Once upon a time, companies did business with… paper. At least, that’s what I’ve heard – from the elders who survived the faxes, the carbon copies, and the infamous filing cabinets. Back then, contracts were mailed, signatures were wet, and ‘cloud’ meant something that ruined your picnic.
Fast forward: digitalisation and, amplified globalisation arrived. The two accelerated each other: as processes became digital, markets became global; and as markets expanded, companies digitalised faster. Organisations were suddenly everywhere, connected, and always on.
Today, we’re standing at the next tipping point: AI. Success will not depend on who builds the most advanced algorithms, but on who ensures that AI is usable, trustworthy, and secure. Because this time, the difference is clear: every breakthrough in AI begins and ends with data.
This summer, a mini buzz ran through the data and AI community. A report linked to MIT’s Project NANDA – The GenAI Divide: State of AI in Business 2025 – claimed that despite $30-40 billion invested in enterprise AI, nearly 95% of generative AI pilots fail to deliver measurable business impact and stuck at the starting line. These failures are rarely due to the AI algorithms themselves – more often, the culprit lies in the data and infrastructure behind the AI. The headlines of this were blunt: billions invested, yet only a handful of projects produce results that move the profit-and-loss needle. Sobering, yes, but not surprising for those who remember the early years of digitalisation, when most ambitious IT projects collapsed under weak legacy systems and poor governance.
The report made headlines with its 95% failure rate, but for practitioners the message was not new. The problem isn’t that AI is immature — it’s what lies beneath the hood of most large organisations: data debt, legacy technology, misplaced priorities, and weak governance. At PwC Luxembourg, we’ve seen the same patterns firsthand. The lesson is clear: AI success depends less on algorithms and more on the strength of data management and governance.
- Data debt. Data debt is what happens when organisations treat data as a by-product instead of an asset. Over years, shortcuts accumulate: missing lineage, undefined ownership, ungoverned copies, inconsistent definitions across systems. We are now paying a price for decades of underinvestment and underappreciation of data management (example with this 2016 IBM report). AI models are too often trained on fragmented, inconsistent, or low-quality datasets. The old saying “garbage in, garbage out” still applies, only now the garbage is being processed at machine speed and amplified across the enterprise.
- Legacy IT. Too many enterprises still run on siloed systems, brittle integrations, and architectures. This means new AI tools can’t access data fluidly. Pilots may shine in isolation, but integration fails at scale. It’s like trying to run modern apps on a 90s dial-up internet – the infrastructure just isn’t up to it… What was once “good enough IT” becomes a deadweight when AI requires real-time, enterprise-wide access to data.
- Shiny-object syndrome. Too many companies chase flashy AI demos (chatbots, marketing gimmicks) for short terms wins or communication value. Meanwhile, the research shows the biggest returns often lie in the less glamorous back-office functions such as finance, compliance, and operations where automation and risk reduction translate directly into measurable, or indirect, impact. In fact, a 2025 banking survey found 68% of AI’s ROI came from back-office uses like fraud detection, not customer-facing apps
- Governance gaps. Finally, there is the human factor. Many AI projects begin with fuzzy goals and no clear ownership. Without strong governance and clear accountability even successful pilots can’t scale. In addition, shadow AI spreads organically through organisations – Employees adopt tools like personal AI assistants outside of IT’s purview, creating inconsistent approaches and security risks. One study showed a third of employees use tools like ChatGPT at work without approval, which exemplifies how uncontrolled AI usage can introduce risk.
Taken together, these issues explain why so many AI pilots fail to create value. It is not the algorithms that are immature, it is the enterprises themselves, still struggling to manage their data, their systems, and their change. While facing controversies, this GenAI report remains interesting, because the real value of the exercise is not in being shocked by today’s failures. It is in asking ourselves: how do we fix these issues, and what would success look like? To answer that, let’s project ourselves forward into 2035.
2035 – Turning Weaknesses into Strengths
If today’s failures reflect yesterday’s weaknesses, tomorrow’s successes will belong to the companies that learn, adapt, and build a new foundation for AI. The very problems that cripple pilots in 2025 – data debt, legacy IT, misplaced priorities, weak governance – will, by 2035, be the areas where leading organisations have transformed themselves.
From data debt to data capital.
In 2025, most companies are drowning in data debt: fragmented databases, poor lineage, inconsistent definitions. Today’s fragmented, low-quality data becomes tomorrow’s biggest asset. By 2035, leading firms will have turned debt into capital. Every dataset used in AI will come with its own “passport”: a digital certificate showing its origin, the transformations it underwent, current owner, and quality score.
