Optimizing supply chain costs and increasing the customer acquisition rate with ML

If you’re reluctant to adopt advanced digital solutions like artificial intelligence (AI) and machine learning (ML) algorithms to analyze your supply chain data, you may miss lots of insights that can potentially improve your business operations. Algorithms are part of every digitization step you take during your supply chain journey. Surprisingly, many companies are still at the early stages of getting familiar with the benefits of machine learning in the supply chain.

According to an APQC survey conducted among 1,000 global supply chain business representatives, 46 percent of respondents still don’t use any form of AI. Now is just the right time to give advanced supply chain technology solutions a chance and get ahead of the game.

Supply chain optimization may come in multiple forms and through different techniques and technologies. One form is robotic process automation (RPA) solutions for automating routine back-office tasks.

In this article, we’ll justify investment in software with ML algorithms, as such software can help you plan your expenses more accurately as well as attract new customers. Let’s dive straight into the essence of machine learning.

Read also: Our expertise in developing custom ML and AI software

ML algorithms as a solution to specific business problems

Each algorithm aims at solving a given business problem. Thus, you need to accurately define your business problems and goals to ensure that ML algorithms operate with the right data to add actual value to your organization.

ML algorithms can process any kind of internal and external data including structured, unstructured, and binary data. When processing big data, ML contributes to data science. Data science focuses on collecting, evaluating, and analyzing data to extract knowledge and value. If ML is combined with such solutions as natural language processing (NLP), expert systems, and robotics, it contributes to AI, which aims at creating intelligent and independent systems that can assist humans.

Machine learning in the supply chain can discover a vast number of dependencies between data elements like market and demand trends and without needing to explain them. Thus, ML opens wider options for data analysis by building algorithmic dependencies between data sets from IoT devices, social networks, media, or weather forecasts. Learn how we've embedded ML based on Amazon SageMaker for our big data analytics solution for a 3PL company.

Further on, we’ll have a look at the best practices for solving two supply chain problems by means of ML algorithms: 

  • Optimizing supply chain costs 
  • Acquiring new customers

We’ll define the pros and cons of ready-made ML solutions, comparing them to software developed from scratch.

Read also: BI and advanced analytics solutions for supply chain data analysis

Optimizing supply chain costs with ready-made and custom ML solutions

According to the Supply Chain Resilience Report 2021 by the Business Continuity Institute, the COVID-19 pandemic is currently a major supply chain disruptor. However, instances of cyber attacks, data breaches, and catastrophic weather conditions have also increased over the course of 2021, causing significant threats to supply chain profits.

In line with that, 16.7 percent of 173 supply chain business representatives from 62 countries surveyed by BCI reported a severe loss of revenue in 2021.

Such external stimuli as the pandemic and unexpected weather conditions sometimes are the only motivations for supply chain companies to invest in technological innovations as a last-ditch attempt to save their businesses from failure. However, it’s much wiser for companies to be proactive and not wait for a certain turn of events or financial troubles to push them to adopt next-gen technology.

Supply chain costs are an extensive topic for discussion that includes many variables and covers all supply chain areas. For this section, we’ve scoped out a few supply chain management processes that require regular operating expenses (OpEx):

  • Supply and demand planning 
  • Transportation
  • Warehousing

To achieve tangible supply chain cost optimization, you should focus on optimizing the above-mentioned supply chain management processes first. ML solutions have great potential to assist you in this endeavor.

Read also: How we build custom data science solutions

Supply and demand planning

Manual supply and demand management isn’t a viable approach anymore. Even traditional deterministic supply and demand forecasting methods aren’t capable of providing you with the most accurate and relevant insights.

Traditional forecasting systems analyze historical data and don’t take into account real-time changes in supply and demand. On the contrary, ML-based forecasting systems have a more probabilistic or stochastic approach, providing data models with a wider range of forecast options. And apart from historical data, such systems can process data from a much wider pool of sources.

We can outline the following key reasons why supply chain businesses adopt advanced supply and demand forecasting systems:

key drivers for choosing ml demand forecasting software
Real-life examples of ML adoption

British retailer WHSmith needed a smart demand forecasting solution that could analyze footfall and demand fluctuations due to promotions and holiday seasons. An ML forecasting solution the company adopted helped it use even airport traffic data to better predict customer demand. As a result, the company achieved a considerable reduction in product spoilage and an increase in product availability.

Another example is Polaris, a manufacturing company that implemented an advanced supply and demand forecasting system. The company managed to reduce inventory levels by 15 percent and improve service levels by 10 percent.

