steep discount to resell it. Even if goods are sold to liquidators at or below cost, it’s still better than writing off them as a loss.?
• Donate to charity and reduce tax: obsolete inventory might be donated to charity. Not only is this much preferred to disposing of the items, but it can make organizations eligible for a tax deduction.?
• Write-Off Obsolete Inventory: at a last resort when all other options are exhausted, obsolete inventory must be written off as a loss.?