Unbox the Matrix: India’s AI Potential
Generative AI. Large Language Models. Foundational Models.
Variational AutoEncoders (VAEs). Generative Adversarial Networks (GANs).
Auto-regressive Models.
Artificial Intelligence (AI) has come a long way since the early days of big data, where statistical inference was the primary driver of outcomes. Today, AI models are being trained on vast amounts of data at an unprecedented scale. For e.g GPT-3 was trained on 570GB of text data, the equivalent of ~45 million books. An incredible step-up in scale. And the amount of data being generated as a byproduct is also growing exponentially — estimated to be 180 zettabytes of data by 2025, up from 33 zettabytes in 2018! (FYI: 1 zettabyte = 1 trillion GB)
All this data and large scale training of AI algorithms has supercharged the power and scope of AI apps, with incredible breakthroughs in speech recognition, image and video processing, and recommendation systems seemingly week on week. This revolution, coupled with the availability of pre-trained models and affordable cloud-based AI services has lowered the barrier to entry and democratized access to AI for both individuals and organizations.
From where we are today — it’s not a big leap to imagine a world where machines create art, music, and spin stories to rival the work of humans. Or a world where language models can write entire code-bases or generate custom responses to customer inquiries, without any human intervention.
Incredible / Scary / Existential — based on your point of view!
Pardon my French
Before we go deeper — there’s so much going on in the world of AI that you’re just one digital detox away from being completely clueless on what’s going on. Here’s a quick primer on the 1st line of the article- which makes up a lot of the base of what I’m going to talk about. You can ChatGPT the rest 😅
- Generative AI: Refers to a set of AI techniques that enable machines to create new data/content from an existing dataset. The core idea here is that the AI output is ‘novel’ and not just a rehash of the inputs which we tend to see with traditional machine learning models. Generative AI can create output across formats — text, images, videos (those weird Balenciaga videos) without human intervention
- Large Language Models (LLMs): A type of generative AI built using deep neural networks and trained on massive amounts of data to learn patterns and relationships. Models are ‘trained’ via a process called unsupervised learning, in which the model is given a large amount of data and learns to predict outputs. The most popular examples of LLMs are OpenAI’s GPT and Google’s BERT
- Foundational models: Traditional ML algorithms which are rule-based systems statistical models that rely on explicit programming and hand-crafted rules to generate output. Examples of Foundational models include Markov Models, NMT models and decision tree based systems
While we tend to rave about every new technology — AI truly has all the ingredients to revolutionize the world in many ways (more than the obvious that we see today) and bring benefits to both individuals and organizations alike.
In my view, its likely that a larger chunk of value will initially be captured in Enterprise grade use of AI — primarily due to access to large amounts of data, and the ability to invest in and scale up AI models and their training modules.
What does this mean for Enterprise companies?
Taking a step back, enterprise companies have a simple goal — find a piece of work at a company and take care of it cheaper/better/faster than the target company can themselves.
“The how” is key and so far there have been two main approaches:
- IT Services Approach: Looks to staff actual humans (extended workforce) on an enterprise’s behalf to complete and execute on a piece of work. These are typically built on a geographic cost arbitrage and engagements tend to be bespoke. Yet the volume of work available has led to the creation of behemoths like Infosys, TCS etc. Key to success here is hiring, training and execution since costs scale along with revenues for IT Services organizations
- SaaS Product Approach: Offer product solutions that are ready to hit the ground running right out the box. The holy grail for these companies is to find ways to ‘productize’ the work. ‘Productization’ effectively means turning a piece of work into a standardized, scalable, and repeatable product. The idea here is that once the product is built — it can be replicated and sold at nearly no marginal cost; which makes margins/scaling look very attractive. The challenge here is finding those pieces of work that standard and repeatable across verticals and use cases!
