The 3-Pronged Promise of Cloud ML

When organizations bring machine learning efforts to the cloud, the benefits can include speed, productivity, and technology. The key is to start with a business case, assess data and model requirements, invest in the right talent, and adopt a ‘fail fast’ mindset.  Machine learning (ML) is transforming nearly every industry, helping companies to gain efficiencies, innovate rapidly, and meet customer needs at an unprecedented pace and scale.

Source: The 3-Pronged Promise of Cloud ML

As organizations mature in their use of the technology, growing reliance on large data sets and the need for fast, reliable processing power are increasingly driving such efforts into the cloud.

Many organizations already tap the cloud for access to modular, elastic, rapidly deployable, and scalable infrastructure without heavy upfront investment. In addition, cloud ML introduces cutting-edge technologies, services, and platforms—including pretrained models and accelerators—that provide more options for how data science and engineering teams can collaborate to bring models from the lab to the enterprise. By using ML in the cloud, companies can gain faster insights into changing market conditions, immediate access to the most current ML technologies, and productivity tools that can drive talent efficiency. In that way, time, technology, and talent—factors that often serve as challenges to scaling enterprise technology—comprise the three-pronged promise of cloud ML.

In a recent study, Deloitte conducted interviews with nine leaders in cloud and AI to explore more fully the potential cloud ML represents for organizations looking to increase innovation.

Measurable Value

In Deloitte’s 2020 “State of AI in the Enterprise” survey, 83% of respondents said that AI will be critically or very important to their organizations’ success within the next two years. About two-thirds of respondents said they currently use ML as part of their AI initiatives, and most of these AI/ML programs currently use or will use cloud infrastructure in some form.

Additional analysis of the survey data shows that cloud ML, specifically, drives measurable benefits for the AI program and generally improves outcomes as compared with nonspecific AI deployments.

“AI is a tool to help solve a problem,” says Rajen Sheth, vice president of AI for Google Cloud. “In the past, people were doing a lot of cool things with AI that didn’t solve an urgent problem. As a result, their projects stopped at the concept stage. We now start with a business problem and then figure out how AI solves it to create real value.”

3 Fundamental Approaches

As organizations look to advance their ML programs supported by the cloud, there are three basic approaches they can take, depending on the form they embrace:

• Cloud AI platforms, which allow developers and data scientists to use cloud-enabled tools for training, deployment, and model management.
• Cloud ML services, which offer access to pretrained models, frameworks, and general-purpose algorithms.
• AutoML, which allows companies to customize off-the-shelf versions of pretrained API models with proprietary data.

There is no one-size-fits-all approach. “At the end of the day, it comes down to where you can put ML and AI into a business process that yields a better result than what you are currently doing,” says Gordon Heinrich, senior solutions architect at Amazon Web Services.

Building a model in-house requires a significant investment of time in data collection, feature engineering, and model development. Depending on the specific use case and its strategic importance, companies can employ cloud ML services, including pretrained models, to help speed time to innovation and deliver results. For example, conversational AI is one use case where cloud ML and pretrained models seem particularly well positioned to help solve business, technology, and talent challenges.

Key Considerations

For general cloud ML use cases, companies should think through business, technology, and talent considerations to determine the right approach:

Business context: Define the problem statement. The most effective applications of cloud ML typically start with identifying the broader business problem AI is suited to help solve—whether it be scaling the digital business, enhancing customer experience, or addressing risk management challenges. Executives should then examine the business problem from both an industrywide and a company-specific perspective.

Technology context: Understand which data and models you have and where they reside. Consider how long it will take to get access to the data and to develop and test the models as it relates to time, resources, risks, and deployment needs. Pretrained models or AutoML solutions may offer time and resource savings for initiatives that are less strategically differentiating for your business. Also consider program goals and outcomes. Starting small allows teams adequate time to test and refine models and to operationalize and scale them over time. Think about easy-to-achieve targets, mid- to longer-term business outcomes, and value.

Talent context: Assess data, ML, cloud, and data science expertise across your team. Review what talent you have and what talent you need to achieve your stated goals. Assess areas where cloud ML services could save training time, optimize the potential of your existing talent, or fill in resourcing gaps.

“When talking about talent in the AI/ML space, think of a staircase with four steps,” suggests Eduardo Kassner, chief technology and innovation officer for One Commercial Partner Group at Microsoft. “The first step is the developers who leverage AI apps and agents for sentiment analysis, personalization, robotic process automation, or object detection. The second step is developers who leverage cognitive services via pre-built models in scenarios such as language, speech, web search, or custom vision. The third step in the talent staircase is developers and data scientists who develop and maintain capabilities that include knowledge mining, which require data science and big data skills, in scenarios like audit and compliance, digital asset management, or customer feedback analytics. The fourth step is where skills in machine learning and algorithm development and maintenance are needed in scenarios like predictive maintenance, inventory management, fraud detection, and intelligent recommendations.”

Maintaining Quality, Efficiency

To ensure the integrity of a machine learning model throughout the life cycle, many organizations are adopting MLOps, or machine learning operations. MLOps refers to the steps an organization takes as it develops, tests, deploys, and monitors cloud ML models—in effect, a set of governance practices. As development operations become increasingly automated because of intensifying pressure for frequent model releases, these steps take on greater importance and urgency, especially as companies become more cloud-focused in their infrastructure.

MLOps “is about explainability and getting a notification when a model is starting to drift,” Heinrich notes. “It is about making the data pipeline easier to use with data tagging and humans in the loop to achieve high-quality results. It really helps businesses feel more confident about putting algorithms into production.”

Model bias is a concern in any broader ML governance program, but especially for companies using pretrained models, which typically offer limited visibility into how models were trained and thus how much bias, if any, may be a factor. Under such circumstances, testing for bias after model deployment requires heightened scrutiny using a variety of known data sets so that the results can be detected based on their output. (For more information on AI ethics and avoiding bias, see Deloitte’s Original Postages/deloitte-analytics/solutions/ethics-of-ai-framework.html?id=us:2el:3dp:wsjspon:awa:WSJCIO:2021:WSJFY21" target="_blank" rel="noreferrer noopener">Trustworthy AI framework.)

“There are tools that help explain why the model gave a particular outcome, where there are biases and where there aren’t, and what impact such biases will have on people in general,” Sheth explains. “Organizations will have to dig deep to understand what is happening and supplement the data accordingly. It’s a painstaking process.”

*****

Cloud ML can be transformational, and there is strong evidence that organizations are already reaping its benefits in wide-ranging initiatives across industries, use cases, and archetypes. By putting careful thought into some key considerations and gaining a thorough understanding of what they want to achieve, organizations can reap the three-pronged promise of cloud ML, extending the reach of critical ML talent and technology investments and cutting time to innovation.

—by Ashwin Patil, managing director, Deloitte Consulting LLP; and Jonathan Holdowsky and Diana Kearns-Manolatos, senior managers, Deloitte Services LP[wsj-responsive-related-content id=”0″]

Related Content

• Time, technology, talent: The three-pronged promise of cloud ML
• The AI Era Is Here. Is Your Data Governance Ready?
• The Complex Truth About AI Computing and Value
• It’s AI’s Turn for the DevOps Treatment