The C-Suite & AI: Why leaders need to stop following the herd

Before a recent study by AI research group METR published its findings, most of the developers involved predicted they would be around 24% faster working with AI assistance. The actual result: they took 19% longer. Afterwards, before the data was revealed, the majority still believed they had been roughly 20% more productive.

That gap between perception and performance is not a niche finding about software developers. It is a reasonable description of where a lot of C-suite AI decision-making sits right now, and it is why senior leaders need a clearer framework before they sign anything.

In this webinar, Ant Burton, who leads product, data and strategy at Data & AI Literacy Academy, spoke with Monica Jones, former Chief Data Officer at the University of Leeds and founder of Animo Consultancy, about the leadership thinking behind AI strategy. Jones’s career spans 17 years as a British Army officer, 15 years in NHS digital and data leadership, and senior CDO roles in healthcare research and higher education. The conversation covered four questions that every C-suite leader should be able to answer before committing to an AI programme.

Why AI herd mentality is rational for leaders, and why it backfires

There is a sociological term for what happens when organisations under pressure look sideways and replicate what others appear to be doing: mimetic isomorphism. In uncertain conditions, copying what looks successful elsewhere is not irrational. It is, in fact, a well-documented human response to ambiguity.

The problem is that it transfers the wrong signal. When even the people using a tool cannot tell whether it is helping them, the leaders copying those rollouts from the outside are working with a particularly unreliable read.

Jones put it plainly: “Don’t adopt AI simply because everybody else is. Be clear about the organisational challenge you are trying to solve, and whether AI is actually the right answer.”

The pressure to be seen doing something is pushing leaders to make decisions they otherwise may not. Boards are asking for AI strategies, suppliers are pitching transformation and executives are approving spend. The question that tends to get sidelined is whether any of it solves a genuine problem. As Jones noted, having an AI strategy on paper, launching pilots, and buying new tools is not the same as delivering value. Activity and progress are not the same thing.

Why data foundations must come before your AI strategy

Poor data does not become useful data because AI has been applied to it. This is probably the most common and costly misunderstanding in enterprise AI programmes right now.

When Jones joined the University of Leeds as CDO, there was no central data warehouse and no data function with dedicated resource. Before any AI strategy could be credible, she put three things in place: a genuine data strategy, a central data service staffed as a day job rather than bolted onto already busy people, and a recurring annual budget to sustain it. The AI work came after those foundations existed, not before.

This sequence matters. A Gartner study of over 1,200 senior leaders found that more than two thirds lacked confidence in their data practices to meet their AI needs. Gartner’s forecast is that over 60% of AI projects without an AI-ready data foundation will be abandoned by 2026.

For any leader who has already committed budget to AI, that figure is worth pausing on. If the data foundation is not in place, the question of how to defend those decisions becomes live before the programme has delivered anything.

Jones’s approach at Leeds was to benchmark before building. Using UniForum, a sector-wide tool that measures organisations against peer benchmarks, she established where the university actually sat on administrative staffing and operational efficiency. That gave the AI and automation programme a concrete starting point and a measurable target. The initial use cases were not glamorous: meeting summaries, routine reporting, standard Copilot applications. But because there was a baseline and a clear business problem, progress could be measured rather than asserted.

Her earlier work at Rotherham NHS Trust followed the same pattern. A central data warehouse, built in-house because external procurement was not affordable, eventually drew in GP data, local authority data, mental health data and a longitudinal patient record. The result was a CQC rating improvement from requires improvement to outstanding for digital in three years, driven entirely by data visibility and access, not by AI in any contemporary sense. The lesson she took from Rotherham to Leeds was the same: get the foundation right, then build on it.

The AI halo effect: why polished outputs can mislead senior leaders

AI produces fluent, structured, confident-sounding output. It takes messy material and returns clear summaries. It presents recommendations in a format that reads as authoritative, regardless of whether the underlying reasoning is sound.

In behavioural economics, the halo effect describes the tendency to judge the substance of something by how it is delivered. Tall, symmetrical, vocally confident people are evaluated as more competent, independent of content. A well-formatted AI output carries a version of the same effect, and that problem is amplified at senior levels where challenge is already less likely to land.

For C-suite leaders, this creates a specific risk. The same dynamic that makes it harder for junior colleagues to push back also makes it harder for a room to notice when an AI-generated recommendation has absorbed the biases in its source data and returned them in a more convincing format.

