Clutch.AI was born in late 2016, as an idea at Khosla Labs. Novopay, a fintech that grew out of Vinod Khosla’s startup incubator, is a digital wallet and Core Banking Solutions provider, with 50K+ authorized outlets, $1B+ in GTV, and 20M+ transactions per year. When Srikanth Nadhamuni, CEO of Novopay, realised that lending to their network of retailers could be mutually beneficial, he knew just the man to take up the challenge.
By this point, Subir Mansukhani had left the Mu Sigma Innovations & Development Lab and got started building an AI focused team at Khosla Labs. Before too long, the team started building an AI-enabled Credit Rating Engine for the retailers who worked with Novopay. Many kirana store owners were not graduates; most didn’t have formal (CIBIL) credit scores. Subir’s team leveraged hundreds of data sources, including online indicators, store transaction history, public records, cell phone meta-data, etc. to build this machine learning credit scoring engine.
Quite simply, machine learning refers to a family of algorithms that are able to automatically detect trends in past data and use these trends to make predictions on future, unseen data. This technology, coupled with the recent explosion in volume of data (over the last decade) has given rise to a variety of powerful new applications; from advanced virtual assistants to driverless cars. But there’s a catch — many of these cutting-edge algorithms (most notably neural networks) are so mathematically intricate that their decision making processes are incomprehensible to human beings; they are like black-boxes.
Regulatory bodies, such as the Reserve Bank of India, require loan rejections to be predicated upon a reason. You can’t just say that a model indicated applicants would default. Lenders have usually adhered to these regulations by following a rule-based model, where anyone who doesn’t meet a list of criteria (eg. Credit Score> 600, Age>23, etc.) would be rejected a loan. This approach makes it easier for banks to explain each loan decision, but is restrictive and prevents too many creditworthy individuals from having access to capital, and inhibits optimal growth of assets under management in the long term.
Clutch.AI fosters a culture of citizen data scientists, who can interpret the machine learning models they build without writing a single line of code. The Clutch.AI Zero Code machine learning workbench whiteboxes the black box, explaining how each prediction is derived, with weightage given to each attribute’s influence. Without writing a single line of code, the workbench quickly exposes relationships within data, reveals how easy segmentation would be, and generates a REST API that plugs into applications to predict future data, within minutes! With tools like these, Clutch.AI empowers bankers to provide the RBI as well as the loan applicant with specific answers as to their reasoning behind a loan decision.
The Clutch.AI workbench was used to create Novopay’s AI-enabled credit scoring engine, but it’s an industry-agnostic tool that interacts well with a variety of data from more than just fintechs. The Clutch.AI Zero Code machine learning workbench can be used to build machine learning algorithms to predict data in agriculture or healthcare, and sales or insurance. We’re even in talks with government agencies and cab aggregators to help them solve problems!
Artificial Intelligence used to be a black box; our tools democratize the power of machine learning. With Clutch.AI, everyone with data can instantly see patterns in data, build predictive algorithms, and leverage previously ignored data to grow their business without writing a single line of code.
