HFS Research discussed the potential, hype, and progress of machine learning (ML) within enterprises in a candid conversation with Robbie Allen, CEO of Infinia ML. Robbie is currently writing a book titled Machine Learning in Practice, which includes his views informed by decades of academic and practical experience.
Q & A
Reetika Fleming, Research Director, HFS Research: First, how would you describe your vision for ML in the enterprise?
Robbie Allen, CEO, Infinia ML: Nothing too controversial! There’s definitely a sense that more people are doing ML. In practice, most are thinking of getting started or have only just started. There are relatively few that have fully embraced ML in all its glory. That fact is due partly to the lack of talent, and partly to upwards of ten years of backlog of opportunities in ML. At the current rate of innovation, that’s only going to grow. This much is true—there won’t be a lack of opportunity of implementing ML, at least for the next decade.
Reetika: From your vantage point, is adoption getting real? Any highlights of where it’s becoming live or is in production?
Robbie: The bulk of the applications in ML are around the inputs available since it’s all about the data. There are three forms that data can take—images and video, text, and numerical data. The bulk of press attention and public innovations have been on image and video applications like object recognition. That’s where ML shines. The ML in image processing has gotten very good, and tasks that require visual inspection or recognition will increasingly be automated. Meanwhile, text data is harder than images as a data input, as it is more unconstrained. However, there have been good advances in NLP applications. In the enterprise context, we are likely to see some interesting use cases in NLP. Most organizations don’t have image or video data; text assets and datasets are more common.
Reetika: So, will these be virtual agents and other conversational interfaces?
Robbie: They won’t necessarily be chatbots or virtual agents, though those are popular applications. The analysis of text in emails or contracts, for instance, can reveal great insights.
Reetika: How do you move ML out of labs and projects to get to the promised land of “industrialized” ML-enabled processes across enterprise operations?
Robbie: It’s happening now, and Infinia ML is evidence of that. We are taking cutting-edge algorithms and techniques and deploying them in companies today. There isn’t a lack of opportunity within companies. But there are all sorts of challenges. One of the biggest is the lack of people that understand ML, especially the more advanced techniques like deep learning.
Reetika: How do you see the industry getting over this massive talent gap?
Robbie: There’s so much content out there now in ML and data science; motivated people can find great material from top universities. For a relatively small amount of money, you can go online and get the skills you need. The proliferation of educational content will certainly continue.
On the other hand, the tools that are necessary to implement ML will get easier over time. It requires a high amount of training and understanding today to properly develop and deploy an ML solution. You’ll see this proliferation of new tools and frameworks, and even the existing ones will get better over time. Right now, we’re in a challenging period, but in the next five to 10 years, we’ll be significantly better off.
Another aspect of talent is that when you automate something, you usually don’t implement the solution and fire all the people that were doing that task. Generally, there’s a need to repurpose many of the existing resources to become “ML assistants.” Data entry, data cleaning, data searching, data mapping—those tasks can be automated. To make algorithms better, though, you need somebody validating the accuracy of the predictions from the ML models. Enterprises thus have to flip their organizational setups from staffing the input side to the output side.
Reetika: In your experience, how should clients think about their top use cases for ML? Any dos or don’ts?
Robbie: You need a good understanding of what you need to solve. Many organizations get hung up on goals; that’s step zero. After that, it’s all about the data. Even if you have a clearly articulated vision, without data, machine learning is not happening. If you will drink the ML Kool-Aid and embrace it, you will need to think about your organization differently. The more you can plan for things like the ML assistant role, the better off you will be. When you start a project, paint a picture for employees: we won’t fire you, but you will have a new role.
There’s also a lot of work in preparing the organization to embrace data and a data science culture. Data challenges haven’t changed much since I started doing data-intensive projects in organizations 10 years ago. To take advantage of ML, you need to have data assets that are accessible, sizable, usable, understandable, and maintainable.
The companies that make data accessible will be able to innovate most rapidly. Often people will put together data lakes that are actually data swamps. The data is going in, but not coming out… a lot of organizations are creating data infrastructure that’s not very useful. Starting down the path of data infrastructure modernization is crucial.
Reetika: Do you want to talk a little bit about any initiatives that Infinia ML is taking to help this market mature, in a way?
