Artificial Intelligence for Inventory Management.

Good inventory management revolves around a single contradiction: keeping enough stock in the warehouse to ensure the business keeps moving but not enough stock to drain its limited cash reserves. This contradiction lies at the heart of the role of store manager. It’s a job where boring is best, where every need of the business is anticipated, where many urgent calls are crisis calls of someone else’s making, yet where the inventory team need to find a solution. All done while simultaneously not sinking all of the company’s cash into non-moving stock.

Inventory management is best orchestrated by a small, tight knit crew.

​The members of a tight, well-greased inventory team are a lot like a engine room crew of a naval ship. Confined for most of their waking hours in hot, oily spaces, a loyalty to no one but each other, rarely appreciated by the rest of the company despite the fact that they’re keeping the entire operation afloat and moving.

One item out of stock can bring a business to its knees. Elon Musk relays an anecdote in his fireside chat with Salman Khan, where one of Tesla’s suppliers had, for a myriad of painfully erroneous reasons, failed to deliver USB cables. This simple, $3.00 item brought the entire Tesla production line to a standstill as the team literally drove to every electronic store in the Bay area to try obtain USB cables for the car’s computer system.

So, how does one bring artificial intelligence to the messy, fast-paced world of inventory management? Can it actually be of use? The answer is absolutely yes. But, like in any real-world application of current state-of-the-art artificial intelligence, it needs to be carefully teamed up with the human oversight and used as a part of the system — it does not replace the system.

That being said, the use of artificial intelligence in inventory is yielding hefty and impressive improvements for the companies that are utilising it. We are on the verge of a major upheaval in the way inventory is managed. This revolution is a result of the availability of the huge amounts of real-time data that are now routinely generated on the internet and through the interconnected world of enterprise software systems and smart products. In order to make effective use of this new data and to stay competitive, managers will need to redesign their supply-chain processes. Amazon, for example, implemented artificial intelligence throughout their inventory operations, at an unprecedented scale. In almost every aspect of their operations, A.I methodologies such as time series prediction and reinforcement learning systems are being deployed. User demand, supplier backorders, warehouse optimisation, stock levels are all being guided by either machine learning or more complex artificial intelligence systems.

Amazon has applied A.I to inventory management at an unprecedented level

Those companies bringing artificial intelligence into their Supply Chain are see major improvements.

If companies are looking to keep up with the likes of Amazon and others, they need to be looking at whether (or where) artificial intelligence can be deployed to assist their operations.

​Generally speaking, there are two key implementations of artificial intelligence for inventory. Of course, the likes of Amazon have deployed many more but for now these two can yield strong improvements.

  1. Demand Prediction for Inventory Management. This is the simplest approach and, if implemented correctly, can be very informative. As the name suggest, the general idea is to build a time series prediction model that can estimate what demand will be like for the coming days across all items in your inventory. If your company has its own dev team that are familiar with machine learning, we’ve found that some of the highest performing time series methods are currently lstm/rnn models with sliding windows, old school logistic regression with a few tweaks and finally certain probability models, One of the joys of demand prediction is that you can incorporate external data sources into the system to see if they have an impact on demand. We have built systems that ingest the weather data to see if that impacts on prediction — for example we recently undertook a proof of concept for a Fortune 100 company that uncovered a strong causality between temperature above 30 degrees Celsius and part failure. This was done by simply feeding weather data and their inventory data into a single artificial intelligence model and the AI did the rest.
  2. Reinforcement Learning systems for full-inventory management. This is the more advanced artificial intelligence approach that involves a model taking serious control of the inventory operations, with human checks and balances. Reinforcement Learning is a domain in artificial intelligence where the models don’t simply make predictions or classifications, but actually act on these predictions. It’s about giving an artificial intelligence the option to act on what it’s predicting. This is done by rewarding and punishing the model for acting incorrectly. In this case, we typically establish punishments for letting an particular inventory item run out of stock, we also punish the model for stock too higher value for too long. For rewards, we primarily focus on ordering items within a safe window before the demand. Reinforcement Systems are hard to implement oneself without prior experience — you need familiarity with simulation models and RL to get anywhere with Inventory. But when done right, they yield phenomenal results. One of our implementations saw a 32% reduction in costs across the operation.

Prior to beginning any implementation, you need to look at three factors:

Integration with your existing inventory management — depending on the size of your company, you may be using SAP, Xero or any other myriad of software for your inventory management. We typically integrate the artificial intelligence model with their APIs and build a separate dashboard for the inventory team to quickly obtain insights.

These models are data hungry. Unfortunately artificial intelligence systems are extremely data hungry and typically require a few years of inventory data to build a reasonable model. Whenever clients ask us how much data we’d like, we answer ‘everything you can give us’. This can be the biggest problem for Artificial Intelligence implementations at the moment, simply not enough data.

Inventory items are not homogenous. It’s also important to remember when looking at artificial intelligence models for inventory that each item is different and needed to be treated differently. There are some items are highly predictable and regular in their movement, whereas some items are highly unpredictable but nonetheless equally vital to keep in stock. Think about undertaking significance testing prior to building any artificial intelligence implementations. This allows you to understand what external data is important for being able to predict an items demand and to also see which items are predictable or not.

As you start building, you need to consider the following:

A Proof of Concept (POC) Pilot.

These artificial intelligence models are complex, and not always applicable, especially if the data is insufficient. Therefore companies should create pilot projects for artificial intelligence applications before rolling them out across the entire enterprise. This allows teams to prove the model will provide sufficient ROI, to educate their teams about the incoming technology. Most importantly, introducing the more advanced artificial intelligence systems are like bringing on a new company role — the team needs to adjust to how they interact with this new company role and team member. It’s vital that the existing team have a chance to provide feedback on how to best integrate the artificial intelligence and how they will work with it. A POC pilot is the perfect means to do this.

Business-process redesign. As artificial intelligence inventory projects are developed, think through how workflows might be redesigned, focusing specifically on the division of labor between humans and the AI. In some reinforcement learning projects, 80% of decisions will be made by machines and 20% will be made by humans; others will have the opposite ratio. Systematic redesign of workflows is necessary to ensure that humans and machines augment each other’s strengths and compensate for weaknesses.

For those people looking to bring Artificial Intelligence into their business operation, inventory can yield profound impact. It’s definitely not a be-all-end-all solution, but it does provide powerful insight that can help your inventory team better manage the day to day tasks and to uncover interesting trends in high volumes of data.

By Alasdair Hamilton


Remi AI

Stay tuned on the Remi AI blog as we build out the complete supply chain offering, or you can sign up to our Remi’s Recommended Reading newsletter here.

Or, if you’re ready to start seeing the benefits of A.I-powered inventory management, start the journey here.

Who are we?

Remi AI is an Artificial Intelligence Research Firm with offices in Sydney and San Francisco. We have delivered inventory and supply chain projects across FMCG, automotive, industrial and corporate supply and more.

The official blog of Remi A.I, an Artificial Intelligence studio with offices in Sydney and San Francisco.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store