Building a Path to Production for Machine Learning

Businesses are increasing their reliance on machine learning for revenue and cost savings. Many models fail to make it past the prototype stage. Others fail in production.

A Path to Production is familiar for traditional technical deployments. However, it is not well understood or implemented for machine learning products. This is the break between model potential and meeting the needs of the business. A Path to Production for machine learning models has 3 parts:
1. A Model Development Lifecycle.
2. Creating an architecture for deployment.
3. Building a team to handle model implementation, deployment, and maintenance.

Data Scientists, Data Analysts, Machine Learning Engineers, Software Developers, Cloud Architects and sometimes Machine Learning Scientists; these are all roles that play a part in the technical Path to Production. Senior Leadership, Product Managers, Project Managers, Internal Stakeholders, Users; these roles play a part in the business Path to Production.

In this talk, I will be covering the technical and business requirements to build a path to production. I will present an example of taking a model from business requirements to production. I will break this example down to show both technical and non-technical audience members how to build a Path to Production.