Parcel Delivery Optimization. Cutting Parcel Shipping Costs With Machine Learning

Parcel Delivery Optimization. Cutting Parcel Shipping Costs With Machine Learning

Dorota Owczarek - September 8, 2022 - updated on December 26, 2022

How do you optimize parcel delivery? It’s a question that logistics professionals have asked for years. The last mile of the delivery process is often the most expensive and time-consuming due to traffic congestion and the need to navigate complex urban landscapes. Until recently, there was no better way to optimize last-mile deliveries than simple route optimization. But now, with the help of custom deep learning solutions that optimize processes in real-time, it is possible to cut shipping costs and speed up delivery times. In this article, we will discuss how machine learning can be used to optimize parcel delivery and reduce costs for businesses of all sizes.

Last Mile Delivery Problem and Big Data

The last mile of parcel delivery is the most expensive and time-consuming part of the process, accounting for up to 28% of total shipping costs. In addition to being costly, the need for last-mile delivery is constantly growing and becoming more complex. This is partly due to the increasing number of eCommerce orders being placed, as well as the growing trend of same-day and next-day delivery. To meet these demands, supply chain companies must be able to adapt their operations and optimize their routes in real time.

Total delivery cost and the impact of last mile delivery cost

Total delivery cost and the impact of last mile delivery cost

One way to optimize parcel delivery is by using big data and advanced deep learning models. Big data can be used to identify patterns and trends that can be used to improve the efficiency of last-mile deliveries. Recent advancements in data collection technology and real-time data processing have motivated logistics companies to develop predictive systems that will enhance customer service and optimize logistic processes.

Biggest challenge for logistics providers in last mile

Biggest challenge for logistics providers in last mile (based on The Last Mile Logistics Whitepaper from EFT Eye For Transport)

In the past, route optimization and delivery scheduling have been manual processes that are time-consuming and often inaccurate. But with the advent of big data and machine learning, it is now possible to automate this process and make it more accurate than ever before. By collecting data on traffic patterns, weather conditions, construction projects, and other factors that can impact last-mile delivery, machine learning algorithms can predict the best route for each individual parcel.

What Is Parcel Delivery Optimization, and Why Does It Matter?

Parcel Delivery Optimization (PDO) describes the process of reducing shipping costs of parcel deliveries, streamlining planning, and speeding up delivery times through the use of big data and machine learning.

Parcel shipping and last-mile logistics pose different challenges than standard supply chain management.

How delivery scheduling software works - gathering data on orders and planning accordingly deliveries - from arrival at transportation hub, through delivery scheduling to loading and finding optimal routs for final delivery

How delivery scheduling software works - gathering data on orders and planning accordingly deliveries - from arrival at transportation hub, through delivery scheduling to loading and finding optimal routs for final delivery

Challenges Impacting Last-Mile Delivery Efficiency

The first thing to note is that when we talk about parcel shipping, we generally refer to the delivery process of small packages to medium-sized goods as single shipments. This service is available through postal systems, couriers, and smart parcel lockers. The parcels are usually light and can be easily handled by a single person. A typical package comes in the form of a carton box or an envelope with some cost-effective protective materials. Therefore, parcel deliveries cannot always be stacked on top of each other or placed on a pallet. Instead, it is best to put them in bins, bags, sacks, or shelves for transportation.

As customers, we often prefer home deliveries even though we cannot always be home. This poses the next challenge to parcel shipping which is predicting the best time to deliver a parcel to a specific address and minimizing the number of failed deliveries. If a delivery attempt is unsuccessful, the driver will have to come back another time, which wastes time and resources.

In some cases, drivers may have to travel long distances just to make one delivery. This increases fuel consumption and vehicle wear and tear, which in turn, raises shipping costs.

And finally, the last mile challenge for parcel shipping is traffic. Many businesses and domestic addresses are located in city centers with high traffic congestion. This can cause delays in delivery times and lead to higher fuel consumption as drivers idle in traffic.

Shippers can be small businesses or large companies that need to send items locally. However, some amount of parcel deliveries are still transported internationally. With the growing number of warehousing and availability of distribution centers, shipping companies are optimizing their processes to send goods from nearby warehouses rather than from the manufacturing site or central depot. This saves time and money, as well as reduces the carbon footprint.

All these challenges make it difficult to optimize parcel shipping routes and minimize costs. But with the help of machine learning, it is now possible to overcome these challenges and optimize parcel delivery. Machine learning can be used to predict traffic patterns, plan the best route for each individual shipment, and even choose the best time for delivery.

