AI: Opportunities in plenty, but challenges remain

Indian companies are fast adopting Artificial Intelligence (AI) to achieve a range of business objectives incorporating AI to make core business decisions as well as discrete business functions such as facial recognition and customer service. There is considerable maturity in thinking and deploying AI projects with early adopters showing the way with best practices and vendors offering robust solutions to address customer needs.

The State of AI in 2021, a global survey across industries and company sizes, conducted by McKinsey Global Institute finds that Indian companies demonstrated the highest adoption of AI among all regions in the world. Broadly, emerging economies, such as China, the Middle East, and North Africa have shown a steady increase in AI adoption over the past year.

IDC pegs India’s AI market to reach USD 7.8 billion by 2025 growing at a CAGR of 20.2%. Currently, AI deployment is the highest amongst IT and technology companies followed by financial services companies. The COVID-19 pandemic accelerated the adoption of intelligent technologies across organizations to develop resiliency that can help them prepare better for future uncertainties. While the need for contactless services is driving the uptake of AI services in e-commerce and healthcare services, AI deployment has suffered a setback due to a devastating second wave of the COVID pandemic.

The pervasive deployment of AI has started touching the lives of consumers in India with 61% of respondents in a Deloitte survey saying that they use digital assistants in different scenarios such as at home and work and even while commuting to help them save time, receive reminders and get things done. This has led 71% of Indians into believing that AI will help solve complex problems and enable humans to live richer lives.

Business use cases

AI implementations are being seen in a wide variety of use cases—from eliminating process efficiencies in financial companies, predictive maintenance of machinery, customized manufacturing, and personalization in retail and recommendation engines to using data in real-time business decision-making, AI is getting deeply embedded at the core and in the day-to-day play of business operations.

Organizations are taking either of the two approaches in AI adoption with some focusing on achieving specific business outcomes with discrete AI implementations while others are seeking to undertake a complete overhaul of the business with a digital data-driven core and leveraging AI to maximize benefits. AI adoption is being driven by companies seeking to achieve a competitive edge by integrating it as part of the larger organizational transformation.

For instance, manufacturing companies are upgrading machinery and equipment to adopt IT-OT integration and leverage the benefits of AI, such as predictive maintenance, enhance workplace security, etc. One of the most widely employed use cases is AIaided customer care via chatbots and AI-enabled communication which is making round-the-clock interactions and customer service a reality.

AI is automating many processes, reducing the workloads on employees and doing certain repetitive tasks better and faster than humans. For instance, when a customer touches a contact center, AI can quickly process the existing information to learn customer patterns and forecast behavior; route calls to the best representative; provide the necessary background information to help resolve the customer query quickly and easily; determine the right message to send to the customer at appropriate times, etc.

From capital markets to consumer banks and fintech, financial services are using AI-powered solutions to identify new opportunities and boost revenue, reduce expenditure, and automate manually-intensive processes. Fintechs and investment firms are mostly using AI applications for algorithmic trading, fraud detection, and portfolio optimization, reflecting a primary focus on protecting and maximizing client returns while banks and other financial institutions are using AI to detect and prevent fraud, and sales and marketing optimization.

Banks are also leveraging AI-enabled applications for customer acquisition and retention, and for cross-selling and up-selling personalized products and services. Insurance companies are using AI for repetitive and tedious tasks, such as claims processing, customer identity verification, initial claims routing claims triage, fraudulent claims detection, and claims management audit. Mumbai-based Edelweiss General Insurance is implementing an AI solution in claims adjudication and looking at using AI for intelligent pricing and product innovation. Increasingly, the need for businesses to collect, process, and leverage data for real-time decisions are pushing the adoption of AI. For instance, Vistara Airlines is using analytics to take decisions on route profitability, understand the travel requirements of passengers, and aircraft utilization, and reduce passenger transit time to achieve operational efficiency and higher profitability.

Using data from around the locality and AI-based insights, 7-Eleven takes fundamental business decisions, with insights help to understand the buying behavior and preferences of the people in a locality, identify the needs of a population in a given area to decide the ideal location of a store, what format of the store to set up and what kind of food and grocery to sell in that area.

