Get Started

Scaling Supply Chain AI to Offer Predictive Shipment Visibility

Use Cases & Projects, Dataiku Product, Scaling AI Ayoub Benyahya, Sam Medary

This blog is a guest post from our friends at Shippeo. They’re a global leader and European specialist in real-time transportation visibility, helping major shippers and logistics service providers leverage transportation to deliver exceptional customer service and achieve operational excellence. 

As we are all now painfully aware, supply chains are being squeezed by many challenges and customer expectations are constantly on the rise. Visibility of goods in transit is becoming a huge focus for organizations, exacerbated by what feels like decades of never-ending global challenges.

Within supply chains, some of the challenges businesses are seeing include: 

  • Increased complexity and fragmentation
  • Rising transport costs
  • Visibility blind spots (meaning shipment status is unknown) which lead to many process inefficiencies

These challenges have significant impacts on business performance, causing issues that vary from sector to sector, including things like inventory shortages, late penalties, lost revenue, and impacts on brand reputation.

Thankfully, the ability to track an order as it travels throughout global supply chains in real time is becoming increasingly accessible with the proliferation of faster and more reliable connectivity, cost-effective IoT devices, and more compatible systems and software platforms.

Shippeo’s platform provides end-to-end visibility of shipments as they flow through supply chains. Just like B2C customers want better service and transparency for deliveries, whether it’s for your Nespresso capsule delivery or for your Uber ride, B2B customers want the same real-time traceability for their deliveries.

However, the question that supply chain management and transport teams ask themselves is shifting from “Where is my shipment?” to “When will it arrive at the next destination?” and “Are there any risks of a delay?” As such, the ability to predict shipment ETAs has become increasingly critical.

Of course, ETA accuracy and reliability depends heavily on the quality of the data captured and the sophistication of the technology and computational methodology used. With Dataiku, Shippeo is able to innovate like never before, with improved capabilities around data ingestion, cleaning and transformation, ML model design and training, and model lifecycle management.

Pushing the Boundaries of Technology & Innovation

At Shippeo, we strive to offer world-class ETA predictions for new and existing customers alike. As such, we need to continually retrain our predictive models to ensure we have a good level of performance for our customers and avoid model drift.

From a technology point of view, we struggled with maintaining reproducible pipelines to create and manage large-scale training and prediction data. This means that when we want to retrain or improve our ETA predictive models, we need to review these pipelines and, in some cases, recreate some of them from scratch. During this process, the team loses valuable time that could instead be invested in our research and innovation to continue creating value for our customers. 

We also find it a challenge sometimes to collaborate on the same AI/ML products, since:

  • Data scientists have their own notebooks and training sets.
  • Data engineers maintain different data pipelines.
  • Data scientists and engineers need recurrent alignments, specifications, and thus more time is needed to progress together.

Below is our ETA architecture before having the Dataiku platform among our stack:

before Dataiku

Shippeo ETA Architecture before Dataiku

Dataiku Was a Natural Choice to Bring the Different Teams Together

Shippeo uses Dataiku to industrialize the way we design, train, deploy, and monitor our ETA machine learning models. We determined that addressing this with an MLOps strategy would help us to acquire a competitive advantage.

Below is the new architecture with the Dataiku platform:

architecture with Dataiku

Shippeo ETA Architecture with Dataiku

The Dataiku platform has strong capabilities for data ingestion, cleaning and transformation, ML model design and training, and model lifecycle management. The main advantages we observed upon using the platform are:

1. Multi-Project Management

We now have an environment in which we can create and maintain multiple AI/ML projects, as well as access a plugin store in Dataiku to help us deploy new monitoring tools or add new features, all without custom development.

2. Collaboration and Versioning Enabler

Dataiku facilitates collaboration between data team members to design and train ETA ML models with a versioned project.

3. Reduce Time to Market and Make Iterations Faster

We are able to test our assumptions, try new approaches faster, and assess the importance and impact of additional features on our prediction performances. We can also run recurring training of our ETA models so that we can have continuous learning based on existing and new customer transport flows.

Predictive Shipment Visibility Now

Our new architecture and Dataiku capabilities give us greater confidence that the ETA we offer customers is best-in-class. In fact, these advancements were a key factor in allowing us to recently enhance our ETA model with a giant 32% accuracy improvement.

We are also looking to diversify our offering, by increasing our focus on generating new forms of value by leveraging the vast amounts of supply chain and ecosystem data captured throughout our network. We think we can offer enormous value in this space but only the best tools, infrastructure and expertise. In other words, only with partners like Dataiku. 

You May Also Like

Simplify Anti-Money Laundering (AML) Alerts With Data-Backed Triage

Read More

Key Steps Involved in the Machine Learning Process: A Primer

Read More

Get Rid of Friction & Trigger an AI Chain Reaction With MLOps

Read More