Courier Delivery Problem

The courier delivery problem is a classic vehicle routing problem with many real-world applications. In its most general form, the problem is to find the shortest or cheapest route that visits a given set of customers and returns to the depot. The problem can be further constrained by adding time windows, capacity limitations, and other conditions that must be met.

The courier delivery problem has been studied extensively in the operations research community, and many different graph algorithms have been proposed for solving it. However, most of these algorithms are good for planning and offline optimization but are not predictive in nature. They cannot self-optimize based on the forecasts they deliver. Recent advances in machine learning and data science have led to the development of new deep learning methods for solving the courier delivery problem that is able to optimize the delivery network in real time based on multiple predictions.

Last-Mile Delivery Optimization with Machine Learning - Use Cases

Now that we have seen how the courier delivery problem can be formulated as a machine learning problem, let’s look at some real-world examples of where delivery operations can be optimized with the right technology.

Route Optimization and Scheduling Deliveries

The first and most important use case for machine learning is a route optimization and scheduling deliveries. This involves using AI to plan optimal routes for each individual shipment, taking into account traffic conditions, transit times, location of the distribution center, dropoff time slots, car capacity, weather, and other valuable data points. The goal is to maximize the number of successful deliveries finalized by each courier and minimize the number of kilometers run (reduce cost and time of deliveries).

Machine learning models are an excellent solution for the route optimization problem that can achieve maximum efficiency because they can consider a large number of factors and constraints simultaneously and find the best route for each shipment in real-time. It can help couriers adjust their routes better and avoid congested areas resulting in time and cost-effectiveness of the last mile delivery operations. This is a significant improvement over traditional optimization methods and base route optimization software, which are often not able to consider all of the constraints at once or take into account real-time changes such as traffic conditions.

Predicting Delivery Times

Another critical use case for machine learning in parcel shipping optimization is predicting delivery times. It is crucial information for both customers and businesses, as it allows them to plan their day and ensure that someone will be available to receive the package. It also helps companies track their shipments and ensure they are delivered on time.

Shipment track & trace is one thing and accurate time windows for parcel dropoff are another, and you need it to satisfy growing customer expectations. Nowadays, customers require to have transparent last-mile visibility of their parcels. But as they order time-sensitive goods (e.g., grocery, frozen goods), there is an even higher need for an accurate delivery time prediction.

To be able to provide accurate predictions, machine learning models consider a number of factors, such as designated driver schedules, the distance of the delivery, the number of necessary visits to the depot, traffic conditions, and so on. This information can be gathered from GPS data, weather forecasts, social media data, and other sources to provide even more accurate information for end consumers and even greater visibility with updated ETA notifications that drive customer satisfaction levels.

Predicting Failed Deliveries

When we discuss predicting delivery times, it is essential to note that we assume someone will be at the delivery address to receive the parcel during specific time windows. But what if they are not home, or the business is already closed? Failed deliveries contribute a significant cost factor to the whole process, not only because a second attempt is needed but also because undelivered parcels need to be returned to the depot and eventually back to the shipper.

This is where machine learning can help by predicting failed deliveries before they happen. Models can consider several factors such as customer address, type of delivery, historical data on parcel deliveries, and so on. This information can be used to predict whether or not a particular delivery will fail and adjust delivery schedules during delivery planning.

Scheduling deliveries by predicting an optimal time of delivery for last-mile logistics

Scheduling deliveries by predicting an optimal time of delivery for last-mile logistics

Related case study: Implementing AI model to optimize routes and timelines of deliveries

A company from the logistics sector approached us to create a custom AI model that optimizes routes and the scheduling of deliveries.

Our challenge? The key challenge here was to prepare a dedicated AI-based system designed for carriers to optimize delivery time depending on the destination address. Thanks to the model, we managed to reduce failed and late delivery rates by 30%. Read our detailed case study of this project.

Dynamic Pricing for Last Mile Delivery

It is not uncommon for the end customer that the delivery cost is either nothing (most stores offer free delivery options) or very low (e.g., a few euros/dollars). This creates a big challenge for delivery companies and shippers since their operational costs are much higher.

Traditionally, parcel freight rates are either static (within a specific dimensional weight parameter range) or are calculated based on parcel size and weight, origin, and destination and contracted to the shipper. But as we all know, these factors do not give the complete picture of what it costs to deliver a particular parcel to a specific customer and how to construct parcel spend properly.

