Machine learning (ML) is today's big buzzword, but most organizations struggle with turning ML into business value. Everyone wants it, but very few can cost-effectively turn it into reality.
Lynda Partner, a self-described "data addict” and SVP of analytics and offerings, Pythian explored how to accelerate getting to value with machine learning during her Data Summit Connect 2021 presentation, “Machine Learning: Making It Real, Making It Profitable.”
“When it comes to machine learning the biggest challenge organizations have is figuring out how to justify it,” Partner said. “Machine learning takes that challenge and amplifies it.”
The promise of advanced analytics is giving insights that humans may not see, better predictions, and real-time automation at scale, she explained.
It can deliver better decisions, efficiency, and speed; give employees more time for creativity and new opportunities.
But the reality of how widespread advanced analytics is, which includes ML, is drastically different. The good news is, there is a way to change this, she said.
There are several steps included in an ML project. At a high level the ML journey takes a use case, then workers build and train a model, deploy the model, and manage the model ecosystem.
“Picking a use case should start with the business,” Partner said. “But there’s a communication gap between the business and the data scientists.”
To bridge this gap, both roles along with IT need to be involved in the process of selecting which use case will succeed.
To begin this process, business people need to identify their answers to these few questions:
- I would like to…
- So we could…
- Which could result in?
- This would require…
- What would be the business impact…?
- Is this a good ML candidate?
“Sometimes it doesn’t even need machine learning,” Partner said. “Sometimes there’s a simpler solution.”
After discovering the ML candidates, organizations need to rank them. Having a high business impact plus low data risk and adding the highest tech feasibility, becomes the project with the fastest time to value.
“Accessible, clean, consistent, integrated data is the key component to implement successful models,” Partner said.
To determine which use case is the best, the team should configure a data risk score for several areas such as, source quantity, data quantity, accessibility, stability, and quality.
A technical risk score is also helpful. This includes areas such as modeling complexity, time to develop, ML Ops-ability to deploy, and consumption complexity.
There are other steps in the process that can be accelerated, she noted. During the building and training the model stage, data scientists need access to lots of clean data, integrated development environments, knowledgeable experienced modelers and more.
For companies who are in the early stages in ML maturity, she offered four pieces of advice:
- Spend more time on use case selection
- Invest in a cloud data platform
- Start a data governance program
- Educate more people about ML
For companies who have done the above they should:
- Invest in integrated development
- Invest in MLOps skills, tools, and processes
- Start thinking about model management
“No one group in the company can say they can improve the ROI and make it a success, it needs to involve collaboration between all the people involved,” Partner said.
More information about Data Summit Connect 2021 is available here.
Replays of this and all Data Summit Connect 2021 sessions will be available to registered attendees for a limited time and many presenters are making their slide decks available as well.