The Maze: B2B gen AI has crossed the point where “we are exploring it” is a useful status update. In 2025, 52% of surveyed organizations were either fully implemented or actively implementing gen AI in commercial processes. A year earlier, the comparable visible total was 42%. The shift is not a declaration that every workflow now works. It is evidence that the centre of gravity has moved from intention toward operating reality.
Implementation is growing while indefinite interest is shrinking. Fully implemented use rose from 19% in 2024 to 22% in 2025. The larger move came from organizations currently implementing but not yet organization-wide: 23% to 30%. At the other end, the share planning to use gen AI with no timeline fell from 16% to 9%, and the no-plans group fell from 7% to 5%. The near-term planning cohort stayed at 20%. The commercial backlog is beginning to convert into work.
Production puts a price on vague ownership. A sales assistant can draft an answer in seconds, but someone still has to define when it may quote a price, recommend a substitute, hand a customer to a person or stay silent. Those decisions cut across commercial, product, technology, legal and service teams. Implementation grows when that operating design becomes concrete. It stalls when every exception becomes a committee meeting.
The ten-point implementation gain is a stage change, not a value claim. Combining the two visible implementation states produces 42% in 2024 and 52% in 2025. That arithmetic uses rounded survey shares, so it is best read as an approximate ten-percentage-point change. It does not measure return on investment, model quality or revenue lift. It does say more B2B teams are putting gen AI into the processes where customers research, buy and get served, rather than leaving it inside a pilot deck.
The commercial use case is more demanding than a general chatbot. A customer-facing or sales workflow has to work with account data, product information, prices, permissions, inventory and a human handoff. A persuasive answer that cites the wrong availability or sends an account down the wrong path is not productivity; it is a new kind of leakage. That is why implementation is a more revealing measure than stated excitement. It forces teams to confront data quality, workflow design, guardrails and who owns the outcome.
The remaining distribution explains why adoption will not be a single moment. In 2025, 43% of respondents still sat outside implementation: 20% planned to begin within a year, 14% planned a longer wait, 9% had no timetable and 5% had no plans. Those groups are not interchangeable. Some are sequencing a real programme. Some are blocked by legacy systems or unclear economics. Some may be correctly cautious. Treating them as one pool of “not adopted” hides the decisions that determine who can turn interest into commercial capability.
Why it matters: Ecommerce leaders should stop tracking gen AI as a binary yes/no. The operational question is narrower: which customer or revenue decision has moved into a repeatable, governed workflow, and what evidence shows it improves speed, margin or experience? A company can have a dozen pilots and still have zero commercial adoption. Equally, one embedded use case that improves product discovery, sales preparation or service resolution can create the data, confidence and budget for the next one. That is how a pilot becomes an operating system.