For example, SalesForce ‘productized’ Sales since sales motions always involve prospecting, qualification, closing — regardless of industry. So you can build a product that can provide this solution at scale at a price that beats building it bespoke. But think about insurance underwriting or tax accounting — its hard to find repeatable motions here so IT Services dominates here
In my view, there is a huge opportunity for AI to blur the lines between these two approaches and build a hybrid organization with AI at the core!
Productizing Services
The TL;DR from all the AI hype is that algorithms are now advanced enough to replace redundant human effort on bespoke projects. This means that, with the right inputs, AI can ‘productize people’ and, as a result, expand the universe of use cases that can be ‘productized’ with code — all in a way that is incredibly economical to execute. Imagine a bespoke service for every business need!
In other words, AI can help product-led approaches take a sizable chunk of work away from services-style engagements that leverage lower-cost human talent to build software solutions. This can be a game-changer for companies looking to streamline processes and reduce costs.
However, a lot of the proprietary business data and inputs that these LLM models need are currently in the hands of large IT services organizations. Just imagine an AI copilot learning alongside services organizations as pieces of bespoke work get completed. This process would completely absorb the upside that IT Services has pioneered so far — hire well, train well and manage new hires well.
Yet all these large companies are currently disincentivized (think rate cards and project billing metrics) on multiple levels to give up this data and move away from their current models — just look at how Infosys’ stock price changes with quarterly performance and you’ll get an idea.
Infosys Stock Performance: Not a forgiving environment for experimentation
So the real question is what changes first?
Will product-first LLM models grow fast enough to overcome the current barriers to entry and learn business operations faster, or will one of the IT services organizations play the long game and take a massive short-term hit?
Even an LLM model can’t answer that question with confidence but my hunch is that the product first approach can win if it can scale quickly
What we can say with confidence is that India has a massive advantage on this front. India is the only country where both IT Services and Product SaaS has matured to a significant scale. I’m convinced there is an incredible opportunity for Indian AI-powered enterprise companies to take an unassailable lead in this massive, emerging market and the early bird gets all the worms in this case!
How are companies using AI today?
Through my interactions with startups at WaterBridge Ventures, AI is the hottest buzzword in board meetings across the world and Founders are looking to cash in on this new tech. Understandably a lot of the integrations so far have been gimmicky and superficial given progress and change in AI can be measured in days (hours?)
Regardless, a few recurring themes that we are already seeing playing out -
- Content Generation
- Create content for smaller use cases at little/no cost
- E.g. Take a photo of a product and make eComm ready marketing materials
- E.g. Write my emails for me so I can improve my outbound sales
2. Hyper personalization
- Give the illusion of bespoke by using code to hyper personalize efforts at little/no cost
- E.g. Mumbai Indians sending personalized shoutouts to their fans
- E.g. Build a personalized learning journey tailor made to each student
3. Compiling Content
- Scour the internet/data sources for insights and action items. Pick the needle from the haystack every time
- E.g. Trawl Sales convos to identify user stories should paid attention to first
- E.g. Understand employee sentiment by looking at slack interactivity
4. Manpower scaling
- Expand human like work without any additional human overheads
- E.g. Answer specific questions about particular health/insurance claim
- E.g. Auto add summaries/content labels to metadata based on utility
However, to truly deliver value — AI must be integrated into the core of a product or service. This will enable it to drive tangible improvements in functionality, efficiency, and user experience.
Conclusions
The landscape is moving so quick that it would take a brave person to identify where this space will pivot to next. In my view, productization of IT Services will be a key trend and hyperscaling on proprietary data will be a vital lever. Startups that can can crack trends early will have a massive early mover advantage in the market (time is money when it comes to AI)
Fortunately, India as an ecosystem is perfectly poised to make the most of this wave. With the highest density of IT services in the world and an enviable pool of strong product management talent, the Indian B2B ecosystem could find itself best equipped globally to crack this massive opportunity
We’re looking for entrepreneurs who can think creatively about this conundrum. If you are building in the AI space and are looking to brainstorm ideas — would love to chat at nilesh@waterbridge.vc
#AI #India #B2B #SaaS #Disruption
