How enterprises can get the best out of AI solutions

Some hedge funds and PE/VC firms are making algorithms sit in the boardrooms, taking decisions on behalf of the board, or along with the board, says Sameer Dhanrajani, founder of consultancy AIQRATE and networking platform 3AI. That, he says, is thanks to AI machines crunching decades of data.
But getting there, he says, requires a lot of effort, and will be a long-drawn one.A marathon, he calls it.
Given the excitement surrounding AI, enterprises are seen to be either stunned by its advance and unable to decide what direction to take, or rushing into it and making investments that are not delivering the results they expect. What’s the best way forward? That was what we discussed the other day in a panel that included Sameer, Gaurav Makkar, engineering director at NetApp, and Prinkan Pal, founder of gen AI-powered analytics platform LegoAI.

Prinkan says it’s absolutely critical for enterprises to start off with developing an AI strategy that is unique to it. He says organisations must understand the specific business areas they need AI to provide insights and predictions in, and then go about collecting and organising the data required, and building the models. “The business requirement needs to get translated into something we call as analytical requirement, which then gets translated into an analytical solution. And then it gets translated into a final business solution, which is that automated solution that can be consumed by a non-technical business user,” he says.
Data is at the root of this. Gaurav notes digital transformation of operations is the first step towards collecting data. And once that data is collected, it needs to be organised in a way that analytical tools can be applied to it. Organising that data itself can be a humongous task. Data is of different kinds, including text, video, image. Each piece of data contains other information. A video, Gaurav says, is the core data. But then there is the metadata that describes things like where the video was taken, its level of resolution. And then there’s telemetry data – how long did a user play that video, did they press forward on that video, did they return to the video some days later.
“Different organisations play with each of these classes of data differently. A person in the sales organisation may be interested in how interactive the content is. Someone else may be keen to cache videos that are used more frequently. Someone may find the metadata showing people are playing on low-resolution devices, in which case there’s no reason to create a 4k video,” Gaurav says.
To make such decisions, you need to annotate the data, and today there are auto-annotating tools that are available that can analyse the content and understand what it is. “And then you link the data up, and the more you link up data, it becomes a source of intelligence,” Gaurav says.
In the enterprise context, Prinkan says, another element that needs a lot of thinking through is around security, compliance, and governance. Gaurav too underscores that. “Data can have privacy sensitive information. Should I be sending this data outside a certain domain? Who should handle this data? Ethics comes in,” he says.
Getting people in an organisation to adopt a solution too can be challenging, because of scepticism or fear of job loss. “Which is where I think human and AI intelligence have to work together to give desired results. Just producing glamorous AI systems will not work,” Sameer says.