Operationalizing AI Beyond the Labs

Image credit: iStockphoto/metamorworks

It is no surprise the Cognizant whitepaper “How Companies Can Move AI From Labs To the Business Core” details an uptick in AI startup activity.

AI spending in the Asia Pacific and the Middle East was already set to reach USD 15.06 billion by 2022 before the pandemic. U.S. giants like Google and Microsoft were also investing in AI for the region. Post-pandemic, companies are waking up to the real benefits that this emerging technology offers.

The reality is more sobering. The whitepaper notes a widening gap between enthusiasm and actual deployments. And there are several reasons why companies face roadblocks when moving their AI lab projects into the production environment.

According to Kaustubh Laturkar, assistant vice president at Cognizant Technology Solutions, those who can bridge this growing gap will be tomorrow’s winners.

Wrong questions

It is natural for companies to start with a proof-of-concept (PoC). But when more than 60-70% of AI PoCs do not get implemented in production, you know something is amiss.

Among the industries, banking and financial services (BFS) companies with the largest AI appetites are only marginally ahead of other sectors (31%). Retail and technology companies, which have much to gain with AI in terms of customer experience, have the most significant number of projects stuck in the PoC stage.

Laturkar suggests the first-most challenge is selecting the right use case. “The art and science of selecting the right use case are extremely important for AI,” he says. He adds that most companies tend to choose use cases where AI's advantages are unclear or not significant enough for further investment.

AI is here to augment human beings’ decision-making capabilities; hence, Assisted AI becomes significant for enterprises.

Being conservative on the outcome or too narrow a use case can also be an issue. While the AI may succeed in the PoC stage, scaling it across the organization for other scenarios and different outcomes can be problematic.

When choosing the right use case, Laturkar suggests targeting rule-based applications. He offers an example where one of the largest retailers in Australia was struggling with gift-card fraud. They were using over 1,000 static rules, built over 20 years. 

“We recommended them to pick this as a use case and helped them to do a four-week PoC on AI and machine learning. The PoC delivered the initial results, and now they are looking to scale it into production with more predictive models,” he explains.

Hear Kaustabh Laturkar explaining why we are at the threshold of broader AI use cases and why operationalizing AI will be critical.

Data fitness challenges

Another major problem is that companies are rushing into AI projects without sorting out their data first. Laturkar explains that AI’s strength lies in analyzing multiple data sets quickly and offering insights. This requires companies to integrate with different data sources, prepare the data quickly and have a common data management platform.

“Does the company having a clear-cut vision in terms of bringing data together on a common platform? Secondly, does your data fit the purpose or business? And lastly, is your data ready for AI/ML models? These are some of the key considerations,” he points out.

One approach is to quantify the business value of data. Cognizant has a unique method to do this called DataIQ.

“DataIQ is how fit your data is business-wise to get used for any type of decision purpose. It is an unsupervised machine learning-driven algorithm that we have built internally,” Laturkar explains.

You can identify existing and missing data sets critical to achieving the business objectives with a DataIQ analysis. It helps companies score and value data asset relevance and intelligence, allowing them to focus on data engineering efforts to deliver the desired business outcomes.

The higher the DataIQ score, the more “fit” the data for decision making. It allows you to identify relevant information, reduce noise and redundancy, and understand the data set's complexity. In turn, AI algorithms can use the high-scored data sets for descriptive, diagnostic, predictive, and prescriptive analysis.

A low DataIQ should not discourage companies, says Laturkar. He notes that it is a journey, and companies can adopt best practices to improve their scores. 

Culture shock, talent crunch, and elusive ethics

The enthusiasm may be overwhelming, but the whitepaper notes how executives are still finding it challenging to secure senior management commitment and business buy-in.

Although AI can also create new jobs and widen current job scopes, some of the fears are related to job protection. Essentially, it boils down to having the right mindset and culture. The low number of AI projects reaching the implementation stage is a result of the current risk aversion.

We humans always have this status quo bias. But to drive AI projects into production and ensure high adoption, companies need to create a culture that is open to AI and welcome it,” says Laturkar.

Lack of talent is a perennial problem. Already data science projects are suffering from the difficulty in finding data scientists. Hiring AI engineers and MLOps experts will be even more difficult, as it requires knowledge of both IT and business to frame the right questions and use case.

MLOps is an important and emerging space critical for digital business transformation and addresses how best to deploy and manage AI/ML models. For example, when deploying models on mobile phones, the cloud, or on-premises, you need to cater to each platform’s nuances.

At the same time, the number of job roles for AI is expanding. Cognizant’s Center for the Future of Work (CFoW) identified 42 jobs that will emerge over the next ten years that will depend on mastery of new technologies, including AI. Some of these include machine risk officers, cyber calamity forecasters, and heads of business behavior.

For many companies, hiring these AI talents is no longer an option. It can be too lengthy and expensive. Laturkar suggests companies look at retraining. It also offers the added advantage of creating a robust talent pipeline that can serve as a critical business differentiator.

Lastly, companies need to see AI and ethics as two sides of the same coin. With algorithms helping decision-makers with recommendations and new insights, user and AI model trainers need to root out bias and unethical use.

Yet, many companies struggling in this department. In fact, many are willing to take the risk of reprisals.

The whitepaper notes that 74% of companies rely on customer feedback to address ethical concerns after an AI app launch. Only 58% have relevant policies in place to address post-launch ethical concerns.

Evolutionary AI

One challenge with machine learning is that it takes time — time that companies do not have in today’s fast-paced business environment. Also, desired outcomes can alter and drastically impact AI project outcomes.

Cognizant is taking a different approach with its proprietary Cognizant Evolutionary AI. Based on its patented Learning Evolutionary Algorithm Framework (LEAF), it uses advanced evolutionary algorithms and deep learning to rapidly produce results from “complicated, multivariate problems.”

Cognizant Evolutionary AI also enables transfer learning “where we reuse a model already pre-trained on a gigantic publicly available data,” says Laturkar. The company helped a retail cosmetics company allow potential customers to view the results before their Botox treatments and cosmetic surgeries.

The same approach is now used in COVID-19 and treating diseases “where there is no historical data available,” points out Laturkar. So, Cognizant reused models “for around 14 diseases” and use transfer learning to create a new model for COVID-19. The results helped researchers quickly determine how the disease spreads and what they can do to prevent it.

Overall, Cognizant Evolutionary AI speeds up machine learning and allows companies to reap the rewards a lot earlier. After all, in business, time is always money.

Looking to gain deeper insights into best practices for operationalizing and scaling AI initiatives? Read the featured whitepaper “How Companies Can Move AI From Labs To the Business Core.” It is available for download here


Winston Thomas is the editor-in-chief of CDOTrends and HR&DigitalTrends. He is always curious about all things digital, including new digital business models, the widening impact of AI/ML, unproven singularity theories, proven data science success stories, lurking cybersecurity dangers, and reimagining the digital experience. You can reach him at [email protected]

Image credit: iStockphoto/metamorworks