AI is a term that is now used often in the IT industry. However, there are several challenges involved when it comes to the usage and implementation of the technology. Sachin Vyas, vice president – Data Products, LTI (L&T Infotech) spoke to Dataquest about how pervasive AI is among enterprises, and what are some of the common challenges faced by companies while implementing the technology.
DQ: AI is a term that is now commonly used, but how pervasive is the technology among enterprises?
Sachin Vyas: AI technologies and products with embedded AI are already helping individuals to be more efficient at home, at work, while driving, and while doing many other activities. While the degree of impact or improvement varies depending on the activity, the average person probably uses AI technologies numerous times in a typical day.
A combination of high-speed connectivity, highly scalable cloud infrastructure, advanced computer architectures, and advancement of machine learning models is the driving force behind the proliferation of intelligent machines such as smart assistants, smartphones, smart devices, smart appliances, smart vehicles, smart factories, smart buildings, smart homes, etc.
From chatbots to virtual assistants to intelligent unstructured data processing, AI technologies are helping us more and more, both at home and in business. These technologies are often embedded in the functionality of other technologies we use—in our email platform, for example, where they might suggest sentence completions based on having learned from observing millions of emails being written.
Large scale consumer companies have been extremely successful in leveraging AI for their products. Many enterprises have also adopted AI for their businesses to reduce costs and optimize their operations. New AI algorithms are being continuously developed to improve functions and processes by observing, learning, and experimenting with improvements.
The major driver for enterprises to leverage AI is to automate time-consuming, labor-intensive workloads and free up the capacity of the enterprise’s greatest asset: its people. For example, an AI-powered financial advisor can analyze large volumes of transactions, flag risk alerts of fraudulent transactions in real-time, automate KYC processes, auto-capture details from images to auto-fill forms, automate customer service, and even automate bookkeeping.The true potential of AI is simply incredible, and enterprises have now started to “AI-ify” everything to augment human capabilities.
DQ: What are some of the challenges that enterprises face when it comes to AI?
Sachin Vyas: While there are many large-scale consumer-focused companies that have successfully leveraged AI in their products, there are also many enterprises who face challenges in scaling their AI proof-of-concepts to enterprise-wide production rollouts.
A few of these challenges are fundamental in nature, due to the shift in approach from traditional software-led business to AI-led business. Traditional software development and adoption processes are highly evolved, but AI product development best practices are yet to become mainstream. Few challenges are:
Probability of success: For interactive apps, we can generally predict the success of traditional software by using detailed wireframes and user-journey. But for an AI product, it’s hard to predict that without undertaking experiments. Most AI use cases originate from business teams/functions that reside in silos today. Business teams start with pilots and productionize few use cases. Enterprises lack an approach to setting up CoEs to democratize the technology and its functional manifestations for the organization-wide needs of today and of the future.
Definition of acceptability: It is possible to list out a complete set of acceptance criteria for traditional software including the boundary conditions, but for AI products it’s very hard to be complete with all possible practical conditions & acceptability under those conditions, not only regarding the accuracy of an AI system but also of other variable software performance aspects such as latency and usability.
Data availability: It is possible to create test-environments and associated data for traditional software but it is very difficult to make this approach work for AI products, especially if the use case itself is around conditions for which the available data is sparse. AI applications are data-hungry. To get correct outputs from your AI application, it needs high volumes of data and robust data management practices. Data preparation—identifying the right data sources, building data pipelines and data products, transforming the data to ensure its suitability, and identifying potential signals in the data, can be time-consuming and is one of the biggest hurdles to AI adoption.
Time and change management: We have seen two aspects of this challenge: 1) Users who resist accepting AI as their “co-worker” and can’t see AI as an assistant that can help make their job easier and elevate them in their roles; 2) Users and business owners who seek 100% automation from implementations of AI. The ideal approach is to achieve the maximum possible benefit from automation through reducing manual efforts by embedding AI within user business processes.
At a more tactical level, there’s a different set of challenges associated with the inefficiencies of scaling AI efforts at enterprises. These are:
High cost: The implementation cost for setting up AI, Machine Learning, and Deep Learning is substantially high—and also the value realization is not immediate. Enterprises struggle in the initial phase to justify the ROI, even though it does reap long-term rewards in terms of effort reduction and the time/cost optimization of functions and processes.
Putting in place the right ML team: The most critical task of any AI initiative is to attract and retain the right talent. Given the scarcity of AI-skilled talent, most enterprises prefer to collaborate with partners to set up AI labs centrally rather than opting to build out in-house expertise.
