By: Michael Privat
There is probably a company right now studying your AI product, estimating how long it would take to recreate the core experience, and realizing the timeline may be shorter than some teams would like to admit. Not because your team failed or because their engineers are unusually strong. Building AI-powered features can be cheaper, faster, and increasingly accessible to companies that know how to execute.
I have spent more than 20 years inside large-scale engineering organizations, most recently leading roughly 500 engineers across the US and India, and what I keep watching is a similar expensive mistake playing out across the industry. Companies are pouring money into features and calling that differentiation. Meanwhile, their competitors may be watching closely, rebuilding similar capabilities faster than expected, and entering the market with almost identical experiences six months later.
What competitors may not be able to replicate nearly as quickly is governed data, operational trust, and infrastructure deeply embedded into how businesses actually operate. If your competitor can replicate a valuable capability within half a year, then it may not have been the thing protecting you. It may have simply been your temporary lead in a race that keeps accelerating.
Many executives already understand this in theory. The problem is what happens once quarterly planning starts.
I have watched this happen at companies with real resources and smart people, and the pattern is often the same: velocity gets celebrated, governance gets underfunded, and then 18 months later, the AI initiative that was supposed to be significant is stuck in a cycle of inconsistent outputs and customer trust problems that teams may struggle to explain. The explanation is usually not the model. It is often the data underneath it.
Many Companies Built Features. Now What?
Here is what the numbers look like right now. RAND Corporation reported that over 80% of AI projects fail to reach meaningful production deployment, roughly twice the failure rate of traditional IT projects without AI components (Source: RAND Corporation, 2024). NTT DATA reported that between 70% and 85% of generative AI deployments fail to meet their target ROI, and cited governance gaps in the underlying data as a leading cause (Source: NTT DATA, 2024). And Gartner is predicting that 80% of data and analytics governance initiatives could fail by 2027, specifically because organizations treat governance as a reactive obligation instead of a structural foundation (Source: Gartner, 2024).
That last one deserves a slower read. The issue may not be that companies are ignoring AI. They are adopting it aggressively. Generative AI adoption was reported to have more than doubled between 2023 and 2024, jumping from 33% to 71% of organizations (Source: WalkMe, 2025). The issue is that adoption has lapped governance by a wide margin, and it is difficult to build a reliable AI system on a data foundation that nobody actually trusts. Many enterprise data environments, if you strip away the optimistic language in the board deck, are a complex mix of siloed systems, undocumented pipelines, and inconsistent definitions that different teams interpret differently.
The AI may not be making that problem worse. It may be making it harder to hide.
What I see in the teams that are getting to production and staying there is often not a better model. It is often better data. The companies compounding AI value right now appear to have treated data infrastructure as the actual product, not the thing that supports the product. That sounds like a small reframe. It is not. It can change every prioritization decision you make for the next three years.
Why Governed Data Is Becoming a Potential Moat
Trusted data, deep integrations, and regulatory credibility can be harder to replicate than AI features themselves. None of them may make exciting product launches. Teams may not applaud clean data lineage or governance documentation during an all-hands meeting. But six months later, when enterprise customers trust your outputs enough to build workflows around them, the unglamorous work can become an advantage.
Trusted Data Can Compound
Many companies still treat data cleanup like housekeeping. Something annoying that gets postponed until the mess becomes difficult to ignore.
The organizations getting AI value appear to have accepted something many companies still resist: governed data is part of the product.
When a company cleans a dataset, assigns ownership, documents where information came from, and validates what it actually means, it can become easier to scale, validate, and trust future AI deployments. Right now, 73% of enterprises cite data quality as a major AI challenge (Source: Second Talent, 2025). The companies that address that problem early may be able to operate more efficiently.
Deep Integrations Can Create Weight
There is a difference between an API connection and an operational dependency.
The integrations that can become potential moats are the ones woven deeply into how customers actually work. Removing those systems could force teams to restructure workflows, compliance processes, reporting structures, and operational habits, not simply replace a vendor.
AI capabilities layered on top of shallow integrations can disappear quickly. The interface changes, another model may catch up, and the differentiation can fade. But infrastructure embedded deeply into customer operations can gain weight over time instead of losing it.
Regulatory Credibility Is Becoming a Gatekeeper
Many companies avoid governance conversations until regulation forces them to care.
As AI oversight expands across the EU, the US, and APAC markets, enterprises may increasingly be asked to prove their outputs are auditable, their data handling is defensible, and their governance structures can survive scrutiny from procurement teams and regulators alike.
Companies that invested in governance early may be better prepared for that environment. Companies that prioritized feature velocity first may now be retrofitting compliance into systems that were never designed for it. And increasingly, enterprises unable to demonstrate AI transparency may be excluded from regulated-market opportunities early in procurement conversations (Source: Gartner, 2024).
What Twenty Years Inside Engineering Organizations Taught Me
I started writing code on computers sitting on store shelves in France in 1983, as an eight-year-old who could not afford to own one. That eventually took me through coding parties across Europe, advanced math preparation in the late nineties, and MIT’s lab in 1998. I have spent my career solving real operational and engineering problems ever since, and the pattern across effective teams has stayed remarkably consistent: the organizations that sustain results over long cycles are usually the ones investing in durable infrastructure while others chase whatever benchmark looks impressive that quarter.
The data problems behind failing AI initiatives are often not mysterious. When I walk into organizations struggling with AI implementation, the engineers usually already know where the problems are. The data scientists are compensating for them regularly. Leadership tends to notice only after a production model generates something embarrassing or a pilot collapses once real customer data enters the system.
And often, the root cause sits upstream from the model itself:
- Data gaps nobody resolved
- Conflicting definitions across teams
- Unclear ownership of critical datasets
- Inconsistent pipelines patched together over time
- Governance is treated like operational overhead instead of infrastructure
Those problems can stay hidden surprisingly long inside traditional systems. AI can expose them immediately.
There is also an organizational layer to this conversation that deserves more attention. In companies extracting meaningful AI value right now, data engineering is often not sitting downstream from machine learning teams cleaning up afterward. It operates as a strategic peer because leadership recognized something important early: trustworthy data can be foundational to the entire system.
You can purchase better tooling. You can buy another governance platform. What is harder to buy is the organizational habit of treating data like a first-class asset. That usually has to be built intentionally, and leadership has to reinforce it consistently.
The Next Three Years May Show Which Companies Built Strong Foundations
If you are navigating this inside a large organization, trying to make the case for data governance investment when AI feature velocity is consuming all the oxygen in the room, or trying to figure out why your AI initiative is stalled despite real investment in the technology itself, I would be glad to connect. These are not abstract problems for me. They are the problems I work on.
You can find me on LinkedIn or follow the longer-form conversation on Substack, where I write about engineering leadership, organizational transformation, and what it takes to build technology organizations that can compound their advantages over time instead of just chasing the next launch.
Many organizations are still treating the model like the product, while the potential long-term advantage may be accumulating quietly underneath it. Governed data may become the layer that separates companies creating short-term AI momentum from companies building infrastructure that customers and regulators may continue to trust years from now. The organizations that understand that shift early may look very different in the next three years, and the distance between them and everyone else could become increasingly difficult to close.
