Scaling AI/ML Initiatives: The Critical Role of Data 

  Sponsored by: Snowflake Ritu Jyoti

INTRODUCTION 

Artificial intelligence (AI) is the most disruptive technology innovation of our lifetime. Originally a discipline limited to academic circles, AI is now commercially mainstream. Enterprises are embracing AI/machine learning (ML) and leveraging a variety of data types (structured, unstructured, and semistructured) in all lines of business (LOBs) and industries. For example: 

• Revenue growth: 
  • Marketing and sales teams use AI to better target prospective customers, optimize outreach campaigns, and prioritize leads
    • AI technologies enable social media sentiment mining, programmatic selection of advertising properties, measuring the effectiveness of marketing programs, ensuring customer loyalty, and intelligent sales recommendations.  
•  Cost/operational efficiency: 
  •  AI-powered contact center solutions accelerate time to resolution and improve the customer experience (CX). Natural language processing enables customers to speak easily about what they need without navigating through a frustrating phone tree. 
•   Deep learning (DL) algorithms accelerate
•   Industrial IoT models can now predict when a machine will break down and recommend preventive maintenance, thus avoiding any potential downtime. 
•  Risk mitigation: 
  • Financial institutions improve loan underwriting and reduce risk. AI can also help lessen financial crime through advanced fraud detection and spotting anomalous activity.  
  •  AI is playing a critical role in cybersecurity today. By improving organizations’ ability to anticipate and thwart breaches, protecting the proliferating number of threat surfaces with zero trust security frameworks, and making passwords obsolete, AI is essential to securing the perimeters of any business. 

AI initiatives offer more than just cost savings. According to IDC’s AI StrategiesView 2021 Survey — a global survey of 2,000 organizations, with IT and LOB decision makers and influencers as respondents 

• AI disrupters (organizations that are repeatedly creating new business value and sustainable competitive advantage from AI) report 39% improvement in customer experience and 33% improvement in employee efficiency and accelerated innovation with the rollout of AI solutions in 2020. This is a double-digit surge in the improvement of business outcomes compared with 2019. There is a direct correlation between AI adoption maturity and superior business outcomes. As such, there is a heightened need for enterprises to strategically scale their AI/ML initiatives.  

Data is essential for AI/ML initiatives: 

• AI/ML requires vast volumes of data to train models.  
• Ensuring unbiased results requires diverse data sets.  
• Models must be continuously trained with latest information to maintain predictive performance, particularly in dynamic business environments.  

By taking advantage of consolidated data architecture, successful AI disrupters have been able to exploit the power of different data types and the associated ecosystem to drive innovation and transformation. 

SCALING AI: INHIBITORS/CHALLENGES/NEEDS 

Although many organizations understand the importance of AI and its potential impact on their business, they often struggle to move from pilot to production. As per IDC’s AI StrategiesView 2021 Survey, the main challenges to implementing AI solutions include, in order of importance. 

•  Costs (i.e., hardware accelerators and compute resources) 
• Lack of skilled personnel (i.e., talent)  
• Lack of machine learning operations tools and technologies 
• Lack of adequate volume and quality of data  
• Trust and governance issues  

                      Not only is data at the core of AI, it is also a key challenge. Over half of the organizations report that they lack the volumes and quality of data needed to implement an AI solution. But it’s not only a question of supply. 

In the model development stage, one of the biggest challenges for businesses is getting data into the platform (see Figure 2). This can be difficult, especially if the data is not readily available in the right format. 

Providing a single source of truth of governed, high-quality data is beneficial not only for data scientists but also for analysts and other data teams. To have an AI application that is relevant, accurate, and scalable, businesses need to make sure that their data, both real time or batch, is of high quality and easy to access and share securely within the organization and with the organization’s network of business partners. 

To harness the full power of data with AI/ML, data scientists and machine learning engineers need the latest software frameworks and programming languages. General frameworks such as TensorFlow, MXNet, Caffe, scikit-learn, Keras, and PyTorch are necessary, as well as more specialized programming languages such as Python, Java, and R. 

However, having the right technology is not enough. Machine learning models need the most relevant data, which may not always be inside the organization. Internal data only allows companies to see their own operations or customer information. That doesn’t provide a complete picture. Companies need access to secure data sharing. 

Data may be coming in real time, and so it is important to harness that data as well for real-time predictions in use cases such as fraud detection or product recommendations. 

As users and use cases proliferate, machine learning–powered applications must be able to handle the extra load. If the application fails to scale, performance bottlenecks can diminish the value of using AI/ML. For example, if a customer doesn’t get the product or service recommendations they want in a timely manner, they may be less likely to come back. 

Although developing a scalable system can be difficult, it is critical to do so to handle increased business demand. Failing to scale the system can lead to lost business and missed revenue opportunities. For example, delays can result in abandoned shopping carts or failure to make recommendations in a timely manner. When scaling your system, organizations should be prepared for potential technical problems like infrastructure optimization (processing performance and elasticity), interoperability (such as supported programming languages and ML frameworks), and machine learning operations integration with existing DevOps tools and practices. 

