Artificial Intelligence is helping upGrad enhance productivity across coding, operations and content creation, but its biggest impact will be in achieving the 'holy grail of education' — making candidates job-ready, said Mayank Kumar, Co-founder, upGrad and Borderplus, and Governing Council Member, Tech Entrepreneurs Association of Mumbai (TEAM).
The company is also focusing on making education more personalised, moving away from what Kumar described as the traditional “factory model of education” -_- upGrad began its AI journey several years ago and has since implemented the technology across its value chain to improve user experience, operational efficiency and learning outcomes.
Mayank Kumar and Deepesh Dhakad, Chief Technology Officer (CTO), upGrad, speak with FE FUTECH.
What is your broader view on AI’s potential in the education sector, and what kind of AI strategy are you building for upGrad over the short, medium and long term?
Mayank Kumar:
I believe AI will impact education in three major ways -_- First, it will significantly improve efficiency and productivity. Companies like upGrad will increasingly use AI for automated grading, smarter operations, and faster content creation.
Second, AI will drive personalised learning. Traditionally, education followed a factory-style model where every learner consumed the same content in the same sequence -_- Going forward, AI will enable highly personalised learning journeys.
For example, after attending the same class, different learners could receive different homework assignments depending on their individual comprehension levels and learning gaps.
The third and perhaps the most important area is outcomes — the holy grail of education. AI will eventually help learners prepare better for jobs, improve interview readiness, identify suitable companies to apply to, and guide career decisions more intelligently.
At upGrad, while we have already started working on efficiency and personalisation, our long-term objective is to move towards measurable learner outcomes and take ownership of those outcomes.
How is AI helping upGrad improve efficiency across functions — whether in helping students learn better, improving internal operations, or reducing costs?
Deepesh Dhakad:
Broadly, AI helps improve user experience, operations, and revenue. At upGrad, we look at AI in three buckets.
The first bucket is improving user experience using AI. For example, in our courses, users can now summarise long-form video content. Instead of watching every minute of a video, learners can search through the content directly.
We also help users transcribe and translate content into their native languages. While we deliver most of our content in English, people often understand concepts better in the language they are most comfortable with. For instance, many learners prefer Hindi, while our Southeast Asian customers prefer their own regional languages -_- These are some of the areas where we are seeing strong AI-led improvements in user experience.
The second bucket is operations. We are using AI to reduce complexity in processes, improve delivery efficiency, and optimise costs. A strong example is assignment evaluation.
Many of our courses involve subjective assignments where learners write paragraphs, develop use cases, or create business theses. Evaluating such assignments traditionally required subject matter experts, and it often took two to three days for a batch of assignments to be reviewed.
There was also dissatisfaction among learners regarding grading quality, which led to around 10–11% of students requesting re-evaluation of assignments. We realised AI could play a significant role here -_- We built what we call an ‘AI grader’. It takes the subject and the evaluation rubric and assesses assignments objectively against those parameters. As a result, a process that earlier took two to three days is now completed within hours. In many cases, students receive feedback almost instantly, within 30 minutes.
More importantly, re-evaluation requests, which earlier stood at around 10–12%, have now dropped to below 1.5% . Students are significantly happier with the grading process. The system has also found acceptance across the universities we partner with.
The third bucket is revenue enhancement. Our business model involves generating leads from prospective learners and ranking those leads based on probability of conversion. Naturally, our counsellors prioritise higher-rated leads.
However, we discovered that there was a large pool of lower-rated leads that we could not effectively engage with due to manpower limitations and operational economics. We then began using AI-driven bots to interact with these leads. We found that a reasonable percentage of those users were still interested in speaking with us and eventually purchasing courses.
How much additional business have you generated from these lower-rated leads after deploying AI?
Deepesh Dhakad:
Broadly, we are seeing around 5–8% incremental users coming from that segment, users whom we earlier would probably not even have pursued.
Depending on the average order value and the type of course, that translates meaningfully across business metrics. But in terms of pure incrementality in users, we are seeing a 5–8% increase -
Mayank Kumar:
AI at upGrad is fundamentally about delivering more personalised learning, improving process efficiency, and enabling smarter counselling, thus expanding market access. Earlier, the cost of engaging with lower-quality leads was too high, which meant we could not efficiently serve many learners who were interested in our offerings.