Ready-made ML solutions for supply and demand planning

There are many ready-made ML solutions available for supply and demand forecasting. Primarily, they differ in functionality, target industries, and customizability. We’ll take a look at the common advantages and shortcomings of off-the-shelf ML solutions for supply and demand planning based on real-world examples.

Symphony RetailAI is a cloud-based AI/ML-driven platform that forecasts demand for retailers and fast moving consumer goods (FMCG) companies. This platform offers prompt solutions for the intelligent allocation of products on store shelves based on customer demand.

o9 Solutions is an AI/ML-powered platform for supply and demand planning for industries including retail, FMCG, industrial manufacturing, and automotive. o9 Solutions offer real-time visibility into supply and demand changes, supply and demand risk mitigation capabilities, as well as goods replenishment planning and inventory allocation planning.

Pros of ready-made ML solutions according to customer reviews on Gartner Peer Insights:

  • A thought-out feature set and wide customization capabilities for the industries and business models they cover
  • Ease of use and platform adaptation for non-technical users
  • Onboarding and knowledge sharing for new and existing customers

Cons of ready-made ML solutions according to customer reviews on Gartner Peer Insights:

  • Difficulty finding a solution for niche supply chain industries like agriculture
  • Necessity to compromise your business goals with a one-size-fits-all functionality set
  • Limited number of sources for extracting data to build advanced ML algorithms
  • Difficulty integrating with other software due to the limited set of APIs
  • Need for an in-house team of specialists to fix possible issues with the purchased software to avoid the risk of sharing confidential information with the customer support team

Judging by the above findings, we can conclude that if your supply chain company works primarily with B2C clients, you have a high chance of finding a simple ready-made solution to forecast supply and demand. However, if you work with B2B clients and deal with non-typical consumer products, you should consider custom ML solutions.

Custom ML solution for supply and demand planning

Let’s have a look now at how a custom ML solution for supply and demand forecasting could look like.

ML proves helpful in building effective supply chain optimization models for demand and supply forecasting.

For building a forecasting ML model, we can use time series data. Thus, to build predictions, an ML algorithmic row can combine time series along with the other available data using unsupervised machine learning or supervised machine learning methods.

First, we can analyze historical data on supply and demand with regard to seasonal trends and anomalies. Then, we can collect additional data like the following:

  • Market indexes, currency pairs, and commodity prices. This data is collected to trace dynamics and risks in demand and supply volatility.
  • Weather conditions. Weather data can be imported directly from weather stations. Such data is of primary importance for forecasting agricultural supply and demand.
  • News and social media. Such data can come from news websites or Twitter accounts for analyzing demand trends in a certain industry.

Thus, custom ML software for the supply chain can take as an input all kinds of data to generate accurate forecasts.

ml algorithm building process for demand forecasting

When weighing options, remember that when you hire a team for custom ML development, you’ll be able to consult on the types of data to choose for building ML algorithms that completely cover your needs. When purchasing off-the-shelf software, it may require much time and effort to differentiate between those solutions that fit and don’t fit your business model.

Optimization of transportation services with ML can also ensure significant financial benefits for your company. Let’s have a closer look at how algorithms can improve your transportation services.


Transportation is another supply chain area that often requires large investments. And here ML also comes in handy, as the right ML algorithms can optimize the following aspects of freight transportation:

freight transportation use cases for ml adoption

For instance, such giants as UPS, FedEx, and DHL use lots of ML techniques to optimize transportation services and reduce transportation risks. ML-based systems help these service providers get real-time insights into supply chain performance based on traffic and weather conditions. ML algorithms help dispatchers predict delivery delays and modify routes on the go.

Ready-made ML solutions for transportation

Out-of-the-box ML solutions for optimizing freight transportation come in different forms: as part of a solution suite that companies like SAS and Oracle offer or as smaller and more specific solutions from independent software vendors. We’ll have a look at two examples of more specific solutions.

Vector is an AI-driven platform for streamlining freight forwarding. It uses ML algorithms to analyze documentation, extract valuable information fast, and unburden employees from manual delivery management. Vector helps logistics managers simplify freight coordination, avoid shipment delays, and reduce the number of spoiled goods.

FarEye is last-mile delivery optimization software based on ML algorithms for building optimal last-mile delivery routes. This system also provides real-time delivery tracking, predictive alerts, and notifications in case disruptions occur.

