Habitat Sees Success With ML Demand Forecasting

Case Study: Habitat Logistics

Posted by Ignacio Vago

on November 01, 2022 · 2 mins read

Habitat Logistics Sees a Cost Reduction Of Over 40% Using ML-Based Demand Forecasting and Optimization Solution

About Company

Habitat Logistics is a Y Combinator accelerated B2B delivery outsourcing platform for restaurants in the United States which works directly with restaurants for a low commission. Restaurants can receive orders from any ordering channel, and have Habitat fulfil these orders for a flat, fixed fee. Their mission is to help restaurants start, keep, and grow their business. Habitat Logistics operates in a Three-Sided Marketplace working as a link between restaurants and final consumers.

Challenge

Habitat was using a set of complex business rules to manually estimate the demand forecast by hour and assign the proper amount of delivery persons. They had hired an external Data Science consultant who applied Machine Learning to computing the forecast and showed improved results but had no expertise with turning that into an actual system on the AWS Cloud that would automatically run in an hourly manner.

Solution

We deployed a team of 5 High-Performance In-House Data Experts to plan, organize, and develop all the necessary AI and data capabilities Habitat needed to solve their challenge effectively and in record time. The goal? Robust systems and capabilities, built to last and scale with Habitat’s business.

We implemented a system that applied demand prediction algorithms to calculate hourly demand forecasts for each delivery area in each city. This system runs continually every hour and adjusted predictions accordingly for the next 7 days. To accurately calculate this demand, data is pulled, integrated and validated from several external data sources such as weather forecasts, special events, etc.

Using these hourly demand forecasts and input, we built an optimization system to adjust delivery person’s shift allocation to minimize costs while maintaining SLA’s.

Finally, we implemented a real-time prediction system to estimate the time it would take to prepare a particular order. This allowed the dispatch system to minimize waiting times in the restaurant for the order to be ready.

Data and ML pipelines and models were implemented combining the use of Apache Airflow for process development, Python for application development and ML Flow for metric and model tracking. The whole Data Architecture for this system was designed, built and maintained by Mutt Data’s team.

Impact

Habitat saw Over 40% cost reduction in their deliveries. Habitat leapfrogged their data journey to operational success through and automated and optimized solution which allowed them to scale and extend their delivery business in a small window of time.

Want To Dive In Deeper?

Mutt Data can help you crystallize your data strategy through the design and implementation of technical capabilities and best practices. We study your company’s mission to understand what has to change so we can help you accomplish it through a robust product and technical strategy with a clear roadmap and set of goals. Talk to one of our sales reps or check out our sales booklet and blog.

Want To Read More On ML in Three Sided Marketplaces?

Check out our recommendation on three articles you need to read about three-sided marketplaces and our blog posts on Improving the three-sided marketplace with machine learning and Delivering your Marketplace’s Orders in Time with Machine Learning.


Case Studies