State of enterprise machine learning in 2020: 7 key findings

An Algorithmia report released on Thursday revealed the challenges associated with increased machine learning use in 2020. Most companies will be in the early stages of machine learning developments in 2020, but to get to more advanced stages, organizations must overcome a variety of obstacles, the report found. 

SEE: The impact of machine learning on IT and your career (free PDF) (TechRepublic)

Algorithmia’s 2020 State of Enterprise Machine Learning report surveyed 745 tech professionals to determine how organizations plan on deploying machine learning in 2020, and the key issues that accompany the journey. The biggest challenges associated with machine learning deployments involved scaling, versioning, and budgeting, according to the report. 

“AI and machine learning is going to be the most impactful technological advance that we’re going to see in our lifetime,” said Diego Oppenheimer, CEO of Algorithmia. To help organizations in their machine learning efforts, the report broke down their data into the following seven key findings:  

Seven key findings 

1. Rise of data science for machine learning

“The role of the data science is to grab a bunch of the data these companies have been collecting and make sense of it,” and technological advances have caused companies to generate more data, which results in the need for more data scientists, Oppenheimer said.

This increase in demand will continue into 2020 as machine learning becomes more prevalent: Nearly 60% of organizations will employ between one and 10 data scientists, the report found.

More than half of those organizations have at least one machine learning project in place, but these deployments are expected to double by the end of 2020, Gartner found. With machine learning projects expected to increase, the report found that the enterprise will begin seeing new data science job titles including machine learning engineer, machine learning developer, machine learning architect, data engineer, machine learning operations, and artificial intelligence (AI) operations. 

2. Cutting costs takes priority 

The report also looked at what companies want to get out of machine learning. Across the board, the top three use cases included reducing company costs (38%), generating customer insights and intelligence (37%), and improving customer experience (34%), the report found. 

“Machine learning has the ability to, in a lot of cases, reduce errors, which can help a company make more money and save money,” Oppenheimer said. “Like in jobs where there’s a lot of data entry or processing, where there might be a lot of humans involved, where it’s error prone and it’s slightly slow, machine learning can automate a lot of that and make it more precise. It liberates those humans who are doing basic data entry to do higher level tasks, which humans are better suited for.”

While medium to large companies, in particular, are primarily focused on cutting costs, small companies are more interested in improving the customer experience, the report found. 

Smaller companies are trying to retain customers and have steady business–a problem that larger companies may not have. When thinking about how to use machine learning, optimization is a huge use case, Oppenheimer said. 

For example, when a customer comes on the line and is bumped from one customer service employee to another it becomes  “a frustrating experience for everybody, and that’s actually really, costly for the organization,” Oppenheimer said. “There’s a lot of work being done now in terms of as you’re asking your questions, the agent on the other side is typing it into a Google search box. A lot of customer care information is coming up relevant to them, which is using all that data science to serve the customer faster. These are all things that help make the customer experience better, which ultimately makes the customer [loyal].”

3. Overcrowding at early maturity levels and AI for AI’s sake

Machine learning projects will still be in early stages at organizations in 2020: 21% of businesses said they would be evaluating use cases, and 20% identified themselves as early-stage adopters in machine learning production, the report found. 

Respondents said they would be at different stages by the end of 2020, however. Some 23% said they would be working with models in production, and 22% said they would be starting to develop models. 

“You won’t imagine a business in the future that is not using machine learning and data science to optimize their business,” Oppenheimer said. “The problem is that a lot of teams go into it without understanding what the final result needs to look like. The truth is you need to understand what the business optimization needs to look like.”

4. Long road to deployment

Companies take a long time to deploy machine learning. For just a single machine learning model, respondents said they can spend up to 90 days on deployment. Nearly 20% of companies said they take longer than 90 days, the report found. 

The process can take a while because machine learning projects are so new that current data scientists may not be completely familiar with the process, which circles back to why new data scientist job titles will surface in 2020, Oppenheimer said. 

The road to deployment is longer with larger companies, according to the report. The main reason is because the bigger the organization, the more approvals, and people are needed to oversee the project, Oppenheimer said. 

5. Issues with scaling

Scaling models was cited as the biggest challenge (43%) by respondents in the report, up from 30% last year. This challenge is likely associated with decentralized organizational structures, which often result in tooling, framework, and programming language friction during scaling, according to the report. 

“One big obstacle is that there’s a lot of tooling,” Oppenheimer said. “The employees building the models are not usually the best people to be using the scaling. Organizations [need to] realize that these teams need to have different skill sets, and then be able to realize that the frameworks advance really quickly. The machine learning space is moving at lightning speed.”

One solution offered in the report involved creating innovation hubs within organizations. These hubs are dedicated to innovation projects like machine learning and can work in an agile fashion to standardize machine learning efforts, according to the report. 

6. Disparity between budget and machine learning maturity

Machine learning budgets are increasing overall, but vary based on the project’s stage of maturity, the report found. 

Companies at the mid-stages of machine learning maturity increased their machine learning budgets between 1% and 25%, and 39% of those at advanced stages of innovation did the same. Some 30% of organizations at advanced levels of machine learning maturity said they increased their budgets by between 26% and 50%, the report found. 

What this data shows is that “If you can prove success, you’re going to get a bigger budget,” Oppenheimer said. “We’ve seen companies across the board increase their budgets in machine learning and data science. But for companies that have been doing it for the longest, they are getting to a certain maturity level where they’re building it into bigger parts of their business and therefore requiring larger budgets to associate with that.”
 
7. Determining machine learning success across organization 

The top two metrics for determining machine learning success are business metrics and a technical evaluation of machine learning model performance, the report found. 

“At the end of the day, it’s about results,” Oppenheimer said. “A lot of it is around having people involved in the process where the end is understood.”

Teams must determine why they want to implement machine learning projects and find those end goals, rather than just implementing AI for the sake of AI, he added. 

For more, check out Managing AI and ML in the enterprise on ZDNet. 

Also see