By building on top of elastic, intelligent infrastructure that requires near-zero management, organizations can more efficiently handle large volumes of data and process data without bottlenecks, regardless of number of users, in a cost-effective and time-saving way. There are several inherent benefits to this approach, including: 

• Organizations can be more agile and creative by having a pipeline that allows for fast executions of every stage (data preparation, experimentation, model training, and deployment). 
•  
• It never hurts to optimize for costs and value. Scaling helps optimally utilize available resources and makes a trade-off between marginal cost and accuracy. Streamlined architecture that supports multiple personas eliminates the need for redundant systems. 
• The pipeline should be as automated as possible so that data science professionals can focus on more complex tasks (e.g., generating value from data rather than building integrations or managing infrastructure, thereby creating a faster path to production deployments). 

Effective AI requires data diversity. Similarly, the full transformative impact of AI can be realized by using a wide range of data types. Adding layers of data can improve accuracy of models and the eventual impact of applications. For example, a consumer’s basic demographic data provides a rough sketch of that person. If you add more context such as marital status, education, employment, income, and preferences like music and food choices, a more complete picture starts to form. With additional insights from recent purchases, current location, and other life events, the portrait really comes to life. 

Source: IDC, 2022 

Unfortunately, due to these challenges, organizations are spending more time on tasks that are not actual data science. For example, IDC’s AI StrategiesView 2021 Survey found that the organizations spend the largest percentage (21%) of their total time in AI/ML life cycle in data collection/preparation (see Figure 5). The personas involved and surveyed include the data scientists, data architects, data engineers, machine learning engineers, application developers, and operations staff. 

Snowflake’s Data Cloud allows organizations to consolidate multiple data types and structures from many sources into a single source of truth. This consolidation makes it easier for everyone involved in the AI/ML life cycle — from data preparation to model building to application deployment (see Figure 6) — to share data and collaborate effectively to derive valuable insights with speed. 

RECOMMENDATIONS 

Today, enterprises are confronted with a complicated set of business challenges, including an increasing pace of business, an expanding volume of business data, the need to think about shared data strategies to truly derive value from data, a growing scope of global commerce, and a multitude of risks for customers, employees, and suppliers. The volume of customers and suppliers, along with regulatory complexity and multi-industry businesses, means complexity is common in global businesses. 

Enterprises are rationalizing, modernizing, and transforming their enterprise application portfolios. Machine learning, natural language processing, assistive user interfaces, and advanced analytics coupled with curated data sets are advancing traditional applications to become intelligent. 

These intelligent applications enable more employee insights by automating transactions that were previously stalled and bringing more data into the equation so organizations can make better decisions immediately. Organizations need a data strategy for AI, which will vary greatly depending on the size, nature, and complexity of their business and AI strategy. To accelerate innovation and time to value and enjoy a sustainable competitive advantage, technology buyers are advised to: 

•  Build a talent pool of industry domain and technical experts like data engineers, data scientists, and machine learning engineers. 
•  Get employee buy-in and trust for the data strategy with inclusivity and transparency. 
•  Create a workflow for bringing in third-party and/or net-new data sources into the organization, including testing, buying, and seamless integration with existing internal data sets and processes. 
•  Ensure the process is cross-functional across IT, procurement, legal, compliance, and security. 
•  Select a secure and governed data platform with support for all data types to support the entire AI/ML life-cycle workflow. 
•  Ensure flexibility in programming with support for multiple programming languages like Python, Java, and Scala, as well as leading machine learning workflows like TensorFlow, PyTorch, and scikit-learn. 
•  Embrace an intelligent data grid that helps: 
  •  Automate and enforce universal data and usage policies across multicloud ecosystems. 
•   Automate how data is discovered, cataloged, and enriched for users. 
  •  Automate how to access, update, and unify data spread across distributed data and cloud landscapes without the need of doing any data movement or replication. 

CONCLUSION 

Many companies adopt AI as they undergo digital transformation — not just because they can, but because they must. AI is the technology helping businesses be agile, innovative, and scalable. 

Successful enterprises will become “AI first” organizations able to synthesize information (i.e., use AI to convert data into information and then into knowledge), learn (i.e., use AI to understand relationships between knowledge and apply learning to business problems), and deliver insights at scale (i.e., use AI to support decisions and automation). AI is becoming ubiquitous across all the functional areas of a business. IDC forecasts the overall AI software market will approach $596 billion in revenue by 2025, growing at a CAGR of 17.7%. 

Data is the heart of AI initiatives. Organizations need to strengthen their data strategy for AI and adopt a secure, governed, collaborative, and scalable data platform that helps data science professionals focus on data science and scale AI initiatives seamlessly. 

Category: Uncategorized