The AI mistake companies are making — and how they can fix it, according to a BCG tech leader

Like a archetypal day that's gotten awkward, immoderate companies struggling to win astatine artificial intelligence mightiness beryllium trying excessively hard.

They mightiness instrumentality connected excessively galore projects oregon neglect to recognize that AI windfalls often travel from rewiring however radical work, not from "super-cool" AI engines oregon large-language models, said Sylvain Duranton, planetary person of BCG X, the tech physique and plan part of Boston Consulting Group.

Those types of missteps tin balloon into big-time frustrations for concern leaders, helium told Business Insider.

Duranton said that if CEOs' large question astir AI successful 2024 was which exemplary to use, their inquire successful 2025 is "Where's my money?"

Indeed, helium said, determination are often challenges astir implementing wide usage of AI.

"Scaling this happening from a tech standpoint — it is hard," Duranton said.

To assistance companies salvage their AI efforts, helium said, his "golden rule" is that organizations allocate astir 10% of their effort and wealth to the algorithms — to physique AI engines oregon bid LLMs. Another 20% should beryllium reserved for information and technology. Essentially, that's to marque the AI enactment successful a company's tech environment, Duranton said.

The bulk of the effort — the remaining 70% — should spell to changing the mode radical work, helium said.

"Assuming you person a exertion that tin scale, you request to bring that into the hands of the people. It's a monolithic alteration effort," said Duranton, who's based successful the company's Paris bureau and oversees BCG X's planetary service of astir 3,000 technologists, scientists, programmers, engineers, and others.

Some companies are struggling

Companies' vexation is real. In the last months of 2024, BCG surveyed immoderate 1,800 C-suite execs from large companies successful astir 20 countries and recovered that portion 75% of respondents ranked AI among their top-three priorities, lone 25% reported seeing "significant value" from the technology.

To find much value, Duranton recommends that companies not effort to bash everything each astatine once. He said the scope of alteration that companies request apt can't beryllium achieved with tens oregon hundreds of usage cases.

"That's not the plan. The program is to absorption connected a precise fewer things, and the things that matter," helium said.

Duranton said companies sometimes besides look to "incremental initiatives." He said helium thinks that's often a mistake. Instead, helium said, companies should location successful connected a fewer "quintessential" things.

For a retailer, Duranton said, this mightiness beryllium utilizing AI to guarantee a brick-and-mortar store has the cleanable merchandise premix for that determination to amended withstand competitors adjacent and online.

Retail CEOs recognize the stakes, Duranton said. They'll often archer him thing like, "I cognize that if I don't bash that amended than others, I'm cooked," helium said.

Duranton said different imperative for a retailer mightiness beryllium to make an AI cause that tin store for customers — 1 that's truthful bully that users won't privation to power to a competitor.

"With those 2 things, you person some a strategical docket and an AI agenda," helium said, referring to making definite inventory is dialed and the buying bot.

The trick, then, is to support the absorption connected those efforts, Duranton said.

"Those quintessentials, that's wherever you enactment each your money, each your energy," helium said. That's necessary, Duranton said, if companies privation to instrumentality the 10%, 20%, 70% attack helium recommends.

AI's Bermuda Triangle

He said scaling AI is besides often hard due to the fact that companies tin consciousness unit to compromise connected expenses, prime of results, oregon the velocity astatine which they're produced.

"You person a benignant of Bermuda Triangle, wherever it is either costly and applicable with a bully latency, oregon you person to compromise connected 1 of the three, and you person to optimize," Duranton said, referring to AI results.

He said it's often casual to show immoderate tech wizardry successful a demo. What's difficult, Duranton said, is handling millions of requests each time and producing timely and applicable results.

"It's a antithetic ballgame," helium said.

Ultimately, to win with AI, Duranton said, companies volition request to bring on people, not conscionable bots.

"Invest successful change-management, not conscionable technology, and person your fearless and strongest leaders beryllium successful charge," helium said.