How we scaled supply chain AI to improve predictive visibility of shipments

We all know that the last few years have been tough for many global supply chains. At the same time, customer expectations around shipment delivery times and status updates only climb higher. For this reason, along with decades of global disruption events hammering our supply chains, has seen shipment visibility become an increasingly important topic for organizations.

‍

Common supply chain challenges include:

- More fragmentation and complexity

- Increasing costs of transport

- Process inefficiencies, caused by blind spots, where shipment statuses are unknown

‍

The impacts of these challenges on business performance are significant, potentially causing inventory shortages, late penalties, lost revenue, and damage to brand reputation.

‍

Fortunately, it’s now more accessible than ever to track orders as they travel throughout global supply chains, in real time, thanks to the availability of faster and more reliable connectivity, affordable IoT devices, and more compatible systems and software platforms.

‍

Our platform offers end-to-end shipment visibility at any point across the supply chain, so that enterprise supply chain management can have the same shipment delivery experience for their supply chain shipments as they’ve come to expect at home from online retailers, offering better service, transparency and real-time traceability.

‍

Although knowing where a shipment is brings a raft of benefits, the question that supply chain managers and transport teams ask themselves is shifting from ‘Where is my shipment?’ to ‘When will my shipment arrive?’ and ‘Will it be on time?’. For this reason, being able to predict ETAs for shipments is becoming more and more important.

‍

The accuracy and reliability of these predictions depends on the quality of the data captured, and how sophisticated the technology and computational methodology used is. Dataiku, a specialist company helping organizations to systemize their use of data and AI, is helping Shippeo increase its pace of innovation, by offering improved ways to ingest, clean and transform data, design and train ML models, and manage model lifecycles.

Taking technology and innovation to the next level

Here at Shippeo, our teams are dedicated to offer our customers world-class ETA predictions. This means constant retraining of our predictive models, to make sure our customers experience a good level of performance and avoid model drift.

‍

From a technology perspective, it was difficult to maintain reproducible pipelines to create and manage large-scale training and prediction data. As a result, whenever we needed to retrain or improve our ETA’s predictive models, it was also necessary to review the pipelines and even recreate them from scratch. This process means valuable time lost for the team, which could be better invested in research and development to generate more value for customers.

‍

Collaborating on the same AI/ML products can also be quite a challenge, because:

- 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

‍

Here’s what our ETA architecture looked like before we incorporated the Dataiku platform into our stack:

Working with Dataiku a natural fit for Shippeo

Dataiku is used by Shippeo to industrialize how our ETA machine learning models are designed, trained, deployed and monitored. After some investigation, it was clear that addressing this need with an MLOps strategy would also create a competitive advantage.

‍

Here’s what our ETA architecture looks like now, with the Dataiku platform integrated:

‍

Dataiku’s platform offers great benefits around data ingestion, cleaning and transformation, ML model design and training, and model lifecycle management. When using the platform, we noticed a few key advantages:

‍

1. Managing multiple projects

It has given us an environment where we can create and maintain multiple AI/ML projects, along with access to Dataiku’s plugin store, which helps us deploy new monitoring tools and add new features without any custom development required

‍

2. Enabling better collaboration and versioning

The platform improves collaboration amongst data team members to design and train ETA ML models with a versioned project.

‍

3. Less time to market with faster iterations

We’re now able to test our assumptions and try out new approaches faster, and to assess the importance and impact that additional features make on our prediction performance. It’s also now possible to run recurring training for ETA models, enabling continuous learning based on both new and existing customer flows.

‍

A foundation for future value

These new Dataiku capabilities and architecture gives us even greater confidence that we’re offering customers a truly best-in-class ETA. These advancements also made it possible for us to recently enhance the performance of our ETA model with a giant 32% accuracy improvement.

‍

We also plan to expand our offering, by focusing on how to generate new forms of value through leveraging the huge amounts of supply chain and ecosystem data we capture across our network. We believe there is a lot of value in this space but only by using the best tools, infrastructure and expertise, i.e. only with partners like Dataiku.