AI helps us address that gap. It not only improves user experience and reduces costs, but also expands access to education for segments that we earlier could not economically serve.
Can you briefly describe upGrad’s AI journey — from early experimentation and proofs of concept to production deployment?
Deepesh Dhakad:
Over the years, upGrad has experimented with many AI initiatives, some successfully and some unsuccessfully. For instance, the auto-grading initiative started around mid-2024. At that point, the technology itself was still relatively nascent.
We spent considerable time improving accuracy and testing the system rigorously. But interestingly, the larger challenge was not technology — it was human acceptance. Since we work closely with universities, it was important for institutions to feel comfortable with AI-based grading systems. Older institutions naturally take longer to adopt newer technologies.
Today, adoption is easier because AI has become mainstream. But two or three years ago, the ecosystem was much more hesitant. Over the last year, we have also invested heavily in AI-enabled engineering systems.
Today, nearly 60–70% of the production code we write is AI-generated — possibly even more. We also built a ‘bug board’ system to address long-standing production defects. While critical defects are always resolved quickly, lower-priority defects tend to accumulate over time. Over the last three months, we aggressively targeted that problem using AI, and reduced the backlog by nearly 65%.
When I joined, our engineering organisation was almost twice the current size. Today, we operate at nearly 50% lower engineering strength while delivering 1.5x to 1.6x higher output. In effect, productivity has increased nearly threefold.
What are some of the AI initiatives currently underway at upGrad?
Deepesh Dhakad:
One initiative I am particularly passionate about involves improving counselling intelligence.
Today, our academic counsellors spend significant time engaging with learners. These interactions often involve five to eight calls spread across several weeks before a learner either purchases a course or decides not to proceed.
A large part of that process is subjective and depends heavily on the counsellor’s experience, capability, and judgment. What we are trying to build is an AI-assisted system that creates a three-dimensional profile of every learner.
The first dimension is identity — understanding who the learner is, including academic background and professional context.
The second is financial identity. Since our courses range from ₹10,000 to ₹4–5 lakh, understanding financial suitability is important for personalised recommendations.
The third is aspiration identity — understanding what the learner wants to achieve through upskilling.
Today, much of this intelligence exists informally within counsellors’ experience. We want to convert that into a structured AI-driven intelligence layer that stays with the organisation throughout the learner’s journey with upGrad.
That can enable much deeper personalisation and smarter course recommendations, benefiting both learners and the business.
How do events like Mumbai Tech Week help organisations solve such complex AI problems?
Mayank Kumar:
It is difficult to learn everything independently. You can either make mistakes yourself and learn, or you can learn from others. Events like Mumbai Tech Week (MTW) create opportunities to understand what others in the ecosystem are building, exchange ideas, and improve our own thinking.
Of course, such events also help engage with policymakers, contribute to broader ecosystem development, and connect with talent. But personally, I see them primarily as powerful learning opportunities.
Deepesh Dhakad:
Many companies spend heavily travelling globally to learn from technology ecosystems -_- Events like Mumbai Tech Week bring that ecosystem together in one place.
You get exposure to diverse AI use cases across companies — whether it is what upGrad is doing, or what companies like Razorpay or Urban Company are building.
It is also a great platform to interact with young talent bringing fresh ideas and perspectives. Under one roof, you learn best practices, discover new approaches, and connect with exciting people and companies.
Prime Minister Narendra Modi has appealed to the country to save energy using various austerity measures. What impact do initiatives such as ‘learn from home’ or work-from-home models have on the online education ecosystem?
Mayank Kumar:
One thing we observed during the COVID period was that when commuting time was reduced because people were working from home, they spent more time on upskilling and education.
Education becomes what I would call the ‘least guilty spend’. People feel comfortable investing in themselves. So whenever work-from-home flexibility increases, we do see a positive impact on learners investing time and money into self-development and becoming more future-ready, especially in the AI era.