This is where machine learning can help by dynamically pricing each delivery based on a number of factors such as distance, size, supply and demand, and so on. This information can be used to calculate a real-time price for each delivery that reflects the actual cost of delivering the parcel. Dynamic pricing is when the price of a product or service fluctuates in real time based on demand. Dynamic pricing models are heavily applied in logistics and shipping. Thanks to this approach, transportation companies can evaluate their approach to pricing strategy and make it more efficient with a higher profit margin.

Large shippers, on the other hand, can use this information to get greater control of their delivery costs and carrier performance, forecast shipping budgets, help in carrier selection, and optimize their shipping operations. They can also utilize this insight to manage multiple carriers that provide parcel delivery services skillfully.

Improving Smart Parcel Lockers Systems

When we talk about last-mile delivery optimization, we cannot forget about smart parcel lockers systems. These are becoming increasingly popular within the parcel market as they offer several advantages for both customers and businesses.

For customers, parcel lockers provide a convenient way to receive their parcel shipments as they can pick them up at their convenience, which improves overall customer satisfaction.

For businesses, parcel lockers provide a competitive advantage to efficiently manage deliveries as parcels can be delivered to one location and then sorted and distributed to the individual parcel lockers. This approach reduces the number of missed deliveries and failed attempts as well as the overall cost of last-mile delivery.

Machine learning can help improve smart parcel locker systems in several ways:

  • smart delivery scheduling (optimize routes and schedules)
  • predicting pickup times
  • predicting locker filling rates
  • forecasting failed deliveries
  • optimizing for parcel consolidation
  • determining the location of new parcel lockers

All of these factors can be taken into account to improve the efficiency of parcel locker systems and make them even more convenient for customers and businesses alike. You can read more about AI for smart parcel lockers in our previous article.

Computer Vision Solutions for Parcel Delivery Optimization

One important aspect of parcel delivery optimization with machine learning is image processing. This involves techniques such as image segmentation, which separates an image into distinct regions or objects. In order to train a machine learning model to accurately identify these regions or objects, a large and diverse training dataset of images is necessary. By analyzing and interpreting these images, the model can learn to recognize and classify different elements in an image, allowing for more efficient and accurate parcel delivery processes.

Image processing plays a crucial role in the development of a land parcel identification system using machine learning. Through techniques such as image segmentation, the system can accurately identify and classify different features within an image of a parcel, such as the size and shape of the package, as well as any shipping labels or markings. By training the system with a large and diverse dataset of images, or “parcel deep training,” the model can continually improve its accuracy and efficiency in identifying parcels. This, in turn, can lead to more efficient shipping methods and smarter optimization of the parcel delivery process.

How Deep Learning Models Increase Customer Satisfaction and Benefit Carriers?

The goal of any business is to satisfy its customers and increase sales. In the case of parcel delivery, this means delivering the parcels on time (ever-tightening schedules) and in good condition. Recent advancements within the machine learning fields and the growing popularity of the deep neural networks approach make it possible for carriers and shippers to develop predictive models that can forecast different parcel delivery events. Unlike basic regression models that can be applied for simple forecasting, deep learning models can take into account many different data sources and make much more accurate predictions together with automating advanced decision-making.

On The Path Towards AI-Driven Last-Mile Delivery Strategies - Reoptimize Routes and Schedules in Real-Time

Last-mile delivery optimization is a complex problem that requires a holistic approach to automation and data analytics. It involves integration with all other logistics systems for end-to-end shipment visibility and, in many cases, able to process data in real-time. AI is already disturbing the logistics sector, though not in the same pace for all the players on the market.Current problems of logistics and transportation companies and the inability to move the needle of digitization are heavily rooted in the legacy systems and setups in use. Legacy ERP and TMS do not offer insights that would give a competitive edge or provide resilience in times of market disruptions that we now notice.

Many supply chain businesses are nowadays transforming their IT ecosystem and moving to cloud providers to leverage the data they collect and use it to improve their business. This is also the moment for proper planning of your future AI implementation. You need to approach the digitization process carefully to satisfy the needs of machine learning models like real-time data access and complete visibility of business operations.

Suppose you want to take full advantage of machine learning and other advanced technologies that can help you optimize your shipping operations. In that case, you need to move away from legacy systems in a planned manner. Only by doing so will you be able to improve your business efficiency, better serve your customers, and remain competitive in the market.

