Intro: AI can flag a fraudulent transaction before a customer even notices. It can order thousands of support tickets in the time it takes a human to read one. It can highlight the right clause in a 200-page contract within seconds, or personalise a digital storefront for every visitor based on their behaviour in real time.
What it can’t do reliably is stay consistent. In many organisations, AI is already productive and commercially valuable. The problem is that it’s moving fast – from pilot project to embedded system – and in doing so, it’s becoming part of the experience layer itself, determining what customers see, what employees are told, and how decisions are made. The question is: can people rely on the systems built around it?
Most organisations aren’t short of ideas, pilots or tools, but they don’t have the architecture to sustain them. It’s not how much AI you deploy, it’s how well you govern what you’ve built.
This means increased risk. A chatbot that gives the wrong policy advice, a model that misreads an insurance claim, or a recommendation engine that amplifies bias are all still commonplace reliability failures. Errors wear down trust, which is the bedrock of any customer relationship.
Accenture CEO Julie Sweet made a similar point in a 2025 TIME interview, arguing that dependability is fundamental to digital transformation, and that without it, companies will never move beyond pilots. This is an important point because it shifts the conversation away from novelty and towards stewardship: responsible AI and human-centred design determine whether an organisation can manage change without heightening risk.
It’s no wonder The Traitors TV series is such a hit. It taps into a very modern anxiety: the feeling that someone else knows something you don’t, and that you can’t quite trust what you’re seeing. That dynamic plays out in enterprise technology. In business, you can’t afford to rely on a hunch. Confidence has to be engineered through transparency, verification, and the steady, repeatable signals that prove a system is actually doing what it promised. In that sense, trust is an operational requirement.
The same applies to authenticity. A familiar image, a human detail and a personal cue can make someone feel more credible. But what if that image is AI-generated? What if the emotional signal is synthetic? In an age defined by deepfakes, misinformation, and increasingly persuasive machine-generated content, authenticity cues are increasingly fabricated and we can’t base our trust on feelings – it is a design issue that must be verifiable, intentional and built into systems.
Teams need to decide if AI is safe to use by asking: Which use cases are allowed? What data can a model see? When does a human have to review an output? How are decisions tracked and checked? Who is responsible if an AI‑assisted process causes harm? How are prompts, models and rules changed over time?
Most organisations are layering AI onto estates built from legacy applications, fragmented data, and overlapping systems. But the problem is rarely the age of the technology, it’s about accumulation: integration sprawl, hidden debt, and too many disconnected components. Deloitte’s CTO Bill Briggs cites data showing 70% of technology leaders deem technical debt, the “quick, messy fixes” in their systems, as a hindrance to productivity. He told Fortune that organisations spend 93% of their AI budgets on technology and just 7% on the people expected to use it.
MIT, Gartner and Harvard Business Review all identify integration and data quality as recurring weaknesses in AI rollouts.
McKinsey’s 2025 State of AI report found 88% of organisations now use AI in at least one business function, but nearly two-thirds haven’t begun rolling it out across the enterprise. As a result, Gartner estimates that half of generative AI projects were abandoned in 2025 after proof of concept – not because the models were inadequate, but because the infrastructure around them was insufficient.
This leads to data misuse, security failures, and weak transparency, which cause reputational damage, regulatory exposure, and direct commercial loss. In finance, insurance, and e-commerce, confidence is fundamental to the operating model. In public services, it underpins adoption and channel shift. Strong internal controls don’t just reduce risk, they build credibility by demonstrating resilience and accountability to customers, partners, and regulators.
Guided automation is often a more honest ambition than full automation. In high-value or high-risk processes, keeping humans in the loop isn’t a limitation — it’s the design. AI handles speed, relevance, and efficiency; human judgement covers sensitivity, context, and accountability. That holds across financial services, the public sector, and any large enterprise where productivity gains mean little if employees don’t trust the systems they’re working with.
The most important AI design question may not be “what can we automate?” It may be “what must we govern?”
Transformation used to be about delivery; now, it’s about dependability. Technology can automate almost everything, but it make people trust it. That comes down to design, engineering, governance, and time.