Implementation challenges for AI

No doubt the benefits of AI are many enabling businesses to disrupt the market with differentiated services, new business models, frequent innovation, high degree of personalization while ushering operational efficiency with increased automation, predictive maintenance, higher utilization, and collaborative approaches. Yet AI is not a silver bullet and there are many road bumps before reaping the harvest.

Below, we enumerate common challenges that can derail AI implementations:

Data availability: Data is the lifeblood of AI models and without data, designing and publishing smart AI models is becoming a challenge. Specifically, with stringent regulations that restrict the flow of data, Indian companies are faced with a situation to train data models based on local data even though the target is to be applied in global scenarios. To overcome the lack of data availability, many companies are resorting to using synthetic data. A synthetic dataset resembles the real dataset, which is made possible by learning the statistical properties of the real dataset.

Bias in the data model: AI is essentially software-based logic embedded in a data model. These models designed by humans have an inherent possibility of having a bias towards specific outcomes such as gender, race, or having a prejudiced view about certain issues. If an employer uses an AI-based recruiting tool trained on historical employee data in a predominantly male industry, chances are AI would replicate the gender bias. When a recommendation engine tends to pick songs or products based on your historical preference, it is demonstrating an algorithmic bias in its design principle.

Lack of quality data: When the AI model is designed to be learning and improving with a continuous data feed, access to a large amount of data from a variety of sources is important. Otherwise, when fed on limited data, deep learning algorithms tend to throw up inaccurate results. For instance, natural language processing algorithms such as Google Translate, which is learning from real-world data, tend to pick up existing gender prejudices and provide output such as “he invests” and “she takes care of the children.” Much of the biases of deep learning engines creep in due to fallacies in the datasets. Undersampling skews in class distribution makes the AI model completely ignore minority class, while oversampling leads to the over-representation of a certain group or factors in the training datasets.

Privacy and compliance requirements: Given that vast amounts of customer data are at play, there are regulatory and privacy issues in AI projects. Security and access to data must be meticulously integrated at every stage of implementation, otherwise, data leakage will not only derail the project but also will get the organization mired in litigation and government sanctions. For instance, personal information data such as data related to customer health and finances are very sensitive and can cause serious damage to customers in case it falls in the wrong hands. From capital markets to consumer banks and fintech, financial services are using AIpowered solutions to identify new opportunities and boost revenue, reduce expenditure, and automate manually intensive processes.

Lack of skilled workforce: With AI implementations in full swing in most organizations, there is a severe shortage of people who understand and harvest the benefits of AI. Shortage includes developers and engineers, AI researchers, and data scientists. Besides, technical skills people to lead AI projects such as business leaders, domain experts, and project managers are also in shortage. These people are critical to actualizing projects, people who understand the value of data and conceive projects and push the adoption of data within the organization. Most organizations are seeking to fill the gap by hiring people with AI skills from outside the organization and this is creating a wider gap.

Instead of training their employees, organizations are seeking to reap AI benefits in a hurry to beat the pandemic-induced disruptions and the need to reduce human interface.

Executive support: AI projects must be a collaborative effort across the organization because data sharing is inherent to its success. Executive support not just be at the leadership level but must thrive at all levels of the organization as data sharing must be continuous to close the loop to improve the algorithm and business leaders at every stage must understand this need. For example, if a business development leader funds an AI project while marketing managers are lax about sharing the data, AI’s ability to predict the accuracy of the marketing campaigns will be affected and thereby undermine the impact of the campaign. It is important to appreciate that data sharing will help people at all levels of the business development functions and that it will be able to do the administrative tasks that consume much of managers’ time faster, better, and at a lower cost.

How to get AI projects right

Most organizations understand the need for AI projects to aid business efforts, though there is a lack of clarity in extracting meaningful value from AI. Not surprising, given that AI is still in its infancy and will take time to consistently deliver benefits.

