AI Adoption in Business Reaches a Turning Point. What's Next? Companies are seeing increasing value in artificial intelligence, but now it's more vital than ever to invest in operation and infrastructure systems that can react and adapt to changes in a sustainable way.

Achieving success in AI adoption was never going to be easy, but the past few years have brought significant progress, with McKinsey & Company reporting that many businesses are starting to see the value, including impact on revenues. Put simply, we may be entering a new phase on the artificial intelligence journey.

In 2020, there was actually "no increase in AI adoption," according to McKinsey's The State of AI (published in November of that year); rather, companies were "capturing value from AI at the enterprise level" in terms of revenue and cost reductions, with some even attributing 20% or more of their earnings before interest, taxes, depreciation, and amortization (EBITDA) to AI.

Following the Covid-19 outbreak, PwC has found that 52% of businesses accelerated their AI adoption plans, with 86% saying that it was becoming a "mainstream technology" in 2021. Many companies are now embarking on AI journeys, but there's a difference this time: they're not going into it blind. It's now a more familiar technology, and rather than being a "bright, shiny object," is becoming central to organizations.

The problem with AI adoption

The journey to achieving full value from AI will be longer and messier than in most technology transformations. The data, tech and talent involved will impact various functions across your organization, as well as those you partner with.

Its adoption is an ongoing process because AI function, as well as the data being fed into it, needs to be monitored throughout various phases of development, deployment and ongoing adjustments. Machine "learning" does exactly what it says on the tin: As data is added and changed, the AI learns from this information and it changes. Therefore, continuous adjustment and improvements are needed.

As businesses begin to understand and value AI, the key challenges so far have been around deployment, but we are beginning to see these resolved systematically. We now also have a few success cases and good examples for companies to learn and absorb best practices from, but it doesn't stop there.

Related: 5 Things Business Leaders Must Know About Adopting AI at Scale

Stages of adoption

Traditionally (if you can actually use that word in this context), an AI solution has three stages: planning, building or modelling and bringing to production. Now, it's time to focus on a fourth stage, and perhaps the most critical: operation. This is where assuming static data in the lab causes problems, because real-world AI solutions will have to deal with dynamic data, which can shift and change, so stability and robustness are vital for engineering teams to consider. The operation phase is also a key time for ensuring stability.

Stable and sustainable AI solutions require more than a model in production: In the operation phase, classic components should include monitoring abilities, observabilities, dashboards, feedback mechanics, data annotation and more. At more advanced stages, the operating team needs to think about retraining models and deploying them in the production environment, as well as advanced data screening, noise-handling and bi-directional feedback between the AI and the user.

The next phase

The operational stage will be critical over the coming two to five years. Those companies that have already planned, built and deployed successful AI models must now invest in maintenance and operation. Only with live feedback, dynamic data, continued testing and growth in the real-world environment can it continue to make an impact.

Related: What Fraudsters and 'Black Swans' Have in Common, How AI Can Mitigate the Effects of Both

Going forward, there will be an increasing need for the tools, products, methods — and crucially, people — to operate AI. This must be looked at across entire organizations so that teams can react to data changes in a scalable way. Looping in a whole data science team for a three-month project every time there's a shift in data is simply not sustainable; after all, the purpose of AI adoption is to automate processes and make life easier — not to use more manpower and cause more problems.

We're moving to a world where it shouldn't take 18 months to bring AI to production and operating it shouldn't be a hassle. The solution? Investing in operation and infrastructure and building an operation suite that can react to changes in a sustainable way.