CAI SIX PART WEBCAST SERIES | STEP 5

Enhancing Your Data & Analytics Environment with AI/ML

The explosive growth of data heightens both opportunity and risk.

Most organizations have far more data than they can process reliably without machine learning and artificial intelligence (AI), but 47% of executives report problems in integrating cognitive systems with their existing processes.

They know they must succeed. Global Solutions estimates AI will drive 95% of customer interactions by 2025. It’s no wonder that three in four C-level executives told Accenture they could go out of business if they can’t scale-up their AI capabilities in the next five years. The question is how to do that when talent is scarce, ethical issues abound, and technologies evolve faster than CIOs can implement them?

CAI, which has provided business technology services for four decades, examined this issue during its six-part series, “Activating Data & Analytics for Real Business Value Creation.” This summary highlights key points from the fifth session “Enhancing Your Data & Analytics Environment with AI and Machine Learning.”

The series features Steven Stone, CEO and Founder of NSU Technologies, and Tom Villani, Senior Vice President, Digital Innovation, at CAI. Stone, a former CIO at Lowe’s and L Brands, shares real-world examples and proven solutions to help companies accelerate their own transformations. We invite you to review past sessions and register for the sixth program for free at www.cai.io, and to contact Tom.Villani@cai.io to complete a complimentary Data & Analytics Readiness Assessment.

PART 1

Why AI and Why Now?

Analytics capabilities provide us with the real insight we need to guide our businesses. Artificial intelligence and machine learning supercharge that capability. They speed things up to enhance analytical results.

Modern graphical processors used to power AI and machine learning are exponentially faster than older computing platforms. As a result, AI can reveal patterns and trends that humans can’t even see.

As labor costs rise, businesses know they must increase productivity while improving both consistency and quality of business outcomes. Interest in AI has soared during the pandemic as companies embraced more automation as a way to offset lost revenue through greater productivity.

AI is a broad term that can be broken down into four categories: reactive machines act on what they see, like IBM’s chess-playing Deep Blue machine; limited memory machines look into the past for simple transient information to inform decisions, such as cars that use data from cars around them to control speed; theory of mind machines interact with emotion and thought, such as a robotic nurse; and finally, self-aware machines—which remain in the distant future—think and act on their own, like the malfunctioning Hal 9000 unit in “2001: A Space Odyssey.”

Similarly, there are four categories of machine learning: supervised systems that predict future outcomes based on patterns in data with help from an operator; unsupervised systems that can eliminate “noise” of incoming data and segment the data without the direct control of an operator; semi-supervised systems, which sit between supervised and unsupervised, use limited sets of data to train itself and can classify data into groups, such as with an MRI; and reinforced learning systems in which self-sustaining machines can consider the results of the initial action.

Amazon's current machine learning algorithm has decreased the "click-to-ship" time by 225%. (Forbes)

IT professionals find new ways daily to drive revenue and efficiencies using these technologies. In retail, for example, McKenzie recently estimated recommendation engines drive 35% of product sales for Amazon. In healthcare, AI has helped to augment telemedicine during the pandemic by analyzing a patient’s history and comparing it with incoming data from wearable IoT devices.

Manufacturers use robots and co-bots to reduce R&D costs in pharmaceutical firms while predictive analytics help improves maintenance and supply chain operations in other companies. Financial services firms can now make much faster decisions on loan applications for their customers.

PART 2

Preparing for AI

Enabling AI comes with prerequisites. There must be strong executive sponsorship and a clear strategy on a plan for AI and machine learning. Once in place, organizations can focus on collecting vast quantities of high-quality data, breaking down silos, and eliminating unreliable data that could lead to bad decisions.

Having the right talent is another necessity for the development and deployment of reliable AI processes. If the process is bad, AI will only produce bad results faster. The organization’s technology infrastructure must have the computing power to handle large volumes of data. And IT leaders must understand the impacts of bias, whether in the data or the algorithm, so ethical governance is something organizations must master before implementing AI.

