New Capital One and Forrester Study Reveals Key Challenges Of Democratizing Machine Learning
In a study commissioned by Capital One, Forrester Consulting surveyed 181 data and analytics and line of business (LOB) decision makers at North American companies about democratizing ML and the opportunities it presents for their firms.
The study highlights that as the demand for ML-driven insights outside data science and IT roles increases, democratizing machine learning becomes more critical. For the successful democratizing of ML, businesses would have to speed up and scale the deployment of ML applications across organizations. However, there are some key challenges related to governance, building trust in data, and communication.
The findings of the study show that ML is increasingly tied to business success. Of those surveyed, 88 percent of decision-makers said they believe ML is a key element of business success. While LOB leaders are extremely confident about the potential positive impact of ML (95 percent), data role leaders are less enthusiastic (81 percent).
Most respondents (86 percent) reported their companies were already democrating models for ML use, and even more respondents (91 percent) said data engagement across teams was increasing.
While LOB leaders are excited about ML-powered tools, the findings of the study show that the available tools and capabilities are too technical for them. Only 27 percent of LOB respondents reported having access to user-friendly tools, compared to 39 percent of data respondents. Overall, 67 percent of respondents agree the lack of easy-to-use tools is slowing across-enterprise adoption of ML. This shows a need for more intuitive and low-code/no-code ML applications.
For data leaders, ML democratization is not as easy as it sounds. The biggest challenges include applying governance policies within AI/ML (50 percent), the cost of computing to train and run models (46 percent), using the correct algorithmic technique or approach (45 percent), and having adequate data and model security (45 percent).
Some of these challenges, such as model security and governance policies, are partly a result of the newness of Ml initiatives. However, some organizations are simply not prepared to handle the increased data traffic of ML applications.
A high percentage of respondents (95 percent) shared that they need reliable data input or data pipeline to generate consistent output. The challenges related to trust and security can be solved with well-communicated governance. However, organizations need to strike the right balance between governance and limiting the capabilities of platform and process. According to the Forrester study, the key to balancing this issue is to have ambient governance, where LOBs have the freedom to do what they need without being overly concerned about permissions. Ambient governance helps organizations meet their regular compliance requirements while instilling confidence in employees to engage with data.
The findings of the survey point to cultural challenges being more pervasive than technical challenges in the ML democratization process. Sixty-four percent of responses believe that a lack of comprehensive, department-specific ML training is slowing organizational adoption of democratization workflows. The gaps in data literacy can be closed with comprehensive training and communication.
Both LOB and data respondents pointed to data analytics and IT as the business functions that would benefit most from increased ML democratization. The study also points to an increased focus on business intelligence (BI) and customer experience (CX) roles as important benefactors of ML democratization.
The respondents believed that the three most important ways to measure the success of ML democratization are to evaluate the increase in operational efficiency, increase in revenue, and improvement in the ability to make insight-driven decisions.
The Forrester study is a reminder that while business leaders are aware of the benefits of ML democratization, some serious concerns remain. The ML revolution is happening, and the companies that are able to make ML applications more accessible and easy to use are set to gain a significant competitive advantage.