Meet The Team: Data And Business Analytics

What we do as a team By Tarik Agayr, Lead Product Manager - Data Science & Analytics at Shippeo

Over the years, we've been gathering a significant amount of extremely valuable data on shippers’ logistics operations with real-time tracking information. Our team was created to take advantage of this data to build new "data products".

Data products are products whose primary objective is to rely on data to achieve an end goal. At Shippeo, these fall into 4 different categories: ‍

- Raw data extraction: We make the data stored in Shippeo available -as is- to everyone who wishes to extract and use it. That's the Shippeo EXPORT feature, which we keep feeding as the data model gets enriched with new information

- Analytics: We provide users (both internal and external) with analytics to get insights from the data, and help them with their decision-making. This includes INSIGHTS (overall operational performance, tracking compliance), as well as many dashboards that the team builds for our operations team to answer questions such as: “How relevant are the alerts that we are sending?” “How do the telematic providers we are connected to compare in terms of coverage, frequency of positions sent, latency?” etc... Our INSIGHTS offering will grow during months to come with in-depth analysis of the performance per lane

- Algorithms: We also build algorithms to add a "predictive layer" on top of the "real-time layer" of the Shippeo platform. We are given some data (order, itinerary, position data...), from which extract a lot of parameters in order to generate more parameters by transforming the raw data to predict an ETA for each stop of the itinerary, using Machine Learning methods

- Decision support tools: We develop tools to provide users with insights contained in the data to allow for informed decision-making. For example: we are about to release a tool called RE-GEOCODE, which detects addresses that are not properly geocoded (because of typos contained in the address details, missing digits in the postal code, etc), and suggests to the user alternative locations which are more likely to be the correct ones, based on historical tracked trips.

How we develop such performant algorithms to establish precise ETA‍

There are three key factors we need to satisfy in order to carry out successful ML projects.‍

Firstly, defining clear objectives : defining what we want to achieve and what we believe is the best, most effective way to get there.

Second, relying on an in-house team with mixed skills: functional, data engineers, data scientists. Our project is not outsourced to external service providers and we involve knowledgeable Subject Matter Experts throughout the entire ML project lifecycle.

Finally, we use clean data with exploitable records only. The data we have at our disposal satisfies our requirements both in length (sufficient number of records to train ML models) and in width (thorough list of variables that can influence the output), and our team has full knowledge of the content of the data and the inherent rules.

We process and transform raw data which is of a high quality using cutting-edge techniques. We've been doing this for years. We’ve tried different approaches. We now know what works.

What we’re currently working on

Our team is getting bigger and we're getting better.

We’re currently on a on a quest to decrease the global average error on our ETA as much as possible (0 would be ideal but sadly unachievable), and also control the variance. We’re also building more dashboards, to measure everything that can be measured and incorporating more analytics that will support informed decision-making.

At term, we want to provide our customers with the undisputed best ETA of the market and start working towards automated decision-making.