AI at work in 2025. Lots of adoption. Not much effectiveness.
Over the past few weeks there’s been a flood of high profile reports on AI in the workplace. McKinsey, IBM, Salesforce, Microsoft, Ramp & others have all tried to answer the same question. Where are we really, as we head into 2026? To save you reading them all- here’s the TLDR:
Adoption is high. Value is patchy.
Around 90% of firms say they use AI in some way, but around 75% are still stuck in pilot mode. On paper, AI adoption looks like a success story. Almost every organisation can tick the “we are doing something with AI” box. But the bar is low. When you ask, “where have we scaled AI into multiple functions with measurable impact”, the pool shrinks fast. Lots of activity at the edges. Very little wired into the operating system.
“Which workflow, if we removed AI tomorrow, would meaningfully slow us down?”
“What is the single metric that proves it is working?”
If you can’t answer those for even one area, you have experiments, not effective use.
There is a real profit gap.
Roughly 75% of firms say AI is helping with innovation, but only 40% can see any impact on profit. Most workers say they feel more productive. However these personal time savings rarely show up as organisational value. Companies pay for licence & change programmes-the upside stays vague. AI that only speeds up tasks does not automatically move the profit needle. The firms seeing financial impact are the ones that have redesigned the work around what AI can do, rather than sprinkling it over existing steps.
Shadow AI is ahead of your governance.
More than half of employees are using AI tools without formal approval. You should assume AI use in your organisation is higher than any official policy suggests. People use whichever tool gets past the firewall and use it to fix their own pain points. That is not automatically a problem. Often it is where the most useful experimentation lives. The issue is that all of this sits outside your data strategy, security model & learning loop. You inherit the risk without capturing the insight. Knowledge stays with a handful of early adopters instead of becoming part of the operating system.
The real blocker is not models. It is the “data tax”.
Most Chief Data Officers now say AI is a top priority, but only a minority feel their data is clean enough to support AI driven revenue.The limiting factor now is rarely the power of the model. It is the state of data living in legacy systems, spreadsheets, shared drives, one off tools and old exports. Definitions are inconsistent. Ownership is unclear. Governance flips between too rigid & totally absent. In that environment, it does not matter how clever your model is, you wont get high quality results. High performers are doing something that looks boring but is extremely effective. They are spending money on making their data boring: clean, structured, accessible, and ready to plug into multiple workflows.
High performers think in operating systems, not tools.
The organisations seeing real gains from AI were over 3x more likely to have redesigned their workflows around what AI can do, V’s just adding tools on top. This is the real dividing line. Most firms are in a tool first phase. They buy licences, bolt AI into existing processes, and struggle to see real impact. The organisations seeing meaningful gains have shifted to an operating system first mindset. They break down end to end workflow’s and rebuild them on the assumption that AI can handle large chunks of execution.Then add human checkpoints in at key stages and treat data & process as shared infrastructure.
End-of-year TLDR
If we zoom out, we’ve moved from the hype phase to the friction phase. Almost everyone now “uses AI”. Very few do so effectively. Access is no longer the issue, it’s messy data, shallow pilots and a lack of operating-system thinking. Your advantage in 2026 won’t come from buying more tools. It will come from a real understanding of what AI can and can’t do well and an operating system 1st mindset.
McKinsey – The State of AI in 2025; McKinsey – Agents, robots, and us; IBM – 2025 CDO Study; Salesforce – State of Service 2025; Strategic Report / BCG – State of AI in the Workplace; Ramp – Business AI adoption flatlines