Key takeaways from Data Science Salon Austin
As a team, we enjoyed listening to a variety of sessions, attending the opening reception, and the conversations that we had at the phData table—with technical innovators, business decision makers, and more. After the conference, we compared notes. Here are some of the key themes we saw over a very busy two days:
Companies see the business value. We heard a lot of talk about business value—how to capture it, quantify it, and increase it. But we didn’t hear anyone questioning whether ML can deliver business value (a welcome change from just a couple of years ago). We appear to be at the point where most companies understand how vital ML solutions are. Now it’s about scaling up.
There are still many pain points. Here are the common ones we heard:
- Deploying solutions at scale. Business leaders talked about excitement about early successes with ML, followed by frustrations as their needs outgrow those first hand-built solutions.
- Operational issues. Many attendees expressed their need for repeatable architecture—using process and automation—to train, deploy, and monitor systems of models, track experiments, and scale up.
- Too many one-offs. We heard frustration about implementations that require each ML model to be hand-built, deployed, and maintained with a custom solution.
- Security concerns. With each hand-built solution, there is an increased risk to mistakenly make an S3 bucket public, for example, or create an over-permissive role.
- Too much time to realize value. In many use cases, developers have to dig through random notes and lines of code to piece together what they need.
- Accumulating technical debt. Models put into production without uniform monitoring, for example, contribute to technical debt and a greater chance of failure in production.
Challenges are the same across industries. Whether they were in finance, healthcare, and technology, companies all faced similar issues. We’ve seen this across industries in our work at phData, too.
Even technology companies struggle. We won’t name names, but several technology companies talked about their difficulties implementing and scaling ML systems. Just because an organization has expertise in one technical discipline, it doesn’t mean they’ll automatically succeed with ML solutions.
Our approach: Automation, people, and experience
There’s a lot of frustration around scaling ML solutions. And there’s a perception that you can fix these problems by throwing more software at them.
Our approach at phData is to think about automation, people, and experience. Yes, it takes software tools to automate ML processes. But it also takes the right people. And the experience to understand how these elements work together. If your organization is struggling with this, we’d love to start a conversation. Reach out to the phData team at firstname.lastname@example.org to see how we can help you.
See you in Boston!
On April 13 – 17, 2020, we’ll be at ODSC East in Boston. We’re proud to be a sponsor of the event, and even more excited to be presenting at it! Both my colleague Robert Coop and I will be giving presentations related to Machine Learning. If you’re attending, be sure to stop by our booth and say hi!