Deploying models is time-consuming: For most enterprises, it takes much longer to deploy ML models into production than to develop them.
Monitoring, governance, and management: Scaling security is often an after-thought. Scaling up a model that works with a smaller dataset to handle larger real-world scenarios is one of the biggest challenges.
DQ: Are there enough skilled individuals in AI in India?
Sachin Vyas: I believe finding right skilled talent for AI projects is still one of the major challenges. It is difficult to get the right combination of different skills: data and algorithms, statistics, IT, creative thinking, research, business impact articulation, storytelling, and understanding of cloud infrastructure and operations. There continues to be persistent gap in the market between AI talent and opportunities available to tap. Reskilling/Upskilling of resources to make them ready for AI projects is of core importance in India.
DQ: What are the top trends that you foresee for AI?
Sachin Vyas: In the last decade we saw how AI has been leveraged to process videos and images, understand speech, and react to conversation. I believe, in the next few years, Applied AI applications can add value in a variety of interesting fields. For example:
Healthcare: Healthcare has been a major beneficiary of AI during pandemic times and is rapidly evolving even further. COVID-chatbots enable individuals to have access to reliable medical advice round the clock. They have gone a step ahead in assisting people in self-diagnosis in preliminary stages. AI has undoubtedly played a significant role in this pandemic: predicting pandemic spread, speeding vaccine development by allowing faster trials, allowing recognition of masked faces, and analyzing test results. It enables more efficient & faster diagnosis to surface indicators of disease based on high quality large datasets related to patients’ biomarkers or medical imagery.In summary, AI in healthcare will continue to drive advancements that were nearly impossible for physical medical professionals to achieve.
Digital Content: Another major industry that boomed during the pandemic was Media & Entertainment (especially on-demand content viewed via Over-the-top (OTT) channels) and this trend is not going away soon. AI will continue to help in the international distribution of content, analysis of content by AI tagging engines to create metadata that can track various metrics like ad placements, building more personalized content, media scheduling, reducing customer churn, improving user experience with next show recommendations, etc. In fact, Netflix has claimed that their AI recommendation algorithms can save $1 Bn each year. Some of the other interesting areas such as digital content marketing are leveraging AI to draft marketing copy(e.g. Jarvis,Wordtune, Kafkai), and create digital ads and videos (e.g. Pattern89, Synthesia)
Customer Experience: AI is revolutionizing customer experience by completely redefining how companies interact with their customers. AI can hyper-personalize customer journey touchpoints from on-boarding to customer care, increasing engagement, customer satisfaction, and loyalty. For example, the Know Your Customer (KYC) process during customer on-boarding is complex when dealing with every document manually and now even digitally (mobile-scanned copies in emails, messengers, portals, etc.). An efficient AI solution can help in understanding KYC by automating document processing to expedite the customer on-boarding process digitally and give back business users their time to perform higher value-added tasks.
AIOps: Simply put this is AI for IT operations. The usage, complexity and bandwidth of IT systems have increased drastically, leading to higher volumes of newer issues discovered in operations. AI makes IT operations intelligent by leveraging advanced data processing & analytics to reliably uncover issues in the IT landscape and provide a just-in-time response to resolve them. There are many more such areas and it is a most exciting time in the field of artificial intelligence.
DQ: Tell us about your AI bot solution?
Our ‘AI bots’ solution, built on LTI’s Mosaic AI comprehensive platform, intelligently automate repetitive tasks for straight-through processing, boosting productivity and optimizing operational costs. The bots are built to simplify business functions/processes to deliver business outcomes that redefine our clients business models and customer engagements. The key focus areas are:
Experience Transformation: “Uberize” content delivery through an immersive & intuitive user experience for an extremely personalized and contextualized conversational experience focused on delivering outcomes.
Intelligent automation: Embed intelligence into automation, thereby automating workflows that require human-like intelligence for decision making.
Empower & Enable Legacy: The value of AI to enterprises is significantly amplified when the knowledge that resides in the traditional legacy system is digitized and contextualized.
DQ: What is your target market?
Sachin Vyas: While ‘AI-bot’ solutions are specific to business functions (employee engagement, customer service, etc.) and limited to a few industries; our target for Mosaic AI is more holistic across use cases and industries. Our advanced Mosaic AI platform is for enterprises that have a fair understanding of AI concepts and model lifecycles and have a team of data scientists experimenting/working on AI models. We target practitioners across regions/geographies and across industry domains who want to work in one collaborative environment and plan to move towards more open-source ML Development for cost savings.