Pros of ready-made ML solutions according to customer reviews on Gartner Peer Insights:

  • Proper integration capabilities with TMSs and warehouse management systems (WMSs)
  • Well-established core features like freight documentation management and last-mile delivery route building

Cons of ready-made ML solutions according to customer reviews on Gartner Peer Insights:

  • Takes a long time for support teams to get familiar with each client’s specific product to solve issues, and even then the solution may be inefficient
  • Frequent system upgrades are a potential threat to the established workflow
  • Non-typical customization requests have to go through the core development team and may not be approved

In ready-made solutions, ML algorithms usually solve particular business problems. Thus, it may be difficult for you to find an all-encompassing ready-made machine learning platform with a suite of algorithms to resolve all of your transportation issues. However, if you aim at solving only a few specific problems like last-mile routing or automation of freight documentation, you can consider ready-made software.

Custom ML solution for transportation

Let’s overview how a custom ML solution for transportation services can benefit your company and optimize the speed of the supply chain as well as your supply chain costs. We’ll overview possible solutions for route optimization and choosing a carrier.

Solving the route optimization and load planning problem

Given lots of delivery destinations and many orders, building optimal dispatch routes poses a complex combinatorial optimization problem (COP). To solve this problem, we have to build specific and fast algorithms. For this purpose, we can choose a set of stochastic local search (SLS) algorithms. Stochastic search is a probabilistic mechanism for defining optimal delivery routes.

Within a SaaS transportation management system (TMS) we built for one of our clients, we also implemented a route optimization algorithm. Logistics managers only need to add required delivery data to the system to train the algorithm.

Plus, logistics managers can set prioritized criteria for any routing algorithm with sliders for delivery price and speed. In this way, the algorithm can build an optimal route for each delivery. For instance, for one customer, the delivery can take longer but cost less. For another, it can be faster but more expensive. Thanks to our solution, our client can reduce the time spent planning a single route from two to six hours to 30 minutes.

route optimization algorithm by yalantis

Load planning is a tough transportation issue too. We helped a Mexican 4PL company solve the load planning problem by building a custom load planning system. The system can identify production plan feasibility and accurately calculate the number of containers necessary for delivering raw materials. We built a solution based on the Google OR-Tools suite of algorithms, which we smoothly integrated with the client’s databases to ensure an uninterrupted load planning process.

Solving the problem of choosing the right carrier

An ML solution can streamline the process of choosing a carrier based on rates and reputation. Those can be calculated using a recommendation system trained on feedback from the supplier and receiver upon delivery completion or failure.

At the core of the recommendation system engine lies collaborative filtering algorithm. Thanks to this algorithm, system users can use pre-defined formats to send feedback reactions like from one to five stars or one to ten points. Collaborative filtering algorithm provides each carrier’s rating based on recommendation system users’ responses (on the sender’s and receiver’s sides). Thus, a recommendation system can calculate ratings for all carriers the supply chain company cooperates with to offer a final list of recommended carriers.

recommendation system algorithm

When a supplier rates the carrier’s service quality, this rating can be also used for training a special system that will be able to predict the chances of successful delivery for that carrier.

One more benefit in favor of a custom ML solution is that if the algorithm doesn’t work as planned, it can be fixed in a timely manner according to new criteria. This is possible if you work with a specialist with deep ML expertise. In case we don’t have enough feedback about carriers, we can implement factorization machines (FM), a supervised learning algorithm, to reduce the problem dimension and improve calculations. FM can estimate parameters when only sparse data is available as well as scale to fit large data sets. 

A similar solution can be applied for choosing suppliers or assigning drivers to deliveries.

Let’s move on to another supply chain management process worth optimizing with ML algorithms to reduce supply chain operating expenses.

Read also: Choosing transport management software: types of TMSs and must-have functionality

Warehousing and inventory management

We’ve reached warehousing, a third supply chain management process in our ML optimization list. Key areas in which ML algorithms can prove beneficial for warehouse management are:

warehousing use cases for ml adoption

With automated algorithms, a warehouse can run like clockwork. Warehouse operators have a system to easily locate orders, which are distributed according to customer demand. The flow of inbound and outbound goods is streamlined, and forklifts don’t move around the warehouse empty.

For instance, Continental Barum, a Czech manufacturer of rubber tires, wanted to minimize inter-warehouse movements for picking and allocating manufactured products. With the help of a product distribution algorithm that optimizes product storage on warehouse shelves, the company managed to eliminate redundant and costly movements in the warehouse.

Read also: Space optimization in storage and transport facilities amid the e-commerce boom

Ready-made ML solutions for warehousing and inventory management

The most efficient way to implement warehousing ML algorithms is to make them part of a WMS. We’ve picked two AI-powered WMSs for our analysis of off-the-shelf ML solutions. 

