Even if you're relying on partners, you need an AI owner

Take an iterative approach when beginning AI investment, says Andy Gray of Kortical

Technology partners are great: they take the management of complex and time-consuming tasks out of your hands, allowing you to focus on your business. However, relying on them for artificial intelligence is a mistake in the long run, said Andy Gray of Kortical at Computing's AI & Machine Learning Live event this week.

Investment in AI is scaling up quickly. Last year, just five per cent of firms listed on the London Stock Exchange had an AI programme: this year, it was 61 per cent. Understanding the technology is crucial to building the business case for these investments, which you will struggle to do if a partner is handling it all.

Gray advised delegates to nominate or recruit an AI owner who can build a ‘centre of excellence' across the business. The CoE will needs two things to succeed: a mandate to access any data, and its own budget.

"We find that the biggest blocker for a lot of organisations in adopting AI is internally organising access to data," said Gray.

Data (and money) drives AI programmes, and having a central data lake is an excellent starting point to take advantage of the technology. It is not a necessity, though, and could even stand in the way of progress.

"There is a large investment [in a central data store] before any payoff; there are political challenges - as soon as you announce intent to have a single data solution, all parts of your organisation will fight to be that solution; competitors who go straight to AI will have years of advantage; and if you're not building it with the business case in mind, you might be collecting the wrong data."

A better way to get started with AI quickly is to take an iterative approach, with a small initial investment and gated proof points, so that the investment doesn't proceed without proof of success.

Gray advised beginning with an AI roadmap, working out how AI can be applied to the revenue and cost centres in your organisation. The next step is to begin proofs of value on historical data, then testing performance against the most recent data - proving that the AI model will do what you want it to.

In this model, the large investment comes at the end once the business value of AI has been proven and the return on investment understood.