Stop investing so much in A.I. technology. Invest more into what helps you create A.I. applications faster and easier

The percentage of companies adopting A.I. has been plateauing between 50% and 60% over the past few years. Meanwhile, some leaders have shared their frustration over not yet seeing the level of returns from A.I. investments that they expected. They simply haven’t been able to get it to scale.

Are some companies forgoing the pursuit of A.I.?

We hope not. Significant returns are being realized by leading companies. There is a contingent that is seeing substantial revenue boosts and 20% or more of their earnings before interest and taxes (EBIT) come from A.I. as they apply the technology to improve decisions, speed processes, and automate mundane, repetitive tasks, freeing workers to explore innovations that can vault an organization into entirely new business arenas.

These aren’t just tech firms—they’re companies in sectors from health care to mobility to financial services. And they plan to continue embedding A.I. in everything they do, widening their advantage over competitors that aren’t tapping A.I.’s potential.

What are these companies doing differently? They’re putting a significant chunk of their A.I. investments into capabilities that enable them to do more A.I., more rapidly—with less effort. They’re putting their dollars into building a veritable A.I. factory.

From one-off productions to repeatable processes

At the start of A.I. journeys, companies typically execute a smattering of A.I. pilots and one-off use cases to tackle business problems. At some companies, with each success, they do more of what delivered those victories: for example, building more entirely bespoke models and encouraging more experimentation in silos of their organization.

While it’s wise to continue some investment in these areas, doing more of what yielded initial results isn’t enough to make A.I. use more ubiquitous within your organization, which is the key to realizing maximum ROI.

As one-off productions multiply, so does the complexity and cost associated with teams starting every development effort from scratch, sifting through hundreds of raw data sets, writing bespoke code, and deploying their own development tools and technologies.

To scale A.I., investments must be channeled toward creating reusable assets, a platform, and repeatable processes (a factory, of sorts), which enable A.I. teams to build, deploy, and maintain models in less time and with less manual effort.

In our research, the respondents from organizations seeing the highest returns from A.I. are far more likely to employ this factory approach. And in our work, we’ve seen numerous organizations begin to get this right. Vistra Corp., the largest competitive power producer in the U.S., created an A.I. factory to standardize the deployment and maintenance of more than 400 A.I. models. It saw $60 million in savings in about a year and is on its way to delivering against a road map of $250 million to $300 million in identified Ebitda and a reduction in greenhouse gas emissions.

Similarly, an Asian financial services company was able to reduce the time to develop new A.I. applications by more than 50%, in part by creating high-quality, ready-to-use data products on top of data source systems that could be used in numerous A.I. applications. The company also standardized supporting data management tooling and processes to create a sustainable data pipeline, and it created assets to standardize and automate time-consuming steps, such as data labeling.

Inside the A.I. factory

An A.I. factory is not a physical place, but rather a framework for enabling teams to deliver more A.I. applications in less time and with less effort. It includes, for instance, providing development teams with easy access to the tools and technologies they need via one end-to-end platform that reduces time-consuming and costly integration efforts. It’s establishing one A.I. development and deployment playbook and a standard set of protocols, bringing together best practices from across the company so development teams can use them over and over. And it’s focusing efforts on creating a handful of ready-to-use assets—be it data products or snippets of code—to give teams a head start on the work.

Consider how simply reusing existing code can help drive several aspects of scale. As recently as a few years ago, most A.I. practitioners had to code every A.I. solution from scratch. Today, organizations can download state-of-the-art, open source, pretrained A.I. models to use in application development. We’ve seen, for example, one global energy company build a model that accurately predicted customer cancellations by applying just a few lines of code.

At the same time, emerging low-code/no-code platforms enable employees without extensive A.I. expertise to develop A.I. applications. Organizations seeing the highest returns from A.I. are 1.6 times as likely to be using these time-saving tools.

For code that still needs to be written in-house, leading organizations actively encourage a focus on reusability. They incentivize and celebrate the efforts of data scientists and engineers to contribute their code to a central codebase that other practitioners can easily tap into for other applications.

With reusable code available and nontechnical workers able to help develop basic A.I. applications, the most skilled A.I. practitioners are then freed to focus their expertise on engineering robust, high-value applications for the business. This increases their productivity, allowing organizations to do more with fewer of the most elite A.I. talent. It also keeps all A.I. practitioners a lot happier because each can focus on the most engaging tasks in their area of expertise, a key to their retention for the organization.

And, most important, it lets the organization redirect talent dollars and efforts toward other emerging roles required to build out the A.I. factory. One such role is that of the machine learning engineer, who is skilled in turning A.I. models into enterprise-grade production systems that run reliably and automate the machine learning pipeline, from data ingestion to the generation of business insights.

Now is the time to lean into A.I., even during this time of disruption and resiliency. The winning companies are doubling down and proving it’s possible to do more A.I. in less time and with less effort so they can realize the high returns the technology promises.

Alex Singla is a senior partner and global coleader of QuantumBlack, A.I. by McKinsey in Chicago. Alex Sukharevsky is a senior partner and global coleader of QuantumBlack, A.I. by McKinsey in London. McKinsey is a partner of Fortune’s Global Forum.