Enterprise AI: Separating reality from myth

By | 6 minute read | October 7, 2020

With more companies moving to digital delivery and operational methods, artificial intelligence has received a lot of attention and hype in recent months. With businesses under immense pressure to reduce operational costs, meet increased customer demands, or simply to build up their general resilience through technology, many organizations have moved AI higher on their priority list. However, the hype has brought with it a slew of purported enterprise AI platforms offering the crucial transformative powers of AI during these uncertain times.

In the recent C3.ai webinar “Enterprise AI: Separating Reality from Hype,” Forrester Research’s vice president and principal analyst Mike Gualtieri says that 73% of companies surveyed reported a positive impact from AI adoption. By 2025, he says, nearly all organizations will have deployed at least one AI-related use case. Enterprises waiting to fully adopt AI, he warns, may find themselves at significant disadvantage in their ability to provide personalized customer experiences and proactively prevent problems.

Gualtieri says enterprises are moving forward with AI because it can perform highly repeatable processes for four primary types of decisions — real-time, operational, tactical, and strategic.

“AI can squarely be used to augment human intelligence through analytics, but when you do AI for real-time and operational decisions, you do it for repeatable decisions, and that’s where most of the value comes from,” Gualtieri says.

With the increased attention comes hype, generating misconceptions and myths about AI that can trip up its adoption. Apply some healthy skepticism to ensure that your business maximizes the full value of AI. Here at IBM, we have identified five common AI myths that often get in the way of a successful AI deployment.

Myth 1: Artificial intelligence is a tool that enterprises add into their existing processes and systems.

Reality: Enterprises must develop new processes and recreate existing ones to realize the full benefits of AI.

When implementing AI, the technology is integrated with existing systems and data to remove organizational silos. By being able to merge what is often thousands of systems and millions of data records into a single view, artificial intelligence allows enterprises to make decisions based on a complete overview of all data instead of fragmented sections of the picture. But organizations only see a small portion of the possible benefits if they simply add AI on top of existing processes.

“This allows you to rethink the core and key processes of the business, how you engage customers, and how you manage your operations all the way from customer delivery through your suppliers and supply chain,” says C3.ai President and Chief Technical Officer Ed Abbo.

Enterprises that fail to revise their business processes miss these new opportunities to engage with customers and prevent future problems, especially with the new availability of real-time data and significantly expanded geographic reach. AI provides enterprises with previously unavailable knowledge, but they must work actively to determine how best to use this knowledge. While enterprises can keep their existing business processes and put a new digital layer on top of them to operate more effectively, reinventing processes with AI woven in offers much greater benefits.

Myth 2: It is best to implement multiple AI projects from the start.

Reality: Starting with a production pilot provides the most effective results and creates an opportunity to develop effective change management.

When deciding where to begin, determine which project is most self-contained while also providing significant value, such as increasing revenue, improving customer service, or eliminating a pain point. Consider projects with a significant amount of historic data upon which to build AI models. Abbo says a bank may decide to start using AI to detect money laundering activity and intervene with suspected fraud in real time.

Tony Giordano, IBM vice president and senior partner for data platform services suggests that organizations address all current likely use cases when planning out implementation, with an eye toward what may be needed in the future.

“Make sure as you’re planning out your implementation that you address all the use cases that your organizations might need,” Giordano says. “At IBM we ask: What are all the use cases for your data? That helps architect and engineer your enterprise AI environment.”

After selecting an initial use case, Abbo recommends pursuing an eight- to 16-week production pilot. During the pilot, the enterprise aggregates existing data using a machine learning tool that then can be applied to create a pre-production application that is operated against historic datasets. The production pilot shows the effectiveness of the algorithm against the hold-out period. For example, applying the anti-money-laundering application against already identified money laundering trains the application on a prior period and creates a prebuilt application with proven economic value. After the application moves into production, Abbo recommends moving on to additional use cases, continuing to expand after each successful implementation.

Myth 3: You only need to provide data required for the AI project.

Reality: Lay out all related data before beginning.

Many organizations used to throw all data into a Hadoop cluster, which was expensive and hard to manage, Giordano says. While the intent was to have data for additional use cases down the road, a properly established enterprise AI environment can easily scale with additional data as the need arises.

“One of the mistakes that people would make in the data lake era was to throw everything onto a Hadoop cluster, which is not only expensive but it’s hard to manage what is out there and what it’s used for,” Giordano says. “One of the great things about a proper enterprise AI environment is that it’s always scalable.”

It’s most cost-effective to bring data into the AI platform as it’s needed, he notes. Giordano often sees enterprises approach data design either through over-engineering or not having a data model at all. Enterprises begin to create data governance through mapping different types of data based on the subject areas.

“Information governance is one of the key success factors,” Giordano says. “It must be an active thread in the development methodology, not a QA function.”

Myth 4: Choosing a platform that supports open source is key.

Reality: Determine how quickly a platform updates new open source development before purchasing.

While enterprises should choose a platform that supports open source, leaders must go a step further and ask how quickly they make new open source code available to the platform. Most important innovations in AI come through open source coding. Enterprises that choose platforms with significant open source update delays often lag behind competitors who can access the features weeks or even months earlier.

Additionally, enterprises must consider tools designed for business users and citizen developers, which open up opportunities for more use cases. Because machine learning models rely on business process applications for their intelligence, AI quality depends on the ease of use of tools to implement more use cases. While enterprises must consider the coding needs of data scientists, experts can be very productive with both visual tools and AutoML, which is available in open source. Giordano recommends considering the platform’s integration methods, which provide the ability to integrate machine learning into production applications.

Myth 5: Data governance must involve setting up a large data council.

Reality: Build data governance into development.

Instead of focusing on councils, enterprises should build information governance directly into the development process. Giordano recommends assigning data stewards to work with the team that created the enterprise semantic models so that the data can be mapped to specific business calculations. By taking this approach, enterprises no longer have to conduct expensive renovations of data systems, he says. In addition to saving money, enterprises that integrate data governance into development speed the AI deployment process and time to market.

“It’s the ounce of prevention versus the pound of cure,” Giordano explains. “Threading data governance into your development capability is very cost-effective and helps speed up projects.”

Moving to AI requires creating new processes and methods for managing data. Understanding the best practices and approaches for moving to AI lets your enterprise fully realize its productivity, revenue, and customer satisfaction benefits.

Learn more about AI and how to get started

Most Popular Articles
?>