1.Buying the tool before defining the process
- Why it appears reasonable
- AI tools are inexpensive, simple to trial and easy to demonstrate. Purchasing one is the most visible way to show that the business is acting.
- Why it fails
- A tool placed alongside an unchanged process leaves the cost and delay in that process where they were. In a survey of 3,235 senior leaders, 37% said their organisation uses AI at a surface level, with little or no change to existing processes.
- What it costs
- Subscription fees and staff time with no effect on the accounts. In a survey of nearly 6,000 executives in four countries, 89% of firms reported no effect on productivity over the previous three years.
2.Leaving adoption to individuals
- Why it appears reasonable
- Staff are willing, and many have already started: 55% of people in work report using AI for work or education, against 35% of businesses with ten or more employees.
- Why it fails
- Individual use is unseen by the business, inconsistent between colleagues and unconnected to any measure the business tracks. The gain stays with the individual task.
- What it costs
- The business has no visibility of the client and commercial data being entered into personal tools, and only 11% of the businesses using AI have trained more than half of their workforce.
3.Starting without a baseline
- Why it appears reasonable
- Early gains appear self-evident, and measurement can look like an overhead on a small project.
- Why it fails
- Without a figure recorded before the change, an improvement cannot be demonstrated, sustained or defended when budgets are set.
- What it costs
- In a survey of 1,719 respondents, 80% said AI had improved their own productivity and 37% attributed any effect on operating profit to it. In a separate survey, companies tracking AI value directly in the profit and loss account reported three times the value of those that do not formally measure it.
Five shifts in practice
Avoiding these errors amounts to five changes in how a business approaches AI.
| Instead of | Do this |
|---|---|
| Asking what an AI product can do | Asking which process is slow, unreliable or dependent on one person |
| Trialling many tools across the business | Rebuilding two or three core processes around AI |
| Leaving staff to find their own tools | Providing approved tools, training and a clear rule on what data may be entered |
| Judging success by how many people use AI | Judging it by a measure recorded before the change: hours, cost or days |
| Treating the purchase as the end of the project | Reviewing the same measure at fixed intervals until working practice has changed |
None of these changes depends on advanced or experimental technology. They are disciplines of management: establish an accurate picture, select a small number of priorities, build AI into the process, transfer ownership to the team and sustain the change.
Sources
- Office for National Statistics, Artificial intelligence in UK businesses: 2023 to 2026, July 2026.
- Yotzov et al., "Firm Data on AI", NBER Working Paper 34836, February 2026.
- The State of AI in the Enterprise, Deloitte, January 2026.
- The state of AI in 2026: on the road to ROI, McKinsey, August 2026.
- The Applied AI Index 2026: The Formula for Agentic AI Value, BCG, September 2026.
Full references are given in the report.