Sign up

Where traditional inventory allocation falls short

August 21, 2018

Most inventory allocation systems leave money on the table. Inventory sells out early at one store and is left unsold on the shelves at another.

While inventory redistribution among stores is a fact of life for all retailers, it is still essentially a fine paid on poor foresight. In order to maximize profits, inventory needs to go to the right store the first time around.

If sales forecasts and inventory distribution models do their job, there should be just enough inventory to meet demand with none leftover piling up on the shelves. This is exactly what our inventory allocation system aims to achieve.

Higher sales & lower inventory holding costs

The first component of good inventory allocation is a detailed sales forecast. Sales forecasts used to be based solely on the previous-month’s revenues: the store that sold the most last month would sell the most again this month. Nowadays, however, sales forecasts must account for a number of factors.

Our sales forecasts utilize artificial intelligence to process a large array of influential factors in real time. This includes classic elements like seasonality, store size, store location, and past revenues. It also includes non-traditional internal factors like changes in product mix, pricing, and discounting, as well as external factors like market trends, competition, and even weather.

As an example, our seasonality forecasts compute seasonality at multiple levels of detail. We consider seasonality most broadly in terms of the general item category and country. Then we delve deeper into specific seasonal effects broken down by item subcategory, the individual store and considering the impact of stock-outs.

Our fine-tuned sales forecasts attach a probability of sale to each particular item depending on where and when it is sold. Based on those probabilities, our inventory allocation model decides where and when to send out each piece of inventory.

Specifically, our models allocate inventory on two levels. First, we decide where to allocate the available inventory. Second we decide how much inventory to allocate to each particular store and how much to keep in the warehouse for refills down the line.

Our experience with retailers so far shows that our models make quite accurate forecasts. The result is higher overall sales and lower holding and redistribution costs.

The chart below illustrates these results.  Our algorithm delivered a 6% boost in overall sell-through (compared to a baseline forecast) for one of our larger clients over a six month period.

An easy hypothetical

As a highly simplified example of how our sales forecasts and inventory allocation models work, let’s use a retail ecosystem with only one type of product and three store sizes: small, medium, and large. We then make a table that lists the probability that each store will sell the first or second item it receives.

Such a simplified scenario makes inventory allocation easy. The first piece goes to the large store where it has 90% probability of sale; the second piece goes to the medium-sized store where it has 70% probability of sale; the third piece goes to the large store where it has a 50% probability of sale, and so on.

A complex reality

Of course, things get much more complicated with thousands of items distributed among hundreds of stores. The graph below shows the inventory distribution model for our client, Napoleon, in Italy.

The horizontal axis measures the amount of inventory distributed and the vertical axis measures the percentage of total items distributed among stores. Store sizes go from large to small as the colors move from red to yellow to green to blue.

As the graph shows, the first 25% of inventory goes primarily to large and medium-sized stores that have the highest likelihood of making initial sales. As these stores get more inventory, the probability of selling each additional item goes down, so inventory is instead sent to smaller stores where the probability of sale is higher. Eventually, distribution evens out, with each size store getting a roughly equal share of inventory.

Adding the human touch

While these sorts of artificial intelligence models easily outperform old static formulae, they still benefit greatly from human input. That’s why we make it easy for clients to integrate their experience into our models.

Perhaps a client has learned that a minimum amount of inventory must be available at all stores so customers are not put off by empty shelf space or unavailable items. We can easily impose this sort of constraint on the model at our client’s behest.

Or perhaps a client is aware of a major upcoming fashion event that will mean a large bump in traffic for certain stores. It’s no problem to adjust the inventory model for a limited time to capitalize on isolated events.

This mix of artificial and human intelligence yields the ultimate inventory allocation model — one that matches actual demand more accurately than any other system. No business can afford to leave money on the table, but sticking with outdated static inventory models does exactly that.

About the author

Will Freeman is a content expert at Evo.

He is a former economic journalist and part-time entrepreneur.

His interests include economic development, China, India, cryptocurrency and blockchain, and financial technology in general.

Hey! Was this page helpful?

We’re always looking to make our docs better, please let us know if you have any
suggestions or advice about what’s working and what’s not!

Send Feedback