Jones’s experience navigating this was partly structural and partly cultural. At Leeds, a data analyst network grew to between 400 and 500 self-identified data analysts spread across every part of the university. That network gave a large number of people the knowledge and the forum to challenge what they were seeing, rather than defaulting to whatever a central team or supplier produced.

She also pointed to a case from her time at Cancer Research UK, where she challenged the standard approach to clinical data collection in the Stratified Medicine Programme. The convention was to build a fully bespoke database for each clinical study, built precisely to the study protocol. Jones argued that routine NHS data could do most of the work faster and with less burden on the people providing it. Sustaining that position required significant advocacy against institutional convention. It worked, and the approach became a precursor to what is now standard practice through Genomics England.

The common thread is not technical expertise. It is the willingness to hold a position long enough to test it, and to build environments where others feel safe doing the same. Burton referenced David Epstein’s book Range on the value of generalist thinkers in novel problem spaces. AI sits firmly in that category: the people most likely to spot when a confident-looking output is wrong are often not the specialists, but the people with enough breadth to notice when something does not add up.

AI literacy vs AI training: what C-suite leaders actually need

Training and literacy are not the same thing, and the distinction shapes whether an AI programme embeds or stalls.

The data Burton cited is instructive. Organisations that provide at least five hours of AI training see around 80% of staff become regular users. That figure falls sharply with less. But training alone does not determine whether AI becomes genuinely useful: a separate study found that positive reception of AI across organisations rose from 15% to 55% when there was visible leadership support for it, beyond the initial mandate.

Literacy, as Jones defined it, applies at every level. Leaders do not need to become technical experts. They do need enough understanding to ask intelligent questions of suppliers, challenge assumptions in AI-generated outputs, and recognise when they are being sold something that will not work. That requires some investment in their own AI literacy, not as a compliance exercise but as a prerequisite for making good decisions.

For the wider workforce, the risk Jones named at Leeds was real: staff felt that students knew more about AI than they did, and that left people anxious and avoidant rather than engaged. The university responded with a combination of the analyst network, apprenticeship levy-funded qualifications in Data Citizen and AI Champion pathways, and a clear green/amber/red framework setting out what AI could and could not be used for. People need to know what is expected of them, and they need to believe the organisation is investing in their capability rather than simply mandating adoption.

On governance: AI accountability sits too often in IT functions because that is where the technical knowledge lives. Both Burton and Jones were direct on this. Delivery can have a technical home. Accountability for AI decisions has to be business-led. At Leeds, the AI strategy was led from an academic perspective, working alongside the data function Jones headed, because the problems it was designed to address were academic ones. The function follows the problem.

Frequently asked questions

What should a C-suite leader do before implementing an AI strategy?

Start with a clearly defined business problem rather than a technology decision. Establish what data your organisation actually holds, assess its quality and whether it is structured in a way that AI can use, and benchmark your current position against a measurable outcome. Having an AI strategy on paper is not the same as being ready to deliver value from it.

Why do so many AI projects fail in large organisations?

The most common cause is insufficient data foundations. Gartner forecasts that over 60% of AI projects without AI-ready data foundations will be abandoned. Poor data quality, a lack of governance, and no central ownership of data mean that adding AI does not improve the underlying picture; it amplifies existing gaps.

What is the difference between AI training and AI literacy?

Training gives people the skills to use specific tools. Literacy gives people the understanding to evaluate when and whether to use AI at all, to question the outputs they receive, and to make informed decisions about where AI adds value and where it introduces risk. Both matter, but literacy is what makes the difference at leadership level.

How do you challenge AI recommendations as a senior leader?

Build diversity of thinking into the room. Create conditions where the people most likely to spot a problem feel able to raise it, regardless of seniority. Be sceptical of polished output: fluent presentation is not evidence of sound reasoning. Invest enough in your own AI literacy that you can ask substantive questions rather than deferring to technical specialists by default.

Is AI governance the responsibility of IT or the business?

The business. Delivery functions can sit within IT, but accountability for the decisions AI informs, particularly where those decisions affect people’s jobs, services or outcomes, must be owned by the relevant business function. Governance that lives only in IT tends to become a compliance layer rather than a meaningful check on how AI is actually used.

Monica Jones is the founder of Animo Consultancy and former Chief Data Officer at the University of Leeds. This session was part of Data & AI Literacy Academy’s ongoing leadership webinar series.

‍

‍