Beyond the Algorithm: Overcoming the Six Hidden Obstacles to Data and AI Success

SECTION 1

Why so many data and AI initiatives still fail in 2026

Data and AI still dominate roadmaps, but the conversation has changed. After the wave of generative AI experimentation, then the arrival of agents able to execute entire processes, executive teams now demand measurable results. Yet the majority of pilots never make it to production, and industry analyses converge: it is almost never the models that fall short, but everything around them, strategy, data, organization, and adoption. Vendors promise great results; beneath the surface, many organizations still struggle to translate investment into tangible value. Why do so many of these initiatives fail, and what can be done to make them succeed? In our view, the answer lies far less in the technology itself than in the way organizations prepare for it.

The paradox is striking. Tools have never been so powerful or so accessible: frontier models available via API, AI agents orchestrating multi-step workflows, unified data platforms bringing analytics and operations closer together. And yet the ability to turn these technological building blocks into lasting business value remains the exception rather than the rule. Our conviction: before choosing a tool, you must understand what it is meant to accomplish and why. That is what directs funding toward real potential and builds solid foundations for adoption.

The calendar adds an unprecedented constraint: since August 2, 2026, the obligations of the European AI regulation (the AI Act) have applied to high-risk systems, risk management, data governance, technical documentation, human oversight, and CE marking, with penalties of up to 3% of global revenue. What used to be good practice has become a regulatory requirement. From our field experience, six recurring obstacles derail the most promising efforts. Here they are, along with practical ways to overcome them.

SECTION 2

Strategic vacuum: tying every AI initiative to business objectives

Launching AI initiatives without a clear connection to business objectives is like putting to sea without a destination. The motivation is often reactive, "doing AI" because competitors are doing it, or because an executive saw a convincing demo. The result is predictable: scattered experimentation, a return on investment that is impossible to measure, and technically successful projects that fail to move the needle on any core business priority. Without strategic direction, teams lack focus, resources get diluted, and leaders remain uncertain about the value actually delivered.

### Prioritizing use cases by value and feasibility

The answer is to develop an AI strategy explicitly tailored to the organization's context and ambitions, jointly sponsored by technology and business leadership. Map potential use cases along the value chain, then prioritize them along two axes: expected business value and technical and organizational feasibility. This discipline directs funding toward the initiatives with the greatest strategic impact, and screens out showcase projects that consume resources without ever delivering anything. In other words, do not pursue AI for its own sake: identify where it can genuinely differentiate your business, and let that answer drive the roadmap.

A further sign of maturity is distinguishing assistive use cases, copilots, content generation, from agentic use cases that automate processes end to end. The latter promise greater value, but demand significantly stronger data, security, and governance foundations.

SECTION 3

Flawed data foundations: quality before scale

AI systems fundamentally depend on data. Poor quality, unrepresentative datasets, unclear lineage, inconsistent management practices: these flaws doom models from the outset and produce inaccurate results, biased outputs, and ultimately a loss of user trust. With RAG and agents, the problem has actually widened: it is no longer just structured data that matters, but also the documents, knowledge bases, and business repositories that AI consults in real time to produce its answers. Garbage in, garbage out has never been more literal than in the age of generative AI.

### Assessing whether your data is AI-ready

Before scaling up your ambitions, audit your data landscape. Is the data fit for the intended purpose? Does it faithfully represent the populations you are modeling? Are its quality and traceability verifiable? Does your platform expose data as documented products, with interface contracts and freshness indicators that AI pipelines can consume with confidence? Do you have the skills to maintain them? Answering these questions honestly is often sobering, but it is far cheaper than discovering the gaps in production.

If the answer to any of these questions is no, prioritize building these foundational capabilities before industrializing anything: quality improvement, clear asset ownership, robust governance, and alignment of practices with security, privacy, and ethics standards.