Data ownership stewardship will be fully industrialised, supported by automation. Instead of analysts cleaning data manually, AI agents will continuously monitor and repair data quality issues in real time. This won’t be a luxury, it will be the minimum requirement for regulatory compliance and for business credibility. Regulators (and customers) in 2035 will demand this rigor. For instance, the EU’s AI Act already requires that training data be traceable, of high quality, and free from errors – by 2035 such standards will be globally expected.
Golden records will evolve into an advanced “single version of truth” system at the enterprise level: a continuously governed, AI-maintained fabric that connects every domain together. Golden records won’t disappear, they will be federated into a broader, unified data foundation. From this foundation, every AI model will draw with confidence, knowing it is using the same trusted data as every other system in the company. Data Governance will not be seen any more as a gatekeeper slowing down innovation, it will be a generator of AI-ready assets, ensuring that every algorithm is trained on trusted, reliable inputs.
From legacy IT to AI-native architectures.
The second barrier to AI today is legacy IT. Companies rely on siloed systems and brittle integrations that can barely handle batch reporting, let alone real-time AI. By 2035 companies will run on AI-native infrastructures built for real-time data flow.
Instead of fragmented exports and daily CSV pipelines, enterprises will operate on logical data fabrics – unified layers where governance, security, and access are embedded by design.
For business users, this will feel like magic: data flowing seamlessly across departments, geographies, and why not across corporate boundaries. Imagine a scenario where a finance analyst can instantly pull data from operations, sales, and external sources in a single dashboard, without worrying about format or permissions – that’s the level of fluidity we’re looking at. But beneath the surface, deep investment in interoperable standards, metadata management, and governance frameworks will make such fluidity possible.
From shiny pilots to systemic value.
In 2025, companies often launch AI pilots in marketing or sales, chasing headlines or quick wins. The result is proof-of-concept theatre: slick demos that fail to scale.
By 2035, the success stories will look quite different. The use cases that dominate will not be chatbots or gimmicks but autonomous processes at the heart of the enterprise. Finance closes in minutes instead of weeks. Risk models recalculate exposures in real time. Supply chains adjust dynamically to geopolitical shocks, energy prices, or even the weather.
Success will no longer be measured in novelty but in resilience, efficiency, and trust. The real heroes of AI adoption won’t be the flashiest pilots but the invisible process transformations that keep companies competitive in an uncertain world.
From governance gaps to trusted AI ecosystems.
Finally, data governance. Today, too many pilots suffer from vague ROI objectives, lack of accountability, and shadow operations spreading outside enterprise controls. By 2035, trust will be the differentiator. Companies that can confidently answer ‘Why did the AI make that decision?’ will win customers and regulators over.
Regulators will demand explainability and auditability for every AI decision that affects customers, employees, or markets. Companies will need to prove not just what their AI decided, but why and on which data. Without strong governance frameworks, they won’t even be allowed to operate.
The companies that embed governance today will thrive tomorrow. Trust will be the true competitive advantage: trusted data, trusted AI, trusted companies. Customers, investors, and regulators will all converge on the same question: Can we trust this organisation’s AI? Only those with embedded governance will be able to answer yes.
Conclusion
Seen through this lens, the failures of 2025 are not wasted – they’re tuition for the lessons that will define the next decade, showing us where we must invest to unlock the next decade of AI. They taught us that investing in data quality, modern architectures, strategic use-cases, and strong governance is the only way to unlock AI’s promise. By 2035, the story will be very different: AI won’t just be running pilots, it will be running companies. And the ones that succeed won’t be the ones with the flashiest algorithms, but those that invested early in data governance, data management, and data architecture.
After all, business has always been a story of infrastructure. Once upon a time, it was paper and filing cabinets. Then it was digital platforms and global connectivity. Now it is AI, but AI cannot stand without the scaffolding of governance.
The lesson is clear: the future will not wait. If we want the companies of 2035 to thrive, the work must start today – turning data debt into data capital, legacy systems into AI-native architectures, and fragmented governance into trusted AI ecosystems. Only then will AI deliver on its promise, not as a series of flashy pilots, but as the operating system of tomorrow’s enterprise.
What we think
The story of AI is not one of disruption, but of reconstruction. It’s about rebuilding our organisations on trusted data foundations. The companies that invest today in data governance and architecture won’t just adapt to AI, they’ll define how business works in the next decade.
AI projects fail not because of algorithms but because of weak governance. Strong data management and clear accountability are the keys to scale.