If you want to learn more about how machine learning can help optimize your last-mile delivery operations, don’t hesitate to contact us. Our team of AI experts is ready and waiting to partner with you to create a stellar AI solution that drives efficiency and improved customer experience.

About the author

Dorota Owczarek

Dorota Owczarek

AI Product Lead & Design Thinking Facilitator

Linkedin profile Twitter

With over ten years of professional experience in designing and developing software, Dorota is quick to recognize the best ways to serve users and stakeholders by shaping strategies and ensuring their execution by working closely with engineering and design teams.
She acts as a Product Leader, covering the ongoing AI agile development processes and operationalizing AI throughout the business.

Would you like to discuss AI opportunities in your business?

Let us know and Dorota will arrange a call with our experts.

Dorota Owczarek
Dorota Owczarek
AI Product Lead

Thanks for the message!

We'll do our best to get back to you
as soon as possible.

This article is a part of

AI in Logistics
51 articles

AI in Logistics

Artificial Intelligence is becoming an essential element of Logistics and Supply Chain Management, where it offers many benefits to companies willing to adopt emerging technologies. AI can change how companies operate by providing applications that streamline planning, procurement, manufacturing, warehousing, distribution, transportation, and sales.

Follow our article series to find out the applications of AI in logistics and how this tech benefits the whole supply chain operations.

check it out


Insights on practical AI applications just one click away

Sign up for our newsletter and don't miss out on the latest insights, trends and innovations from this sector.


Thanks for joining the newsletter

Check your inbox for the confirmation email & enjoy the read!

This site uses cookies for analytical purposes.

Accept Privacy Policy

In the interests of your safety and to implement the principle of lawful, reliable and transparent processing of your personal data when using our services, we developed this document called the Privacy Policy. This document regulates the processing and protection of Users’ personal data in connection with their use of the Website and has been prepared by Nexocode.

To ensure the protection of Users' personal data, Nexocode applies appropriate organizational and technical solutions to prevent privacy breaches. Nexocode implements measures to ensure security at the level which ensures compliance with applicable Polish and European laws such as:

  1. Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation) (published in the Official Journal of the European Union L 119, p 1); Act of 10 May 2018 on personal data protection (published in the Journal of Laws of 2018, item 1000);
  2. Act of 18 July 2002 on providing services by electronic means;
  3. Telecommunications Law of 16 July 2004.

The Website is secured by the SSL protocol, which provides secure data transmission on the Internet.

1. Definitions

  1. User – a person that uses the Website, i.e. a natural person with full legal capacity, a legal person, or an organizational unit which is not a legal person to which specific provisions grant legal capacity.
  2. Nexocode – NEXOCODE sp. z o.o. with its registered office in Kraków, ul. Wadowicka 7, 30-347 Kraków, entered into the Register of Entrepreneurs of the National Court Register kept by the District Court for Kraków-Śródmieście in Kraków, 11th Commercial Department of the National Court Register, under the KRS number: 0000686992, NIP: 6762533324.
  3. Website – website run by Nexocode, at the URL: whose content is available to authorized persons.
  4. Cookies – small files saved by the server on the User's computer, which the server can read when when the website is accessed from the computer.
  5. SSL protocol – a special standard for transmitting data on the Internet which unlike ordinary methods of data transmission encrypts data transmission.
  6. System log – the information that the User's computer transmits to the server which may contain various data (e.g. the user’s IP number), allowing to determine the approximate location where the connection came from.
  7. IP address – individual number which is usually assigned to every computer connected to the Internet. The IP number can be permanently associated with the computer (static) or assigned to a given connection (dynamic).
  8. GDPR – Regulation 2016/679 of the European Parliament and of the Council of 27 April 2016 on the protection of individuals regarding the processing of personal data and onthe free transmission of such data, repealing Directive 95/46 / EC (General Data Protection Regulation).
  9. Personal data – information about an identified or identifiable natural person ("data subject"). An identifiable natural person is a person who can be directly or indirectly identified, in particular on the basis of identifiers such as name, identification number, location data, online identifiers or one or more specific factors determining the physical, physiological, genetic, mental, economic, cultural or social identity of a natural person.
  10. Processing – any operations performed on personal data, such as collecting, recording, storing, developing, modifying, sharing, and deleting, especially when performed in IT systems.