@cite:principes-d-architecture-de-donnees

SECTION 4

Portfolio blind spots: steering the data and AI investment

In large organizations, AI initiatives emerge organically in every department, often without central coordination or visibility. This sprawl produces duplicated efforts, competing projects, wasted resources, and a leadership team unable to arbitrate the allocation of capital and talent. The phenomenon has worsened with generative AI: any team can now build a convincing prototype in a matter of days, multiplying invisible workstreams and uncontrolled inference costs. Without a clear view of the entire data and AI portfolio, no one can effectively manage the investment or track the value actually realized.

### Establishing a regular portfolio review

Establish a clear, prioritized, and actively managed portfolio of all significant data and AI initiatives, capturing for each one the strategic objectives, expected outcomes, required investment, and dependencies. Then institute a review cadence in which leadership assesses the portfolio's alignment with overall strategy, tracks real value creation, and unsentimentally stops projects that fail to demonstrate it. A well-managed portfolio brings discipline and transparency, and ensures resources flow toward the work that delivers the most value.

SECTION 5

Organizational friction and "Wild West" data practices

Traditional organizational structures create major obstacles when no one clearly owns the data. The symptoms are well known: data copied and modified on the fly by different teams for their local needs, which breaks lineage and trust; critical considerations like bias or inclusiveness neglected in models; inconsistent, or entirely absent, security, privacy, and ethics guidelines for data collection, retention, and use.

### Placing data ownership within business domains

Design roles and structures that explicitly support data ownership: place it within the business domains that understand the data best, and balance this distributed accountability with central enablement functions providing shared tools, standards, and guardrails. Complete the setup with targeted AI literacy and KPIs that allow experimentation in a "safe to fail" environment, in order to support continuous learning. Teams closest to the data gain autonomy, while the enterprise keeps consistency and responsible practices.

### Lightweight, tooled governance: now required by the AI Act

Against Wild West practices, institute lightweight but disciplined governance: clear, practical policies, automated checks embedded in pipelines, and responsible AI principles guiding development and deployment. The AI Act makes this investment doubly worthwhile: the data governance and documentation requirements for high-risk systems rest on exactly these same foundations. An organization that has laid them turns compliance into a formality rather than an emergency program.

@cite:gouvernance-des-donnees-transformation-digitale

SECTION 6

Adoption challenges: treating AI as change management

Even the best-designed AI solutions fail if no one uses them. Adoption stumbles over lack of engagement, fear of replacement, internal politics, a perceived lack of personal benefit, or inadequate training. With agents automating entire swaths of processes, the question of team trust becomes even more central than with simple copilots: employees need to understand what the system does, within what limits, and how to take back control. Adoption is never a detail to be handled at the end of a project; it is a workstream in its own right, with its own objectives and its own metrics.

### Internal champions, training, and explicit benefits

Treat adoption like any other change management challenge: engage users and stakeholders early and often, clearly communicate the purpose and "what's in it for them," and identify and empower internal champions who lead by example day to day. Invest in training programs that demystify AI and give teams the skills to work confidently with new tools, including the ability to review, question, and correct model outputs rather than accepting them blindly. The goal is simple to state and demanding to achieve: teams that trust the tools because they understand them.

SECTION 7

Building lasting foundations for measurable AI value

Focusing on acquiring the latest technology while neglecting strategy, data foundations, portfolio management, organizational design, and disciplined practices: that is the root cause of the high failure rates of data and AI initiatives. No model, however capable, can durably compensate for these blind spots. Technology is the visible part of the effort; the foundations described here are what actually determine whether it pays off.

The good news is that the six obstacles described here can be addressed with deliberate effort. By shifting focus from purely technical implementation toward these foundations, companies dramatically improve their chances of success: they move beyond the hype and start generating lasting, measurable value from their investments. This is precisely the approach Adservio applies with its clients, diagnosing the foundations, prioritizing the portfolio, then progressively industrializing the use cases whose value has been demonstrated.

Disclaimer: the statements and opinions expressed in this article are those of the author(s) and do not necessarily reflect the positions of Adservio.

@cite:les-sept-peches-capitaux-de-la-transformation-ia-lecons

FAQ

Frequently asked questions