Better known as the startup nation, India is in the middle of a massive digital transformation, fuelled by the phenomenal growth of tech startups. India’s startup story has seen an Up & Up phase since the last six years. According to our latest study, there are more than 5,000 companies in India today that claim to provide analytics as an offering to customers. This includes a small number of companies into products and a larger chunk offering either offshore, recruitment and training services. Our research indicates that India accounts for 7% of global analytics companies and an average analytics company staffs around 179 employees on their payroll.
While on one hand, it is true that India is seeing a great success in big data and analytics with a number of startups in this space, there is another facet of entrepreneurial India that we would like to discuss in this article. What are the hurdles in starting a data science startup and the pitfall future startup enthusiasts should avoid? Broadly speaking, the hype around artificial intelligence and machine learning has reached fever pitch, but there are limited advancements in this field. Startups have to navigate the key barriers such as lack of training data, accuracy of models and qualified talent to build and thrive.
Analytics India Magazine spoke to startup heads to find out the key challenges in building a data science start-up in India and gives a primer for budding startup enthusiasts. Divyesh Patel, Co-founder of Turing Analytics and Aniruddha Yadav, Founder and CEO Gauge Data Solutions weigh in to separate the hype from reality and share their inputs on how to avoid the technology trap, build strong product ideas, bench strength and whether one should you consider VC buy-in.
Strong Product Idea: A strong product idea would be the one, that can leverage available tools and data to deliver an industry solution with high accuracy. Patel believes there is a lot of scope for innovation in this process and the real advantage would be in focusing on a niche where one has a strong domain expertise. Yadav, seconds the same view. With data science startups and technical founders, one must avoid the “technology trap”. “They must find a problem where the market demands a solution and then apply data science to try solve it,” he said. Here’s what we gleaned – your product should be scalable and the market should have a demand for that solution.
Tech Stack: Let’s face it, running a tech startup requires a lot of effort and financial planning. A lot of money is plowed in steep licensing of software. Open source technologies and tools plays a huge role in helping startups overcome the software costs that may or may not be aligned with your business’s requirements. Patel recommends sticking to popular libraries like Tensorflow or Pytorch. “It helps you to focus on solving the business problem rather than spend a ton of time in doing engineering,” he emphasises. Yadav echoes a similar view and has a word of advice — open source is good, but check the various licenses that come with it and their implications.
Building Bench Strength: Despite all the overreaching hype and buzz surrounding data science, it is near impossible to find someone with experience in data science, Patel reveals. Another harsh reality start-ups are grappling with is — how would one convince the person to join your new start-up v/s working for Google? Having someone in the founding team with deep data science knowledge is a huge plus and would help select the right candidate for building the engineering team, shared Patel. Another oft-treaded route most startups are following is — hiring good engineers and training them in data science. Startup environment can be exciting and exacting at the same time. So, how does one keep the momentum going? “Communicate your vision to potential hires and make sure that this communication is always ongoing. Your team should feel that they have a stake in the product and a culture of excellence should always be fostered,” Yadav added.
VC Buy-in: Money does make everything go round, especially for startups that are seeded with VC money. However, most entrepreneurs don’t like losing control/autonomy over core business decisions.The general view is that investment decisions should be taken strictly on basis of the business requirement. On one hand it may be relatively easy to attract investors in a particular area, but it could be harmful to the startup in long run.
When and How to Scale Effectively: Here’s an honest view: scaling is a trade-off between exploration and exploitation. One can be easily be lulled into believing that the product fits the market and can be scaled, Yadav lets in. Check out these points from the entrepreneurs who have already dipped their toes in the market.
- Patel believes that the decision to scale should be taken based on business momentum and financial situations, however to comment on specific aspect related to data science
- If good accuracy has been achieved in solving at least one client’s problem and the solution can be customized quickly for others, one should consider scaling up as quickly as possible.
- One must always explore to find even better fit after a few iterations and this can continue in parallel, Yadav emphasizes.
However, before you dive in, here are a few things to remember. There are plenty of barriers in setting up data science startup and foremost is the task of keeping up with thousands of new research papers/ blogs to filter out real advancements in the fields from other noise. It is a daunting task for every marketing research executive to keep up with new patents being published and startups making a headway in the same area. Another huge barrier is the cost of generating training data that can be very high in field like medicine, reveals Patel. Your product would only be as good as the models. “Quite often, publicly available solutions have around 70-80% accuracy on benchmark. Hence, it’s tempting to believe that one could improve upon it a little and start shipping an industry solution. But, in reality it is an uphill task. And anything below 90% accuracy would be unacceptable in the market,” said Patel. Finally, the lack of quality talent can make it hard to attract and retain employees and eventually scale the startup.
Try deep learning using MATLAB