By Sanjiv Gossain
July 8, 2021
In the near future artificial intelligence is going to be as fundamental for business success as cloud is becoming to running a company’s information technology. After many years of gradual growth, we’re now finally seeing cloud take off. And AI is where cloud was five years ago. People are realizing there’s value here, and business leaders see that AI can help them fundamentally change how their company works, how they get work done, and how they serve customers. In a recent survey of 1200 companies, Cognizant and ESI ThoughtLab found that 64% of executives believe AI will be important to the future of their business. For the largest organizations, the figure was a stunning 85 percent!
But while AI is an important set of technologies, and the algorithms are very important and at its heart, what really matters is how it impacts business and business outcomes. The technology a means to an end. And AI is going to change how companies buy, sell their own products and services, and work with other companies and indeed their entire set of stakeholders. It’s going to impact the way decisions are made, and all aspects of work life, as well as many people’s personal lives. It will help define the next era of business.
A huge part of the benefit from AI will come because businesses have assembled, maintained, optimized, and updated the right data. They’ll get that data both from inside their business but also by bringing in and leveraging data from outside their business (social, demographic, geospatial, and many other forms of data). Doing that will become, from now on, an ongoing and never-ending activity. The only way a company can build an effective model for an AI program is with well-defined data sets, used thoughtfully and deliberately and built upon continually.
Almost every company has in the past siloed much of its data in diverse and often unconnected systems and storehouses. So to understand what’s happening inside an organization will almost always require breaking down some of those barriers. The study found that companies spend about 35% of their AI budgets on “data modernization.” And in addition to combining once-disparate information, more and more projects will also need to go further and integrate third-party data sources. One Cognizant client in retailing is using machine learning to help predict ongoing covid infection rates and regional weather conditions, along with future sales and demand trends. To achieve that, the company is combining aggregate medical data from leading healthcare providers with a variety of sources of its own internal data, like same-store sales.
And inevitably, companies will increasingly add more subtle forms of data into their models–information that may be psychographic, geospatial, or real-time. And almost all AI programs going forward will include “the voice of the customer” by factoring in data from social media and call centers.
But getting genuine and useful results from an AI project is not easy. The technologies themselves are commoditizing quickly. A competitive edge will come from having the very best programmers, data analysts, and other specialists. Cognizant finds that about 40% of organizations that make investments in AI don’t reap any gains from it, at least at first. And about 70% of initial projects require significant rework before real advantage can come from them. Our research found that it takes companies on average 17 months to achieve a positive return on their AI investment, and the overall average return on such investments is only 1.3 percent. So while lots of companies are investing in AI, mostly they’re still doing “proof of concepts”. And many are not yet getting much benefit or return.
At Cognizant, we help clients derive value from their AI investments wherever they are on their AI journey. When they are just starting out, we work with them to identify scenarios where AI can make an impact. We rapidly develop a proof of concept, build a hypothesis around what an algorithmic model could be, and in the process translate the data problem into a business problem. We put models to place and work to prove them out.
If a client is further along, we work with them to start getting benefits on what I like to call enterprise scale. Once they actually start getting value out of their data, after they’ve “modernized” their data sources, the journey really begins. Then they typically begin to see a path to much more benefit from AI. It’s critical to use the right data to validate and “train” an AI model. But once the model gets going its impact can be powerful. It’s important to note that a model isn’t just created once–the data that drives the algorithm must be regularly cleansed and updated to ensure it is relevant.
The biggest early benefits are often found in sales and marketing. There’s lots of low-hanging fruit around customer experience, and in supply chain operations. And then of course there are numerous point solutions for specific companies, depending on their industry.
At Cognizant we have developed a variety of AI approaches for specific industries, to help them scale their efforts. For example, we can help insurance companies process claims with minimal employee involvement. In retail we developed an approach to dynamic pricing and demand forecasting. A similar set of algorithms and systems work for manufacturing and logistics companies. In healthcare we help companies offer patients an intelligent health experience. What we call “hyper-personalization” is an approach we use in several industries, including retail, media, and manufacturing. And we can help banks integrate environmental, social and governance (ESG) factors into decisions around lending and other operations. Many AI approaches can be applied across all industries.
Another part of banking where AI can help is preventing fraud. We’ve done work on what’s called “synthetic fraud.” That’s when someone takes a piece of another person’s identity, like their Social Security number, and puts it together with a different person’s real name and address or other info. Over time they slowly start applying for small loans and miscellaneous financial services to build up a history. Then they start using that “synthetic” identity for substantial-sized fraud. It’s difficult to detect before it gets bad because the transaction sizes are so small. We worked with a North American credit card issuer to apply graph technology and AI to look at many small pieces of information and connect the dots. So far we’ve helped the company reduce losses by about $25 million.
In life sciences, we helped a medical provider improve the accuracy of diagnostics for skin cancer. Using AI-based image analysis we can help predict whether a skin abnormality will be benign or malignant. It augments the doctor’s own opinion, and helped improve diagnostic accuracy by about 85%. In a completely different kind of application, we helped a mining company optimize its inventory management, so it was better able to predict when it would need new parts, like an expensive mining pot. That made their mines more efficient.
Another client is a logistics company in the Middle East. It was not satisfied with its ability to give customers accurate predictions on how long a package would take in transit and when it would arrive. If you tell someone it will come between 2pm and 3pm but it doesn’t get there until 6pm, that can be very frustrating. Using AI we helped the company not only do a better job predicting transit times, but improve how it loaded packages on their trucks and defined the best routes to drive. Just getting the loading order right can significantly improve the accuracy of delivery predictions. Even though this company was relatively early in their own AI journey, we were able to improve the accuracy of delivery times shown to customers in the company’s app by about 74%.
Often we need to spend quite a bit of time understanding the client’s business problem before we can translate it, in effect, into data. We need to study the decision drivers and parameters that inform a certain kind of company decision. Then we help figure out where the data comes from and where it’s stored. Then our data scientists can get to work designing, testing, and validating a software model for an AI solution. It’s been harder during the pandemic but it turns out much of that work can in fact be done remotely.
The pandemic drove many companies towards AI, because the future often felt so unknowable. We think 2020 will come to be known as the “tipping point” for artificial intelligence. That seems to be what companies think. Our survey found that annual spending increases on AI will double from 4.6% in 2020 to 8.3% in 2023.
Suddenly AI has become a very broad field. People throw around that phrase “artificial intelligence” fairly freely. The image of robots and automated beings that AI conjours up for many is simplistic and misleading. What’s really going on is that more of software is in effect become artificially intelligent, with algorithms helping people make better decisions, and predicting actions and behaviors, and learning as they go. The way we think of it is as applied intelligence. How do you apply intelligence to data and make better business decisions? Another thing we talk about is intelligent decisioning–taking data, applying algorithms that activate the intelligence that’s hidden in it, and making the result available to employees so they can make better decisions.
The good news is that companies using AI generally say they are achieving a variety of business goals, including better productivity, profitability, employee engagement, and customer satisfaction. The further along they are the more likely they are to say they are seeing increased revenue, making better decisions, increasing market share and innovating faster. As we see spending and commitment increase over the next three years or so, the very shape of business will look and feel very different, both to managers, and most importantly, to customers.
Here are the key things to remember:
- Data quality is critical to AI success.
- Good areas to start are sales and marketing, customer experience, and supply chain.
- Many different types of data will make AI algorithms more effective.
- The data you’ll need will not always come from inside your company.
- Data needs to be updated all the time.
- As technologies commoditize, the companies with the best engineering and data science workforces will win.
Sanjiv Gossain is the Global Head of AI, and Google Cloud Business Group at Cognizant.