Vision AI: pioneering the future with the University of St Andrews

Seeing beyond the obvious

Picture the daily realities Vision AI is built for.

In a bustling mountain resort: a winter morning can escalate quickly. An unexpected fall on an unmonitored slope, or lift queues spiralling into unmanageable crowds. Some create real risk, some degrade the customer experience. What if you could identify hazards pre-emptively, detect unusual activity instantly across vast terrain, and direct resources before incidents arise?

In a high-stakes security environment, an airport terminal or a casino: personnel scan countless screens for a single subtle anomaly. The struggle is not just seeing, but distinguishing genuine risk from everyday noise and acting with precision. What if your systems could cut through that noise, freeing expert teams to focus their judgement on intervention rather than exhaustive surveillance?

In a fast-paced restaurant: Managers are often in a constant battle against operational friction. An unattended checkout, a forgotten table in a busy section, or an uneven distribution of staff can lead to lost revenue, frustrated customers, and a drain on profitability. These seemingly small inefficiencies accumulate. What if your existing cameras became an intelligent observer, giving managers real-time insight into customer flow, bottlenecks and staffing, so they can adjust as service unfolds?

These are not only business problems; they are human ones, touching safety, efficiency and satisfaction. Vision AI was designed to turn raw visual data into clarity — the predictive insight needed to get ahead of these hurdles rather than chase them.

The view from the course

Most teaching cases are finished objects: a bounded problem, clean data, and a settled answer in the teaching note. Even the more open ones - a market assessment, a marketing strategy -tend to run alongside a company already on a steady trajectory. The analysis is real enough, but the ground does not move while you stand on it.

This engagement was the opposite, by design. Vision AI was live and still being shaped, with no tidy market to analyse and no fixed application to defend. The technology was developing rapidly, redefining its own assumptions as it went and forcing students to keep pace. For a strategy course, that was the whole point.

It changed how the module had to run. The reading list and the weekly focus needed to be adapted at a short notice and then as we went. There were no clear answers to students questions to hand: instead, more questions and discussions. We were not working with a successful or failed case to analyse the ‘factors’. The skill being taught was not how to apply a framework to a clean problem; it was how to keep reasoning when the problem keeps moving, hot to match adoption dynamics with technology and regulation on the move.

The instinct, from an innovation strategy course vantage point was to push towards adoption: where and who would realistically take Vision AI up; what stands in the way. Fujitsu could have insisted on its own framing, a single market analysis, or on telling only the success story: instead, we met in the middle. That willingness to adjust, on a live product, at both ends, is what kept the case so real for our students.

What the students found

Handing a live product to a sceptical audience is uncomfortable by design, and useful for the same reason. The teams did not treat Vision AI as a finished solution to admire; they challenged its assumptions, questioned its ethics, and stress-tested it through the pre-mortem exercise.

The ethical questions came up early and stayed central. Behavioural monitoring in workplaces and public spaces raises issues of privacy, consent and trust. Beach safety monitoring seemed easier to start with – but challenges quickly diverted the attention to ski slopes instead. For us at Fujitsu, having those concerns put sharply by people outside the project was exactly the value of the exercise. Responsible deployment is a condition of adopting this technology at all.

The pre-mortems made those issues concrete. Each student took the strategy their team had developed, imagined it had already failed, and worked backwards to explain why. Although working independently, taking different perspectives (a finance lens, a head of security’s chair, a restaurant manager’s floor) the diagnosis converged: the technology was rarely the problem. Trust, measurable value, and fit with how people already work proved more decisive. One analysis argued that in a high-security setting the real question was never whether the system could generate insights, but whether those insights could be trusted enough to act on: a false alarm wastes resources, a missed signal is a genuine threat. Another imagined a restaurant deployment abandoned within weeks. Not because it failed technically, but because alerts during a busy night without a complete redoing of the fabric of activities just did not work.

The pre-mortems, as a whole, mapped something a product roadmap rarely does: not whether Vision AI works, but whether the people and organisations around it would let it.

Vision AI for strategic advantage

That scrutiny does not diminish the opportunity; it sharpens it. Across mountain resorts, high-security facilities and dynamic retail settings, Vision AI delivers on a few core strengths:

Unveiling the invisible. Human observation is finite. Across a wide slope, a factory floor or a crowded terminal, Vision AI watches large-scale visual streams for the subtle anomalies and pattern shifts that precede an incident — the fall, the micro-fracture, the unattended bag — turning reactive damage control into proactive intervention.

Creating responsive environments. It converts static spaces into adaptive ones that read context and movement: the restaurant that notices a table going unserved, the resort that redirects staff as queues build. Environments that anticipate, rather than merely record.

Augmenting human potential. Far from a substitute, Vision AI is a strategic co-thinker — surfacing information and flagging risk so that the security team, freed from scanning every screen, can spend its judgement where judgement matters. Designed ethically, this is human-AI collaboration, not surveillance: a distinction the St Andrews work insisted we keep in view.

The path to value

Our approach to Vision AI is deliberately staged and evidence-based. We integrate with the systems already in place — the existing CCTV and access-control infrastructure — to minimise disruption and protect current investment. We begin with a low-cost Discovery Pilot on a single high-impact use case, move to a Proof of Value phase that demonstrates measurable business impact, and scale only once results are evidenced. What the collaboration added was a discipline we now treat as non-negotiable: agree with the customer, up front, what success will look like. The absence of those metrics was one of the clearest failure modes the students identified.

The future of competitive advantage is here, and it is looking back at us through the lens of AI-powered cameras. Vision AI turns passive infrastructure into active, intelligent observers that surface issues and insight as they happen, across security, retail, manufacturing and beyond. But the lesson of working with St Andrews is that the hardest part of an innovation like this is rarely the engineering. It is earning the right to deploy it well — and that is a question worth getting right before the technology, not after.

If you're ready to move beyond traditional surveillance, transform raw visual data into predictive insights, and secure your competitive advantage, we invite you to partner with us on this journey.

Contact us at: https://mkt-europe.global.fujitsu.com/Fujitsu_Thought_Leadership_Contact