Three cloud ML approaches for enterprise AI strategy

A blog post by Diana Kearns-Manolatos, senior manager, and Jonathan Holdowsky, senior manager, Center for Integrated Research at Deloitte.

Artificial intelligence (AI) and machine learning (ML) are helping global organizations solve numerous business challenges, ranging from operational efficiency and risk management to customer experience and more. And, while ML is powerful in its own right, when combined with cloud technologies, it allows organizations to better innovate at speed and scale. 

Deloitte’s 2020 State of AI Study reveals that 83% of organizations expect AI to be critical to their business’s success in the next two years. In fact, our recent analysis of the survey found that cloud drives measurable benefits for AI programs, improving processing efficiency (49% versus 42% overall), decision-making (45% versus 39% overall), and competitive advantage (39% versus 26% overall).

Why is that? The answer goes beyond the well-known data scalability and infrastructure elasticity that provide a resilient “scale up and scale down” option for organizations, though that is part of it. Cloud ML brings to bear additional ML services—pretrained models; vision, speech, and video APIs; frameworks; and general-purpose algorithms (fraud detection, inventory management, and others)—as accelerators for developer and data scientist teams.

As organizations look to explore cloud ML, there are three approaches to consider:

• Approach 1: Cloud AI platforms. This model allows developers and data scientists to use cloud-enabled tools for training, deployment, and model management by bringing new and existing ML models into the cloud. The benefit of this approach is that it helps manage the cost of spot training, gain greater elasticity to scale infrastructure needs, and improve orchestration across the model management life cycle.
• Approach 2: Cloud ML services – Pretrained models and more. This model allows organizations to access pretrained models, frameworks, and general-purpose algorithms (vision, speech, language, and video APIs; virtual assistants and bot frameworks; and algorithms for fraud detection, inventory management, and call centers) to accelerate development of new business applications without the need for a large data science team to build the underlying model.
• Approach 3: AutoML – Customized pretrained model. This model allows organizations to create their own custom models (i.e., off-the-shelf versions of pretrained API models) with proprietary data, allowing for ongoing model refinement that, in some cases, may even include a low-code or no-code environment.

And, by virtue of being in the cloud, organizations can achieve downstream benefits across the model development, training, deployment, management, and monitoring life cycle with an enterprise infrastructure for deployed MLOps, including model versioning, autoscaling, continuous monitoring and training, and retraining and deployment.

Organizations across industries are having tremendous success driving efficiency (time); tapping into the power of pretrained models (technologies); and extending and augmenting their developer, data engineer, and data scientist teams (talent). For example, one financial services organization used cloud ML to power personalized recommendations for high-net-worth and ultra-high-net-worth clients. Another technology and media organization used cloud ML to empower contact centers with virtual agents and conversational AI to engage users, resolve and escalate issues, and guide cross-sell and upsell opportunities. In another example, a government organization used cloud ML to capture and analyze drone data to monitor equipment and inform predictive models.

While the potential of these two technologies coming together creates strategic opportunity, perhaps equally as important is the human impact. Cloud ML minimizes organizations’ dependency on having a large team of data scientist PhDs to advance ML programs and empowers data engineers and developers to create new, innovative products and services.

Given the time, technology, and talent barriers broken down by cloud ML, it is no wonder estimates expect today’s $2–5 billion cloud ML market to reach $13 billion by 2025.

As AI programs look to innovate new solutions, expand training data sets, and scale solutions across the enterprise, key considerations like strategic business alignment, talent orchestration, infrastructure finance models, data requirements, and innovation culture needs should all be factored in to take advantage of the cloud ML opportunity.

For more on this topic, check out our recent research, Innovating R&D with the cloud and Time, technology, talent: The three-pronged promise of cloud ML.