Typically, organizations are applying AI in discrete uses, an approach that doesn’t produce consequential change. But the other extreme trying to overhaul the whole organization with AI is simply too complicated to be practical. The answer lies somewhere in between wherein there is a cultural change in terms of design thinking, and an understanding amongst executives about the use of data in decision-making, promoting its usage and distilling processes to finetune data and identifying use cases on an ongoing basis.

Below, we enumerate recommendations based on the insights and understanding of our esteemed panelists who have been early adopters and AI practitioners:

Identify the right use cases: Initial use cases are those that are broad enough and where new ways of working is going to create impactful changes. Identify projects with a series of inter-related business activities and that which are likely to have wider impact such as chronic process inefficiencies, rapidly fluctuating customer demand or difficulties in getting products to customers. Design metrics makes changes visible and measurable in terms of improved customer experience or financial gains.

Design re-usable assets: This is a critical component in design thinking and will have a cascading impact of AI initiative. After the initial returns, it is important to build on the benefits on a continuous basis to expand the project in an incremental fashion which will sustain the momentum and demonstrate the impact. Building reusable assets helps promote a collaborative culture which is critical in the success as data sharing and feedback loop is crucial to train and enhance the accuracy of the data model.

Executive sponsorship and team empowerment: Needless to say, success of AI implementations must be backed by leadership sponsorship. It cannot be led by data scientists alone without having a pool of people who share the vision of AI and business goals. A fully empowered team led by business leader and comprising data scientists, change management experts and a group of frontline users are necessary to execute AI projects. To facilitate the momentum and excitement, team members must be assembled and empowered as a single unit, preferably as CoE. Or the team must work together as a coherent group reporting to a single business leader, otherwise there will be islands of efforts and delays as teams wait for approvals and clearances.

Incorporate design thinking: It is important to reimagine systems and processes to create impact. For this, team members must be prepared to spend time with end users such as internal teams and customers to identify inefficiencies, understand the gaps and inject a new vision with the help of AI-enablement. Teams must identify the end goal and work backwards incorporating design thinking to re-imagine the existing set up with clear goals such as delivering a wow customer experience or rehauling the vendor onboard system for faster deliveries.

Invest in capability building: AI implementation must be accompanied by capability building on different fronts including organizational culture and new technology. Teams must learn to work in a truly collaborative environment underlined by a shared belief in the larger good of the organization. This means teams must share data and work in an agile manner to continuously share and improve experiences and thereby the data model. At the same time, it is an opportunity for the organization to embrace new and advanced technologies conducive for AI implementation, such as Cloud infrastructure, DevOps enablement, serverless and micro services architectures. It does not mean that organizations need to rehaul IT systems, but it is a good idea to embrace Cloud platforms that can accelerate project deployments to deliver initial benefits quickly and easily. A McKinsey survey finds that 64% of AI high performers run the workloads in the Cloud.

Embrace risk mitigation: Risk management is a shortcoming for most organizations adopting AI and this spans risks emanating from cyber security, meeting privacy norms and regulatory requirements. Embracing a set of best practices such as embedded security and privacy in data access via multiple levels of authorization, using encryption in data storage and data transit help mitigate risks to a large extent. Also employing innovative techniques such as storing and processing data on customer devices rather than at edge can help risk mitigation in specific use cases. According to McKinsey survey, The State of AI in 2021, high performers are more likely to scan the sampling data for under representation to eliminate bias in the model, have legal professionals in the team of AI committee and a stringent governance system in place to monitor adherence to best practices.

Get set for the future with AI

AI-enabled businesses are becoming a reality as organizations employ the technology to achieve meaningful outcomes. Early adopters are breaking boundaries by eliminating systemic inefficiencies and reducing human workloads while ushering in higher accuracies and profitability with machine learning technologies. This is having a dominos effect within organizations as it equips adopters to envision more benefits, leverage learnings and new assets to quicken the pace of AI adoption and gain further advantage. Businesses are on the threshold of a new era as technologies mature and converge to deliver new benefits facilitating the rise of AI to emerge as the foundation of new business models.

- The article was first appeared in FUTURESCAPE 2.0, a book jointly produced by CIO&Leader and Hewlett Packard Enterprise India.

Image attribution: Freepik