There’s little doubt that artificial intelligence (AI) will be core to the business model of every leading enterprise of the future. But right now, stories abound about how difficult it can be to unlock business value from big data and AI. Take the example of a financial institution that spent R1.5 million and six months to develop a customer attrition model, only for it to deteriorate in less than a year.
This company is by no means unusual. As we speak to financial institutions and other large businesses, they share stories about spending months and millions of rands developing AI models that fail to meet expectations or even to deliver any value. This is a global challenge, with one study finding that 87% of data science and machine learning (DSML) projects do not progress beyond prototype and R&D stage.
Suman Singh, founder and CEO of CyborgIntell, often speaks of his time in the data science trenches, where he learned first-hand why so many data science and AI projects fail. He led a team of 80 high-calibre data science experts – many of them educated at the most prestigious American universities – that struggled to gain the expected ROI from machine learning and data science projects.
Data scientists – overworked and under pressure
Even though they were working excessively long hours, these data scientists were constantly under pressure and falling behind schedule. More than 90% of projects were not delivering the expected benefits. Singh realised these extremely clever people were spending an undue amount of their time doing slow, repetitive, manual work.
That was the situation of a company that could afford to attract and retain a team made up of the best. It’s even more difficult for companies in South Africa competing for a small pool of talent. Building a team that spans the DSML lifecycle is costly and complicated. Outsourcing comes with its own challenges such as intellectual property ownership.
Manually carrying out tasks such as data selection and modelling or operationalising AI is not only mind-numbingly dull for the data science and AI team. It’s also slow, expensive and doesn’t scale. The result is it takes so long to develop, deploy and operationalise ML models that the data the system was trained on is often out of date before it is ready to be deployed.
The solution that Singh came up with was to elevate the level of automation in the DSML lifecycle. Companies today can benefit from a new approach to AI, based on a one-stop, zero-code method for rapidly developing, deploying and operationalising AI applications at scale. This approach slices the time to deploying AI projects from many months to as little as two to four weeks, while helping to reduce risks and enhance ROI.
A 250-fold productivity boost
Automation is up to 250 times faster than manual approaches and eliminates human error from the equation. Where a team of data scientists could build dozens of machine learning models, automation enables them to scale up to millions of models if they wish, improving accuracy. DSML models can be deployed in seconds, while auto retraining, validation and redeployment can be accomplished in hours rather than days or months.
Such a solution reduces the time required to develop accurate, production-ready models to a few hours without writing any code. A scalable AI platform can address a variety of use cases for every enterprise in various industries, from optimising pricing, driving cross-selling and upselling opportunities, and targeted marketing to predicting loan defaults, pricing insurance risks, automating claims approvals, and detecting fraud risks.
For companies in regulated sectors, issues of risk, trust and governance are high on the agenda. They need to be able to explain how an algorithm decides someone is a fraud risk or why a loan application was refused. Today’s solutions enable a company to interpret, explain, and trust ML models. They mitigate bias and manage risk.
The promise of getting a working AI system up and running within weeks is a game-changer for AI. But just as importantly, introducing an automated DSML lifecycle enables the business to democratise AI and put this powerful tool in the hands of more people, including business users. That will be the key to accelerating AI adoption and unleashing its full value in the years to come.
By Bryan McLachlan, Managing Director: Africa at CyborgIntell
