How Novartis AG brought SMART into Smart Procurement with AWS Machine Learning (Part 1/4)
September 8, 2021: Amazon Elasticsearch Service has been renamed to Amazon OpenSearch Service. See details.
In this series:
- Part 1: How Novartis AG brought SMART into Smart Procurement with AWS Machine Learning (this post)
- Part 2: Novartis AG uses Amazon SageMaker and Amazon Neptune to build and enrich a knowledge graph using BERT
- Part 3: Novartis AG uses Amazon OpenSearch Service K-Nearest Neighbor (KNN) and Amazon SageMaker to power search and recommendation
- Part 4: Demand Forecasting with Amazon SageMaker and GluonTS at Novartis AG
The strategic collaboration between Novartis and AWS
On December 4, 2019 AWS announced a strategic collaboration with Novartis AG to accelerate digital transformation of its core business including manufacturing, supply chain, and delivery operations. To date, we have kicked off several initiatives that are serving as building blocks towards a new normal for the healthcare and life sciences industry, transforming Novartis’ operations and enabling efficiency through technology.
Novartis, headquartered in Basel, Switzerland, is reimagining medicine to improve and extend people’s lives. Innovation in medical science while leveraging digital technologies is one of Novartis’ key tenets in transforming treatments and producing products that reach 800 million people globally. With a history of more than 250 years, Novartis employs about 110,000 people and invests in research and development of innovative treatments and medicines.
In an effort to utilize data-driven methodologies and improve decision making across the organization, Novartis worked with AWS Professional Services, a global team of experts that specialize in delivering valuable business outcomes on the AWS Cloud. The outcome of this collaboration is a portfolio of solutions aiming to support Novartis in tackling industry-specific problems through the use of advanced technology. One of them is Novartis Buying Engine (BE), an AI-driven platform that uniquely combines a cognitive storefront with a supply chain automation back-end and enables lab scientists and researchers in making informed decisions faster.
In this four-part blog post series, we will explain the Buying Engine and its business importance as well as dive deep into technical concepts behind its features. In this blog we will introduce the Buying Engine, elaborate on the transformation it brings to Novartis, and uncover the technology used to deliver an AI-powered procurement marketplace with product demand prediction and planning automation.
The business goals of the Buying Engine platform
As a leading pharmaceutical organization, Novartis purchases millions of lab supplies from vendors around the world to run research, development, and delivery operations. Traditionally, Novartis procurement has been decentralized within regional business units, with several buying practices used across multiple enterprise resource planning (ERP) systems. The long-established process of purchasing lab supplies required employees to spend extensive time manually comparing relevant item prices in catalogs on vendor websites and issuing purchase orders with individual suppliers. Information asymmetry on the vendor products coupled with inconsistent buying behaviors resulted in overspending, inventory mismanagement, in addition to research, development, and manufacturing delays.
To enhance and optimize this process, Novartis and AWS initiated the Buying Engine project, building upon an initial MVP created internally within Novartis. The resulting solution hosts products and services from different vendors in a centralized repository. Users can search and compare products, get recommendations, and predict demand and automate purchasing without being limited to a single vendor or product catalog.
The Buying Engine’s commercial goal/outcome is to enable 5% yearly cost saving on procurement spending with progressively increasing the total addressable spend within 2020. Novartis has a long-term vision and scaling plan that allows the Buying Engine to cover a majority of their procurement spending within the next several years.
With Buying Engine, Novartis Procurement is seizing the AI opportunity to drive the digital wave to new heights by transforming the complex procurement processes. This will further help us grow, manage risk, and ultimately achieve sustainable competitive advantage.
– Amit Nastik, Head of NTO Strategy & Operations
The technical aspects of the Buying Engine platform
The Buying Engine is built entirely on AWS and now operates in two major Novartis sites, in Austria and the US. AWS was identified to build such a platform because it provides access to a scalable, reliable, and secure computing infrastructure offering a broad catalog of analytics and machine learning services. The scalable and flexible characteristics of AWS combined with a modular microservice architecture designed for Buying Engine allow for plans to deploy the solution in eight new Novartis sites in Europe, US, and Asia by November 2020.
To platform operates in the following way:
- Procurement products from third-party vendors are ingested in the form of product catalogs
- The catalogs are analyzed, enriched, and categorized by utilizing state-of-the-art machine learning techniques
- The processed products data is ingested in a set of databases that form the Novartis Product Knowledge Base
The following high-level diagram depicts the Novartis Buying Engine architecture, which is divided into three major components (data ingestion, knowledge base, and data discovery).
Behind the scenes, data ingestion is achieved through scalable, serverless data pipelines, leveraging AWS Step Functions, AWS Lambda, and AWS Glue for data processing and Amazon S3 for data storage. A scalable system with high throughput is crucial, as Novartis already handles over three million products for only two out of the eight planned Novartis sites going live. Thus, the whole data ingestion pipeline is streamlined and stress-tested to adjust based on the need. The data ingestion pipelines are, finally, responsible for further data enrichment with custom Natural Language Processing (NLP) models built on Amazon SageMaker.
The foundational component of the Buying Engine is its centralized knowledge base, a sophisticated data storage mechanism. The knowledge base is built on top of Amazon DynamoDB, a key-value document database; Amazon Neptune, a graph database; and Amazon OpenSearch Service (successor to Amazon Elasticsearch Service), a search database. This creates a schema-less information storage, performant querying mechanisms, and graph traversal algorithms. This powerful combination of AWS databases provides the right tools to achieve Data Discovery.
Buying Engine users can leverage a web user interface that is connected to the data discovery layer. The layer exposes functionalities such as search and recommendation, letting users search for products within the catalog using free text. The underlying search engine combines keyword matching and ML techniques to provide a frictionless experience for users. Additionally, the recommendation engine enables users to continue browsing the catalog while accessing recommendations along their journey.
Demand forecasting is another crucial component for cost optimization in procurement. As Novartis works towards building an automated replenishment engine driven by demand; forecasting and optimization of the procurement chain for all goods and services is a core building block of it. Being able to predict demand for each Stock Keeping Unit (SKU) at particular geography several months in advance allows Novartis to make faster data-driven decisions, plan better, negotiate contracts and discounts, and save costs.
In this post, we’ve provided an overview of the Buying Engine and described the transformation it brings to Novartis procurement. To learn more about the technical concepts behind the Buying Engine components, check out part two, three, and four in this series.
Many thanks to Novartis AG team who worked on the project. Special thanks to following contributors from Novartis AG who encouraged and reviewed the blog post.
- Srayanta Mukherjee: Srayanta is Director Data Science in Novartis CDO’s Data Science & Artificial Intelligence team and was the data science lead during the delivery of the Buying Engine.
- Deepanshu Mehta: Deepanshu is Solution Director for the CDO DSAI-Innovation Execution team at Novartis and was the Solutions lead during the delivery of the Buying Engine project.
- Shubha Chaudhry: Shubha is Head of Digital for Procurement, NBS at Novartis and the Product Owner for the Buying Engine Project.