61% of business executives with an innovation strategy say that using AI has identified opportunities in data that would otherwise be missed. (GlobeNewsWire)

To get started, organizations should identify problems that need to be solved, then align those with expected outcomes. Then they must assemble their internal staff and external experts and begin with a project that is important enough to drive engagement, but not so large that it becomes cumbersome. Teams should also understand the organizational impact of the project, such as the effects of automating a process that currently requires four people.

In other whitepapers, we covered the importance of adopting the right data architecture, and that’s even more critical when going deep and wide into AI. Beyond having a formalized data management program, organizations should address issues like data integrity, accuracy, security, lineage, privacy, and timeliness. The elastic nature of the cloud will help with AI’s demand for both data storage and computing power, so that is another key architectural consideration.

Finally, organizations should consider architectural choices between vertical AI solutions, such as SaaS apps that employ some level of AI and machine learning, with a horizontal layer that creates a centralized repository for all models that may be consumed by other applications often by using API’s. Building a horizontal layer is complicated by the abilities of applications to consume third party models. As such, most organizations will need to bring in experts in this field to build what they need.

The AI toolbox is large and growing, with a dizzying array of products. In the area of data preparation, for example, there are tools that can be used by themselves or integrated with solutions aimed at data integration, analytics/BI, or data science and machine learning.

Platforms like Microsoft Machine Learning Studio, Amazon AI Services, TensorFlow, IBM Watson Studio, and Google AI Platform will help data scientists build, train, and manage their models. Both the large platforms and some standalone tools offer what might be called consumer features like speech image processing and natural language processing.

PART 3

How to Succeed with AI

As with other areas of IT, culture plays a major role in creating a sustainable program, achieving business goals, and powering new models. AI can be scary on some levels of the organization, and even leaders may need to be trained on the business and ethical implications of AI. Communication with the appropriate people throughout the organization is critical to building an understanding of the desired outcomes.

Data management, and in particular data quality, will play a make-it-or-break-it role in any AI program. Incomplete or missing data, duplicate data, inconsistent use of metrics, unexpected formats, and even malign data can all undermine success. A good data quality program considers how data preparation software should handle each such exception.

Data governance and data architecture, which were covered in prior sessions, define responsibilities for the required levels of data quality and accessibility. Many data preparation and integration tools use AI or machine learning algorithms to help improve data quality. But it’s not all about quality. No AI program can succeed unless storage capacities are adequate to handle the large quantities of data needed to effectively train models.

About 80% of companies are investing in AI and are facing challenges establishing the framework need to implement a useful AI application. (Harvard Business Review)

A key to any technology initiative is a focus on value. Organizations should spend time up-front to set up KPIs that measure specific outcomes, such as sale increases related to AI. They also need to learn proxy metrics that measure things like functions and accuracy. Finally, there’s a need for Objective and Key Results (OKRs) that are performance-based metrics to measure the effectiveness of an AI/ML objective.

We mentioned staffing needs, but keep in mind there are new roles to consider with AI. These include subject matter experts who help define objectives and use of AI solutions. Data engineers help to integrate data into AI. Meanwhile, data scientists build algorithms that explore data and extract the information needed to make decisions.

Statisticians help decision-makers draw conclusions on the viability of some AI recommendations, while ethicists look at the proper use of data and indications of bias that may harm the results. Because many AI solutions will be public-facing, a UX designer is also needed.

While organizations face huge risks by ignoring AI, they also face risks in implementing it. For example, if there is a bias built into the AI or a model was trained improperly, it could create significant risks for a financial institution. A key to minimizing risk is with a governance framework that specifies processes and roles for managing risk during deployment. An AI ethicist should be brought in to help risk professionals understand the complexity and inherent risks of AI solutions, and to build transparency into AI solutions.

The road to success requires consistent and constant oversight throughout development and deployment. But it leads to the creation of real business value through better analytics.

Have additional questions?

We'd love to talk.

Get in Touch

Webcast

STEP 5

Enhancing Your Data & Analytics Environment with AI/ML

Watch now