Mercado Envíos in a nutshell
I lead a team of designers through the creation of a new logistic network for the largest e-commerce company in Latin America, responsible for delivering 1.8 million packages per day.
Context
MercadoLibre hosts the largest online commerce and payments ecosystem in Latin America. Their efforts are centered on enabling e-commerce and digital and mobile payments on behalf of their customers by delivering a suite of technology solutions across the complete value chain of commerce. This included the logistics solutions for delivering 1.8 million packages per day.
Framing the problem
In order to reduce our delivery lead time from five days to three days on average, we needed to increase the number of logistics companies we worked with. That meant incorporating smaller carriers which specialized in delivering packages within specific areas. But these carriers lacked the software tools that would guarantee the experience we wanted for our buyers. So we built those tools, allowing forty small carriers to join the network within one year.
Research
The first step was learning and understanding how delivery operations across Latin America worked. We did field research in Brazil, Mexico, and Argentina, mapped the user’s journeys, talked with postmen and operators about their daily tasks, and learned their vocabulary. Back in the office, we shared our insights with the rest of the team: developers and product owners. During long workshops, we created User Personas, Jobs To Be Done, and consolidated what we called “The Package Journey”.
MVP - Minimum viable product
Defining the MVP was a difficult process because even the smallest version seemed huge, and involved different technologies and teams. To make this conversation easier, I divided the team into groups based on the User Personas. Each team had to define their own MVP, and this approach helped us focus on smaller problems and come up with simpler solutions. Here is an overview of the result:
The Driver was defined as the one driving and delivering packages around the city. Their work tool was a mobile app, where they could see their route and mark packages as “delivered” or “not delivered”. The first version did not include a map with the route, just a list of addresses.
The Carrier was defined as the Driver’s employer, who controlled productivity and sent help in case the driver needed it. Their work was done from a desktop app where they could see the driver’s locations on a map and KPI’s, like “delivery success”.
The Operation Manager was a Mercado Envío’s employee, controlling multiple Carriers that worked in the same warehouse. Since they dealt with complex data, they used a desktop app, where they could plan routes and assign them to different Carriers.
Rollout
We needed this product on the streets before peak selling season, so we designed, tested, and implemented this MVP in six months. It had a lot of functionalities missing, but the first warehouse worked as we expected. That enabled us to quickly open another thirty-two warehouses which, in turn, were able to deliver 80K packages per day during the 2019 Christmas sales. That volume of packages brought about new challenges, so we started researching and designing the following iterations for the product.
Iterating the Routing Tool
The challenge was huge, and we had a lot of problems to solve within the product. But of one example of the iterative process we follow was the routing tool, which went from being a human-based solution to a machine-learning algorithm.
Version One: Learn from human decisions
During field research, we learned that Drivers had a lot of strategies for defining the delivery order, for example, they avoided dangerous areas of the cities (that weren’t marked on the map) or they changed their path to steer clear of schools at noon.
We understood that we didn’t have enough information to build optimal routes, so we displayed a list of addresses on the app to let drivers choose their own routes. Meanwhile, we started tracking their movements to learn from them.
Version Two: Learn from an external solution
The number of packages kept increasing and each Driver had to visit forty addresses every day, so we needed to help them organize their route. We benchmarked different routing engines that were available in the market and we found RouteEasy.
This software company didn’t specialize in delivery routes, but we managed to define some constraints and build a backend integration that provided a suggested route on the driver’s app.
We knew the routes weren’t perfect, so we still allowed our Drivers to rearrange the delivery order. But we had helped them by giving them an organized list of addresses as a guide.
Version Three and beyond: Learn from iteration
Based on the data we already had available, we started crafting our own machine-learning algorithm to create the routes. This was a complex problem that included the driver’s decision, but also the business goals and the constraints the warehouse faced each day.
Our machine learning model had grown but still needed human decisions in some of the steps. We have iterated and designed these steps to provide a good user experience and also data inputs for our model.
Fast-Growing team and product
With a fast-growing product and operation, our team was also bound to grow. My team went from two designers to nine within a year, which posed the challenges of hiring and onboarding new members. But I learned a lot, managed to keep a nice working environment, and guided the professional growth of my team.
Today, there are 295 warehouses running, that deliver 1.8 million packages per day in five countries. That means that almost 10.000 drivers used Logistic’s app to work every day, and millions of buyers around Latin America get their orders in less than two days.