2. Cookies

The Website is secured by the SSL protocol, which provides secure data transmission on the Internet. The Website, in accordance with art. 173 of the Telecommunications Act of 16 July 2004 of the Republic of Poland, uses Cookies, i.e. data, in particular text files, stored on the User's end device.
Cookies are used to:

  1. improve user experience and facilitate navigation on the site;
  2. help to identify returning Users who access the website using the device on which Cookies were saved;
  3. creating statistics which help to understand how the Users use websites, which allows to improve their structure and content;
  4. adjusting the content of the Website pages to specific User’s preferences and optimizing the websites website experience to the each User's individual needs.

Cookies usually contain the name of the website from which they originate, their storage time on the end device and a unique number. On our Website, we use the following types of Cookies:

  • "Session" – cookie files stored on the User's end device until the Uses logs out, leaves the website or turns off the web browser;
  • "Persistent" – cookie files stored on the User's end device for the time specified in the Cookie file parameters or until they are deleted by the User;
  • "Performance" – cookies used specifically for gathering data on how visitors use a website to measure the performance of a website;
  • "Strictly necessary" – essential for browsing the website and using its features, such as accessing secure areas of the site;
  • "Functional" – cookies enabling remembering the settings selected by the User and personalizing the User interface;
  • "First-party" – cookies stored by the Website;
  • "Third-party" – cookies derived from a website other than the Website;
  • "Facebook cookies" – You should read Facebook cookies policy:
  • "Other Google cookies" – Refer to Google cookie policy:

3. How System Logs work on the Website

User's activity on the Website, including the User’s Personal Data, is recorded in System Logs. The information collected in the Logs is processed primarily for purposes related to the provision of services, i.e. for the purposes of:

  • analytics – to improve the quality of services provided by us as part of the Website and adapt its functionalities to the needs of the Users. The legal basis for processing in this case is the legitimate interest of Nexocode consisting in analyzing Users' activities and their preferences;
  • fraud detection, identification and countering threats to stability and correct operation of the Website.

4. Cookie mechanism on the Website

Our site uses basic cookies that facilitate the use of its resources. Cookies contain useful information and are stored on the User's computer – our server can read them when connecting to this computer again. Most web browsers allow cookies to be stored on the User's end device by default. Each User can change their Cookie settings in the web browser settings menu: Google ChromeOpen the menu (click the three-dot icon in the upper right corner), Settings > Advanced. In the "Privacy and security" section, click the Content Settings button. In the "Cookies and site date" section you can change the following Cookie settings:

  • Deleting cookies,
  • Blocking cookies by default,
  • Default permission for cookies,
  • Saving Cookies and website data by default and clearing them when the browser is closed,
  • Specifying exceptions for Cookies for specific websites or domains

Internet Explorer 6.0 and 7.0
From the browser menu (upper right corner): Tools > Internet Options > Privacy, click the Sites button. Use the slider to set the desired level, confirm the change with the OK button.

Mozilla Firefox
browser menu: Tools > Options > Privacy and security. Activate the “Custom” field. From there, you can check a relevant field to decide whether or not to accept cookies.

Open the browser’s settings menu: Go to the Advanced section > Site Settings > Cookies and site data. From there, adjust the setting: Allow sites to save and read cookie data

In the Safari drop-down menu, select Preferences and click the Security icon.From there, select the desired security level in the "Accept cookies" area.

Disabling Cookies in your browser does not deprive you of access to the resources of the Website. Web browsers, by default, allow storing Cookies on the User's end device. Website Users can freely adjust cookie settings. The web browser allows you to delete cookies. It is also possible to automatically block cookies. Detailed information on this subject is provided in the help or documentation of the specific web browser used by the User. The User can decide not to receive Cookies by changing browser settings. However, disabling Cookies necessary for authentication, security or remembering User preferences may impact user experience, or even make the Website unusable.

5. Additional information

External links may be placed on the Website enabling Users to directly reach other website. Also, while using the Website, cookies may also be placed on the User’s device from other entities, in particular from third parties such as Google, in order to enable the use the functionalities of the Website integrated with these third parties. Each of such providers sets out the rules for the use of cookies in their privacy policy, so for security reasons we recommend that you read the privacy policy document before using these pages. We reserve the right to change this privacy policy at any time by publishing an updated version on our Website. After making the change, the privacy policy will be published on the page with a new date. For more information on the conditions of providing services, in particular the rules of using the Website, contracting, as well as the conditions of accessing content and using the Website, please refer to the the Website’s Terms and Conditions.

Nexocode Team