AI-Based Demand Forecasting Case Study: Parcel Delivery

Case study details

Demand forecasting plays a crucial role in the efficient and effective operation of parcel delivery services. As the e-commerce industry continues to grow rapidly, fueled by shifting consumer behaviors and increased online shopping, parcel delivery companies face mounting pressure to accurately predict and meet customer demands. To stay competitive in this dynamic landscape, businesses must optimize their supply chain processes, streamline operations, and ensure timely delivery of packages.

Inpost SA is specialized in the provision of postal services. The group offers a network of parcel boxes, self-service parcel delivery, and collection points. It also provides services for collecting, routing, and delivering mail, dispatch, and delivery of parcels, express transport, and delivery.

With 4.6 billion zlotych in revenue, the company operates in three segments. APM segment is focused on the delivery of parcels to automated parcel machines. The to-Door segment includes the delivery of packages using door-to-door couriers. The international segment comprises the APM business, which delivers parcels to automatic parcel machines in the United Kingdom and Italy.

InPost, a Polish company developing postal services such as Paczkomaty network, struggled with problems meeting the demand and appropriate pricing strategy for individual sellers, which caused a painful and costly customer churn. The company couldn’t afford it as the competition in post-pandemic reality was just waiting for it.

Now, we will explore the significance of demand forecasting in the context of parcel delivery, focusing on InPost’s demand forecasting case study as an example. By examining real-world examples and best practices, we can gain valuable insights into effective practices that can be applied in the evolving landscape of the parcel delivery industry.

Challenge

In this demand forecasting case study, InPost experienced difficulties with demand forecast accuracy during the pandemic and needed to implement innovative, AI-based solutions that would improve its entire operational performance. The stake was high, as the competitors were intensely aware that the customers’ behaviors coined during the COVID-19 outbreak would stay long after the situation calmed down, and the companies that proved themselves under this pressure would win customer loyalty.

Main challenges were:

No daily sales forecast per each customer
Forecasts are calculated for each cohort in the same way – in excel and just based on historical dynamics for a particular cohort
No offer to private sellers at the best prices

Approach

Addepto team built and implemented a Machine Learning prediction model based on the historical data, macroeconomics factors, and other 3rd party data with rare events such as Covid included. The model predicts demand and – based on it – calculates the optimal prices for private sellers.

The solution automatically forecasts monthly and weekly sales by individual products and companies, based on the demand and supply analysis, an optimal price is chosen.

Unified Platform for Managing Supply Chain
Predictive AI Modules

Goal

The company’s goal was to reduce the challenges related to managing the logistics side of operations, avoid disruption, get the overall view of the factors that influence the supply chain, and – finally – reduce its costs and lower the gas emissions.

Outcome

This demand forecasting case study shows how ML-based solutions enabled the company to improve accuracy in demand prediction, they could optimize the usage of the resources and prices and reduce operational costs while keeping up with their client’s demands.

Challenge

Retailers who rely on approximated demand disappoint customers, bleed profits, and loose market share to more tuned-in competitors

However, tackling the challenges in demand forecasting is not easy, especially while using spreadsheets and intuition. The number of factors that influence demands is simply overwhelming. Moreover, they are in constant flux, making it extremely difficult to grasp them from a bird’s view.

Basing demand forecasts on what happened in the year prior can be highly inaccurate

InPost experienced it during the pandemic. The demand for the company’s services skyrocketed then and – there would be no exaggeration to say – completely changed the game.

Keep up with the clients' needs

To keep up with the clients’ needs, the company needed to implement innovative, AI-based solutions able to improve the accuracy of the provided predictions and – in general – the entire operational performance.

Competitors

The stake was high, as the competitors were intensely aware that the customers’ behaviors coined during the COVID-19 outbreak would stay long after the situation calmed down. The companies that proved themselves under this pressure will win customer loyalty.

Approach

Machine learning to drive predictive analytics and demand forecasting

Machine learning ensures higher accuracy and veracity of the forecasts. Human-driven processes are prone to errors, especially if there are loads of data to be checked, and machine learning doesn’t have such a downside.

During project development, our team worked on:

Machine learning provides a forecast with no-fault. Moreover, it allows the automation of the updates, which simplifies and speeds up the decision-making processes.
Based on the above, Addepto team built and implemented a Machine Learning prediction model that considers historical data, macroeconomics factors, and other 3rd party data (with rare events such as Covid included) to provide forecasts.
The ML model predicts demand and - based on it - calculates the optimal prices for private sellers.
The solution automatically forecasts monthly and weekly sales by individual products and companies, based on the demand and supply analysis, an optimal price is chosen.

Goal

Improving overall logistics processes to avoid any disruption

Company managers needed to get data-grounded business insights to make decisions based on real factors, not the assumptions; processes required to be automated, scalable, repetitive, and bulletproof, so manual labor had to be reduced to a minimum.

Setup and configuration of new ML platform based on cloud infrastructure

Support large amounts of data

Include into the ML processes the Covid-19 effect

Let our clients do the talking

Thanks to the efforts of the Addepto team, the company can now craft more effective business strategies using the data that the solution provides them.

Przemysław Piekarz Sales Analysis Manager – InPost

Outcome

Demand forecasting case study - outcome

ML-based solutions enable companies to forecast demands based on unique target groups’ shopping behaviors while understanding seasonal trends. Based on the data-grounded forecasts, ML allows managers to avoid costly delays while simultaneously reducing the amount of work they need to do. This demand forecasting case study shows how with improved accuracy in demand prediction, they could optimize the usage of resources and prices and reduce operational costs while keeping up with their client’s demands.

Data-grounded demand forecast
Automated price optimization
Analyzing multisource data at once

Before

Price optimization based on intuition
Slow and vulnerable predictive processes
High percentage of delays in delivery
High customer churn

After

Automated data processing
Data-grounded dement predictive analytics
Automated price optimization
Reducing customer churn
Increased customer experience

About Addepto

Addepto, an AI consultancy company, supports logistic businesses in increasing their data's effective usage with various AI-related technologies, such as machine and deep learning.

We help to create data-driven services to impact customer satisfaction, increase revenue and accelerate internal processes.

About us