Robbie: We have what we call our 3D Approach to ML. The first step is data preparation, followed by development, and then deployment. Deployment is usually the least mature function in client organizations. We work to educate companies on these steps and what it takes to move through them. Everyone is suffering from not having highly trained people, so we want to help educate data scientists, developers, and especially business leaders.
HFS RWe recently held live training for business leaders based on the book we are writing, Machine Learning in Practice. We explain the methodology in business terms because there are so many complex subjects that are confusing if you don’t have a Ph.D. in ML. There’s a large amount of content for technical people, just not that much for non-technical people that are making the buying decisions…and that’s why we started writing the book.
Reetika: Speaking about non-technical skills, our recent research, How To Avoid Your Looming Machine Learning Crisis, showed that the biggest ML skillset gap is not exactly something you can train for—analytical curiosity and problem-solving creativity. How do we develop this as an industry?
Robbie: I’m teaching an upper-level MBA class for UNC’s Kenan-Flagler Business School called Machine Learning: Strategy and Execution. There’s definitely high demand and interest for that subject—we even had a waitlist. Because ML is different from classical software development, there are new things people need to learn about: moving through organizational and political issues, having access to data, securing funding, building the ability to execute, and wrapping that into innovative projects. Much of the material we cover is not technical, but rather these “soft” skills.
Reetika: Do you see ML having a two-year time to impact, as our study found? Or is it a longer tail when you talk about industrialized, embedded ML-based processes?
Robbie: At least for the projects we’re doing, we can deliver value in four to six months. That’s our average project length for taking data and automating some tasks with ML. Companies are seeing significant opportunities now. Building a data science culture and changing your organization to use ML opportunities will take time. But, certainly within a year, enterprises can easily be on the path to see advances in processes and products.
Reetika: From your career spent in this industry, are there any myths about the current state of ML that you would like to bust for enterprise clients?
Robbie: You read the tech press and see advances daily and think there’s so much going on, everyone’s got to be doing ML! I feel this myself sometimes. In all my conversations, enterprises are admitting, “We haven’t even started anything yet.” The challenge is that the tech press continues to blow up all the advancements. It’s OK if you’re just getting started—you’re not alone!
Q & A
Reetika Fleming, Research Director, HFS Research: First, how would you describe your vision for ML in the enterprise?
Robbie Allen, CEO, Infinia ML: Nothing too controversial! There’s definitely a sense that more people are doing ML. In practice, most are thinking of getting started or have only just started. There are relatively few that have fully embraced ML in all its glory. That fact is due partly to the lack of talent, and partly to upwards of ten years of backlog of opportunities in ML. At the current rate of innovation, that’s only going to grow. This much is true—there won’t be a lack of opportunity of implementing ML, at least for the next decade.
Reetika: From your vantage point, is adoption getting real? Any highlights of where it’s becoming live or is in production?
Robbie: The bulk of the applications in ML are around the inputs available since it’s all about the data. There are three forms that data can take—images and video, text, and numerical data. The bulk of press attention and public innovations have been on image and video applications like object recognition. That’s where ML shines. The ML in image processing has gotten very good, and tasks that require visual inspection or recognition will increasingly be automated. Meanwhile, text data is harder than images as a data input, as it is more unconstrained. However, there have been good advances in NLP applications. In the enterprise context, we are likely to see some interesting use cases in NLP. Most organizations don’t have image or video data; text assets and datasets are more common.
Reetika: So, will these be virtual agents and other conversational interfaces?
Robbie: They won’t necessarily be chatbots or virtual agents, though those are popular applications. The analysis of text in emails or contracts, for instance, can reveal great insights.
Reetika: How do you move ML out of labs and projects to get to the promised land of “industrialized” ML-enabled processes across enterprise operations?
Robbie: It’s happening now, and Infinia ML is evidence of that. We are taking cutting-edge algorithms and techniques and deploying them in companies today. There isn’t a lack of opportunity within companies. But there are all sorts of challenges. One of the biggest is the lack of people that understand ML, especially the more advanced techniques like deep learning.
Reetika: How do you see the industry getting over this massive talent gap?
Robbie: There’s so much content out there now in ML and data science; motivated people can find great material from top universities. For a relatively small amount of money, you can go online and get the skills you need. The proliferation of educational content will certainly continue.
