Passa ai contenuti principali

Amazon’s SCOT

Amazon’s SCOT

The most amazing application of mathematical optimization in logistics

What is SCOT?

The state of the art in advanced mathematics at the service of the operations management. I like how some important companies implement this scientific development. — Julio Mora Olivares
SCOT is an acronym meaning Supply Chain Optimization Technologies. Probably it is the largest and the most complete application of operational research in the field of logistic.

In summer 2017, I discovered that Amazon was hiring operational researchers in Italy. This intrigued me enough to start some research on the Internet. I found a plentiful source of information on LinkedIn profiles, where people were sharing their working experience, their goals and successes.

I’ve collected all the pieces of information and, with my experience, I was able to reconstruct a whole picture of this project.

Human resources

First things first: people

The team working on SCOT project covers many countries. In my search, I’ve noted places from North America, India, and Europe. This is a quite common characteristic of projects developed in large companies as Amazon.

What really impressed me indeed is to see so many roles involved in the same project. They are:

  • Econometricians,
  • Statisticians,
  • Economists,
  • Operational Research Scientists,
  • Machine Learning Scientists,
  • Software Engineers,
  • Product Managers


So, from the management point of view, the project leader has to handle two independents source of complexity: many roles in many places.

Applications

Fantastic combination of Optimization, Computer Languages. and Inventory Management — Deepak Bammi
Have you ever bought something from Amazon? A book, of course. But also other goods as a Kindle reader or a cover for your smartphone.

I did all of this and every time, before the purchase order, I knew a precise forecast of the delivery date. Typically one day for many Prime products.

A so elevated level of service quality can be achieved with a full control of the supply chain. From my research seems that Amazon invested a lot on:

  • inventory planning and control;
  • last mile logistic network.

Inventory planning and control

The picture above was taken from an Amazon video on YouTube. Probably it gives only a vague idea of the inventory complexity managed behind the button “buy now with 1-click (r)”.

If you want some totals here they are:
Hundreds of millions of items, for hundreds of product lines worth billions of dollars of inventory world-wide.
In a single sentence, the team of inventory planning control of Amazon:
decide and optimize what, when, where, and how much should buy to meet Amazon’s business goals and to make their customers happy.

Last mile logistic network

Another important area where SCOT gives a major improvement is the last mile logistics.

If you think of the entire user experience, the parcel delivery is the only ring of the supply chain out of the control of Amazon. But everybody knows how Amazon is obsessed with customer experience and satisfaction.

This means not only the optimal location of the delivery point to get an appropriate geographic coverage. But also control a complete fleet of people and vehicles physically delivering goods.


Technologies and IT impacts

From the technology point of view, SCOT has many interesting things to say.

Optimization library

SCOT is an optimization library designed and built with the purpose to facilitate the integration, of optimization models, within the Amazon’s infrastructure without any knowledge of optimization.

General purpose optimization engines

SCOT uses well known and powerful optimization engine: CPLEX, XPRESS and GUROBI.

Programming languages

SCOT has some kind of relationship with the following programming languages:

  • ruby
  • VB
  • c#
  • java
  • python (my favorite)

I don’t know what is the main language used for SCOT. Also, I didn’t find any relationship with compiled languages such as C or C++.

SCOT in a video

If you want, you can enjoy this short video from Amazon explaining what is SCOT.

Disclaimer

All the information reported in this document are gained from the internet, especially from profiles published on LinkedIn. I don’t work for Amazon, so this report collects my suppositions, suggested by my own experience in optimization technologies.

Commenti

Post popolari in questo blog

PuLP – Un valido strumento per la didattica

L'insegnamento dei concetti di base della ricerca operativa, ovvero la programmazione lineare, ha trovato nel corso degli ultimi anni diversi strumenti di supporto. Sono ormai parecchi i software gratuiti e open source che permettono agli studenti e agli insegnanti di toccare con mano le nozioni e i concetti spiegati e studiati sui banchi. Ricordiamo, ad esempio, glpk che con il suoi linguaggio di modellazione MathProg permettete di scrivere e risolvere anche complessi modelli di programmazione lineare intera. Oppure citiamo anche lp_solve che con il suo ambiente impropriamente chiamato lp_solve IDE permette di scrivere e risolvere modelli di programmazione lineare direttamente nella formulazione matematica. A mio avviso però le proposte appena citate sono limitate nella potenza espressiva e nelle capacità di integrarsi con altri software o moduli esterni. Queste limitazioni sono egregiamente risolte da PuLP : un modellatore di problemi di programmazione lineare intera basato

Ci arricchiremo con la ricerca operativa?

A questa domanda forse possiamo rispondere sì :-) , rimandando al lavoro molto fresco ed interessante di Giancarlo Volpe dal titolo " Scommesse sportive: un modello di Ricerca Operativa che descrive la “vincita perfetta” " E' possibile scaricare il documento da scribd.com . Dall'apprezzabile contenuto didattico la parte entrale, dove si illustra passo passo come è possibile usare il risolutore di excel per applicarlo al modello descritto. Buona lettura e giocate con moderazione. Un Modello di Ricerca Operativa per Scommesse Sportive

Dispense di ricerca operativa

Ho trovato sulla home page del prof. Agnetis, delle interessanti dispense di ricerca operativa. I temi trattati sono tutti molto interessanti: Appunti sul duale del problema del massimo flusso Appunti sui problemi di matching Appunti su classi di complessità e problemi NP-completi Appunti sul problema del TSP euclideo Appunti sulla generazione di colonne Appunti sui modelli di lot sizing: Wagner-Whitin, Zangwill, Florian-Klein Appunti sui problemi di scheduling Appunti sui metodi metaeuristici di ricerca Introduzione all'ottimizzazione non vincolata   Introduzione all'ottimizzazione vincolata Esercizi di ottimizzazione non vincolata  Condizioni di KKT e Programmazione Lineare  Esercizi di ottimizzazione vincolata   Raccolta di esercizi di PL svolti  Esercizi di esame di PL svolti Esercizi di PLI svolti Appunti sui metodi basati sul rilassamento Lagrangiano Esercizi d'esame (R.O.) di ottimizzazione non vincolata e vincolata Ottimizzazione nella Gestione