Focus is an advanced AI-driven enterprise resource planning system with a separate WMS module (Focus WMS) that includes machine learning capabilities. One of the ML-based features that Focus WMS offers is a solution for tracking inventory by aisle and shelf. Focus WMS promises to reduce order fulfillment times, optimize operating costs, increase inventory allocation accuracy, and improve customer satisfaction.

Consafe Logistics is an innovative AI solution for warehousing and inventory management. Some of the features the system offers include:

  • streamlined storage with goods allocation based on order history
  • building efficient routes to pick goods
  • real-time warehouse performance tracking

Pros of ready-made ML solutions according to customer reviews on Gartner Peer Insights:

  • No need for significant investments to get started
  • Easy and clear functionality set
  • No-code customization opportunities
  • Custom integrations using RESTful APIs
  • Continuous improvement with frequent releases of new features

Cons of ready-made ML solutions according to customer reviews on Gartner Peer Insights:

  • Additional AI solutions apart from ML algorithms that may be redundant
  • Limited integration and customization capabilities (e.g. inability to connect to certain IoT devices)
  • Unexpected interruptions in system functioning
  • Some features are too advanced and unnecessary

If a WMS or yard management system (YMS) is already part of your supply chain management flow, you can consider developing a custom set of ML algorithms to simply enhance your existing software capabilities and functionality.

Read also: How a yard management system can boost yard and warehouse efficiency

Custom ML solution for warehousing and inventory management

Whereas transportation services and supply and demand planning can be optimized using a roughly similar pattern, warehousing is a different story. A custom solution here is even more critical than in the previous cases, as any disruptions or limitations in the warehouse significantly impact the whole supply chain. Thus, the stakes are higher when implementing a warehouse management solution than in the case of solutions for previously discussed supply chain processes.

Plus, custom development offers you a much wider playground for experiments than purchasing ready-made software. As you have no limits on the number and difficulty of ML algorithms you can build to optimize your warehouse operations.

For instance, combinatorial optimization algorithms help optimize the time spent on warehouse operations. With the help of such ML algorithms, optimal warehouse loading schedules can be built to increase the product flow and improve the whole supply chain business cycle.

Warehouse space management can also become a bottleneck. If you use space in your warehouses inefficiently, you may often either lack space or have too many empty shelves. In both cases, you may spend extra money on either leasing or buying additional warehouses or maintaining empty space. Algorithmic solutions for allocating goods in warehouses can help you save money and use existing storage space in an optimal way. Get familiar with our recent solution to the space optimization problem.

In the next section, we’ll drill down a bit into machine learning applications for attracting more customers to your company.

Retaining existing customers and increasing customer acquisition with ML

In a customer-centric organization, streamlining all of the above-mentioned processes may be enough to retain existing customers and gain new ones. A 2021 Oracle survey among 1,000 US consumers revealed a surprising fact: 78 percent of respondents are more likely to buy from a company that implements AI techniques for efficient supply chain management.

Thus, even the mere fact that you implement AI/ML solutions in your supply chain may increase your company’s value in your customers’ eyes. But imagine what could happen if you started using the potential of ML to target your customers directly? Such a decision could be extremely beneficial.

Possible ML use cases that can improve your relationship with customers:

ml use cases for customer acquisition

ML algorithms can potentially help you gain deeper insights into your customers’ preferences, habits, and behavior patterns. Many ready-made AI platforms provide additional AI-enhanced features specifically for your clients, such as chatbots that simulate real communication with customer agents.

Custom ML algorithms for targeting customers are most advantageous if you’re looking for a fast and industry-specific solution. Since each supply chain company has a unique client base, you can build individual algorithms that work only for your specific target audience.

By implementing either custom or ready-made ML algorithms, you not only optimize supply chain costs and attract new customers but also save your company time for handling important strategic issues. Consequently, your business can grow and improve much faster than it could without algorithms in place.

If custom development sounds promising to you, our AI/ML team is here to provide you with a thorough consultation on which algorithms to choose to gain the most value for your supply chain organization as well as to build those algorithms. If custom development isn’t exactly what you’re looking for, we can conduct market research on ready-made AI/ML solutions for your specific business problem so you can make the most balanced decision.

Want to implement ML algorithms that are beneficial for your business?

Get some help from Yalantis AI experts.

Rate this article
5.0/ 5.0 based on 1,153 reviews

We use cookies to personalize our services and improve your experience on this website and its subdomains. We may use certain personal data for analytics and marketing purposes.