Where AI's productivity revolution will strike first

As large cloud and software providers race to unfurl new enterprise AI offerings, a new McKinsey report says banks and retail are the business sectors first in line for the biggest boost from generative AI.

• In these and other companies, 75% of the productivity gains from generative AI will come from just four business functions: customer operations, marketing and sales, software engineering and R&D.

Why it matters: AI providers are starting to deliver specific tools to transform predicted AI productivity gains from theory to reality.

• The higher the pay and education associated with a work function, the more susceptible it will be to AI's impact, according to McKinsey.

Driving the news: Oracle on Tuesday announced an end-to-end platform for generative AI services, a day after Salesforce launched its Cloud AI offering.

• Both are branded as privacy and security-focused and are designed to appeal to regulated industries.

What's happening: McKinsey studied 63 practical uses of generative AI across 2,100 existing work activities to assess its potential.

• McKinsey expects activities that consume about two-thirds of our work time today to be automated over the next 20 years.
• A key example is R&D, where the emerging field of generative design — turbocharged by foundation models that enable faster drafting and testing of products and services — will provide big efficiencies, per McKinsey.

By the numbers: Generative AI could deliver total value between $2.6 trillion to $4.4 trillion a year, a sum greater than the GDP of Germany, the world's fourth-largest economy.

• Retail and packaged consumer goods companies would be in line for $660 billion a year in productivity gains, if "use cases were fully implemented" — which would mean a 44% boost to profits.
• Banks, which tend to have higher profit margins, wouldn't receive as big a boost to profits (9% to 15%) but stand to gain up to $340 billion annually.

Of note: McKinsey held back from attaching a timeline to these gains — a spokesperson said, "How quickly that value will be captured is dependent on how many and how fast companies decide to implement these use cases. It will undoubtedly take years."

Yes, but: While innovators and researchers agree on the huge potential of generative AI, even bigger dividends are possible from non-generative AI — the basic machine-learning and analytics technology that has been steadily evolving for the past 15 years.

• McKinsey estimates extensive use of non-generative AI would deliver $11.0 trillion to $17.7 trillion economic value, roughly four times that of generative AI.

• These rosy scenarios come with a warning: "Significant human oversight is required," and there's no replacement for human "strategic thinking specific to each company’s needs."

What they’re saying: Michael Chui, partner at McKinsey Global Institute, which prepared the report, told Axios he was most surprised to see generative AI’s impact “is highest for occupations which have the highest educational requirements and wage rates.”

• Antonio Neri, CEO of Hewlett Packard Enterprise, told Axios that much of realizing AI's potential pivots around making better use of data. Companies today are "exploiting only about 10% of the data they're accumulating," he said.
• Clara Shih, CEO of Salesforce AI, sees new enterprise AI applications as "democratizing generative AI" by lifting up the standards of work all staff can deliver.
• Using the example of marketing and sales emails, Shih said it's now possible for "every seller in your organization, every service agent, every marketing manager" to prepare text at the level only the "top few percent" could previously.

SAP chief marketing officer Julia White told Axios that generative AI is converting the notion of 1-to-1 marketing from theory to reality. "We can start delivering on that idea in a way that never would have scaled before," White said.

• White said her team is conducting a number of internal pilots, including one that suggests to sales people which demos might be most appropriate using characteristics such as a prospect's location and industry.