One Quarter of Insight in Five Minutes

Over the past quarter we’ve spoken with operators from across the UK’s mid-market AI ecosystem. Here are the key messages from those interviews, bundled so you can digest a quarter’s worth of insight in five minutes.

Aaron Asadi, Enterprise Nation Small businesses were sold a consumer product and never shown what implementation looks like, so most stop at the surface with the same spreadsheets and the same systems with a chatbot bolted on. 79% of members use AI in some form and 84% say they’re ready, but nearly half lack the expertise to do anything meaningful.

Alison Wright, Microsoft UK AI has already entered most organisations through employees, so the sequence is visibility first, approved alternative second, policy third. Microsoft research puts 71% of UK workers on tools their employer hasn’t approved, with nearly one in four entering finance-related data.

Anca Pintilie, formerly Amazon EMEA “The AI is almost never the thing that’s broken”: the failure sits in unmapped operating models. Six teams and 300 to 400 people were running 13 different operating models that had gone unexamined.

Ben Lee, Bidwells AI scales the repeatable document-generation layer of professional services, while the tailored advice clients pay most for sits above that layer and doesn’t scale. He expects AI assistants to reproduce the “Power Platform sprawl” with inconsistent dashboards nobody can locate.

Dr Laura Weis, WPP The distinctive cost is decision doubt, where AI returns twenty confident answers to one brief and teams lose both the ability to choose and the confidence they chose right. Correction load then lands on strong performers, which is where the saving reverses.

Ed de Minckwitz, ServiceNow UK People who are AI-native have worked out exactly how much time they save and have no incentive to say so, because declaring 40% more productivity invites 40% more work.

Josh Clement-Sutcliffe, Zencargo They targeted the happy path in a business that has no simple processes, and it failed on context and variable count. Having fixed the data foundation, the constraint became noise: operators receiving too much information “blank it out”.

Kerri O’Neill, Ipsos Write the philosophy before naming the tools, and train explicitly for scepticism, because research findings on cognitive surrender show people accept AI output uncritically and grow more confident while doing it.

Max Haining, 100 School Access is not activation, and stacking more tool-and-feature workshops produces overwhelm rather than behaviour change. His learning sequence focuses on imagination, then fluency, then judgment, with the last one absent in most organisations.

Paul O’Sullivan, Salesforce UKI “FOMO will get you to a pilot” but only intent produces results, and the fix is scoping data to what one use case needs rather than cracking the whole estate. Focused customers report resolution rates above 80%; Salesforce reports 84% and $100m annualised savings on itself.

Rana Gujral, Behavioral Signals “What does the system do when it’s wrong? Not how accurate it is, not what’s the benchmark score. What happens on the bad day? If you ask about rollback and the answer is a pause followed by ‘well, in practice that hasn’t really come up,’ walk away. If you can’t undo an automated action, you don’t have oversight. You have hope.” Rana also names verification, not prompting, as the skill about to be scarce, and says it comes from reps.

Ross Nichols, Just Move In They inverted the build order so AI writes the code and humans review it, and rebuilt the whole product into an AI-driven platform. The old digital-plus-phone-calls setup couldn’t support tailored financial products or long-term relationships without hiring a large human team. Now they’re running roughly 20,000 moves a month and scaling into new geographies that weren’t economical before. Of note: new joiners with the right mindset are pulling the rest along.

Rupert Kemp, Google DeepMind The starting point is a question rather than a strategy, and deployment should obsess a core user group instead of chasing the whole organisation at once. He cites aibl’s own figure: leaders who start almost every task with AI report 71% ROI.

Sara Maldon, Make Adoption is a staffing and structure problem, solved by full-time embedded people in every department rather than tooling or training, and she deliberately runs a high pilot failure rate because failed projects locate the real friction points. She refuses time saved as a success metric, and concedes that 96% of employees building automations measure breadth, not depth.

Tim Flagg, UKAI The break point is the manager layer, whose job has changed most: reinventing throughput targets, designing human-in-the-loop checkpoints, and connecting tools to today’s tasks. His sharpest mechanism is the unfair test, where pilots fail for structural reasons and leadership concludes AI failed.