15 October 2026 | Ray Poynter
AI can make individuals more productive, but turning those gains into better results for an organisation requires leadership and a plan for adoption.
Many of us will recognise the problem. Someone finds a useful way to use AI, shows a few colleagues, and gets an enthusiastic response. A three months later, very little has changed. The question I keep being asked is how do we turn individual discoveries into benefits that are leveraged by the whole team?
A recent episode of The AI Daily Brief, featuring Nathaniel Whittemore and Nufar Gaspar, prompted me to look at the evidence. They argue that a well-run internal AI champions programme offers an exceptionally good return. That is their assessment from experience, rather than a finding from a comparative study. However, the wider research supports the case for taking adoption and leadership seriously.
Usage is growing faster than financial impact
BCG’s 2026 AI at Work survey covered 11,749 workers in 14 markets. It found that 74% of frontline employees used AI regularly. Here, ‘frontline’ means white-collar employees without managerial responsibilities.
Among regular frontline users, 42% reported saving at least a working day each week. Yet 66% said they received limited or no guidance on what to do with that time. Saving time is useful, but managers need to decide how it will improve the work or the business.
McKinsey’s 2026 State of AI survey tells a similar story. The proportion of respondents reporting that AI was scaling across their enterprise rose from 38% to 44%. However, the share reporting a positive contribution to earnings/profit remained broadly unchanged at 37%.
This does not mean the remaining organisations gained nothing. McKinsey also reports benefits within individual functions and improvements in areas such as customer satisfaction. It does suggest that wider use and individual productivity do not automatically produce a measurable improvement in company profits.
The organisation matters
Microsoft’s 2026 Work Trend Index surveyed 20,000 knowledge workers who already used AI at work, across ten countries. Its analysis identified organisational culture, manager support and talent practices as stronger predictors of reported AI impact than individual mindset and behaviour.
Only 26% said their leadership was clearly and consistently aligned on AI. Just 13% said they were rewarded for reinventing work with AI even when results fell short.
These are self-reported findings from AI users, published by a supplier of AI tools. The analysis shows associations, so it cannot establish that changing culture will cause a particular improvement. Nevertheless, it is a useful challenge to the idea that recruiting enthusiastic users and buying licences will be enough.
McKinsey also found that nearly three-quarters of its AI high performers reported fundamentally redesigning workflows, compared with about a quarter of other respondents. These high performers represented just 6% of the sample. Their experience is worth examining, while recognising that most organisations are in a different position.
Give champions a clear job
Champions can help colleagues turn useful experiments into everyday practice. GitHub’s account of its own approach stresses the importance of peer trust. Its advocates share examples, help colleagues get started and feed back what is working.
There are examples beyond technology companies. Business Insider reported in January 2026 that Citi had around 4,000 volunteer AI Accelerators, supported by approximately 25–30 AI Champions. This illustrates the scale such a network can reach, although it does not establish how much adoption the programme caused.
The AI Daily Brief’s practical advice is to define the role before choosing people. An ambassador who helps colleagues use approved tools needs different skills and support from someone building systems for others. Champions should know their team’s work, have their colleagues’ trust and have time allocated by their manager. Training should use real problems, with continuing support afterwards.
The time commitment depends on the job. GitHub suggests 30–60 minutes a week for volunteer advocates. The podcast recommends one or two days for its champion model. Those are different commitments, and organisations should be explicit about what they expect.
Write a practical adoption plan
For an insights team or research agency, I would start with seven questions:
- What do we want AI to improve? Choose specific outcomes, such as a faster proposal process, better checking of questionnaires or more time to explore alternative interpretations.
- Who will lead the work? Give someone responsibility, an agreed scope and enough time. In a small team, one champion may be sufficient.
- How will colleagues learn? Use current projects. A researcher should leave a session knowing how to apply AI to a task they need to complete.
- What must managers change? Review deadlines, expectations and how saved time is used. Otherwise, people may have little reason to change established habits.
- What are the rules? Specify approved tools, what data can be used and who checks the work. Make it easy to ask for a decision when a proposed use falls outside those rules.
- How will we share what works? Agree a simple way to demonstrate useful approaches, record them and help colleagues adopt them. Review whether they are still useful as tools change.
- How will we judge progress? Look at quality, turnaround times, costs and client outcomes alongside usage. More prompts and licences are weak measures of business improvement.
An opportunity for researchers
Research and insights professionals already have relevant skills. We regularly ask whether evidence supports a conclusion, whether a method is appropriate and what the consequences of an error might be. Those habits are valuable when deciding whether an AI output is fit to use.
They need to be combined with practical experience of AI and the ability to help colleagues change how they work. Being the best user in the team does not automatically equip someone to lead adoption.
My recommendation is to write down what AI should achieve, who will help the team get there and how you will assess progress. Then give that person the time and support to do the job.
I would definitely recommend listening to the podcast I mentioned at the start of this post.