On the other hand, the tools that are necessary to implement ML will get easier over time. It requires a high amount of training and understanding today to properly develop and deploy an ML solution. You’ll see this proliferation of new tools and frameworks, and even the existing ones will get better over time. Right now, we’re in a challenging period, but in the next five to 10 years, we’ll be significantly better off.
Another aspect of talent is that when you automate something, you usually don’t implement the solution and fire all the people that were doing that task. Generally, there’s a need to repurpose many of the existing resources to become “ML assistants.” Data entry, data cleaning, data searching, data mapping—those tasks can be automated. To make algorithms better, though, you need somebody validating the accuracy of the predictions from the ML models. Enterprises thus have to flip their organizational setups from staffing the input side to the output side.
Reetika: In your experience, how should clients think about their top use cases for ML? Any dos or don’ts?
Robbie: You need a good understanding of what you need to solve. Many organizations get hung up on goals; that’s step zero. After that, it’s all about the data. Even if you have a clearly articulated vision, without data, machine learning is not happening. If you will drink the ML Kool-Aid and embrace it, you will need to think about your organization differently. The more you can plan for things like the ML assistant role, the better off you will be. When you start a project, paint a picture for employees: we won’t fire you, but you will have a new role.
There’s also a lot of work in preparing the organization to embrace data and a data science culture. Data challenges haven’t changed much since I started doing data-intensive projects in organizations 10 years ago. To take advantage of ML, you need to have data assets that are accessible, sizable, usable, understandable, and maintainable.
The companies that make data accessible will be able to innovate most rapidly. Often people will put together data lakes that are actually data swamps. The data is going in, but not coming out… a lot of organizations are creating data infrastructure that’s not very useful. Starting down the path of data infrastructure modernization is crucial.
Reetika: Do you want to talk a little bit about any initiatives that Infinia ML is taking to help this market mature, in a way?
Robbie: We have what we call our 3D Approach to ML. The first step is data preparation, followed by development, and then deployment. Deployment is usually the least mature function in client organizations. We work to educate companies on these steps and what it takes to move through them. Everyone is suffering from not having highly trained people, so we want to help educate data scientists, developers, and especially business leaders.
We recently held live training for business leaders based on the book we are writing, Machine Learning in Practice. We explain the methodology in business terms because there are so many complex subjects that are confusing if you don’t have a Ph.D. in ML. There’s a large amount of content for technical people, just not that much for non-technical people that are making the buying decisions…and that’s why we started writing the book.
Reetika: Speaking about non-technical skills, our recent research, How To Avoid Your Looming Machine Learning Crisis, showed that the biggest ML skillset gap is not exactly something you can train for—analytical curiosity and problem-solving creativity. How do we develop this as an industry?
Robbie: I’m teaching an upper-level MBA class for UNC’s Kenan-Flagler Business School called Machine Learning: Strategy and Execution. There’s definitely high demand and interest for that subject—we even had a waitlist. Because ML is different from classical software development, there are new things people need to learn about: moving through organizational and political issues, having access to data, securing funding, building the ability to execute, and wrapping that into innovative projects. Much of the material we cover is not technical, but rather these “soft” skills.
Reetika: Do you see ML having a two-year time to impact, as our study found? Or is it a longer tail when you talk about industrialized, embedded ML-based processes?
Robbie: At least for the projects we’re doing, we can deliver value in four to six months. That’s our average project length for taking data and automating some tasks with ML. Companies are seeing significant opportunities now. Building a data science culture and changing your organization to use ML opportunities will take time. But, certainly within a year, enterprises can easily be on the path to see advances in processes and products.
Reetika: From your career spent in this industry, are there any myths about the current state of ML that you would like to bust for enterprise clients?
Robbie: You read the tech press and see advances daily and think there’s so much going on, everyone’s got to be doing ML! I feel this myself sometimes. In all my conversations, enterprises are admitting, “We haven’t even started anything yet.” The challenge is that the tech press continues to blow up all the advancements. It’s OK if you’re just getting started—you’re not alone!
Register now for immediate access of HFS' research, data and forward looking trends.
Get Started
If you don't have an account, Register here |
Register now for immediate access of HFS' research, data and forward looking trends.
Get Started