The AI paradox in Dutch organizations: more data doesn’t always mean better decisions

Quick summary

Collecting more data does not guarantee better decisions. The real issue is not the volume of data, but the link between the data, the business question, and the organization’s ability to act on the answer. Companies that deploy AI without making that connection often see projects stall or quietly disappear.

• According to Statistics Netherlands (CBS) figures from 2025, around 22.7 percent of Dutch companies with 10 or more employees were using AI technology in 2024, a jump of nearly 9 percentage points in a single year.
• Even so, in 2023 approximately 43.6 percent of previously active AI users had stopped using AI the following year, pointing to structural issues with long-term adoption.
• The most frequently cited barrier: lack of experience (reported by around 74.6 percent of companies that considered AI but did not implement it).
• AI only creates value when Data Engineering, Data Science, and Business Intelligence work together around a specific business question.
• Twentynext’s approach, built in part on CRISP-DM and supported by ISO-certified management processes, shows that structure and business focus are what make the difference.

Why is data use growing while decision quality isn’t?

A BI manager at a mid-sized business services firm will know the feeling. There are dashboards, reports, and data feeds coming in from CRM, ERP, and external sources. Yet when it’s time to make a concrete decision about capacity, customer prioritization, or product development, senior leaders still fall back on an Excel sheet someone pulled together the night before.

That is the AI paradox: organizations have access to more data than ever, but the quality of their decisions does not automatically improve. The reason is not the technology itself. The problem is that data, AI models, and decision-making processes often exist in separate worlds.

According to CBS data (2025), the share of Dutch companies using AI rose from around 14 percent in 2023 to roughly 22.7 percent in 2024. But that same CBS AI monitor 2024 also shows that more than a quarter of those companies had already stopped using AI the following year. In 2023, the dropout rate was about 43.6 percent.

That figure matters. It suggests that for many organizations, AI adoption follows a hype cycle rather than becoming part of a sustainable transformation. And the root cause is rarely the technology.

Twentynext, a data and AI agency based in Eindhoven and active across the Netherlands, sees this pattern repeatedly among clients who start AI initiatives without a clear business case. The technology is there. The ambition is there. But the connection between data and decision-making is missing.

What the CBS figures really say about AI adoption

AI adoption means embedding AI technology into business processes in a lasting way so decisions become faster, more reliable, and more scalable. That is very different from running a proof of concept or experimenting with a single AI tool.

The illusion of broad adoption

The rise in AI usage to around 22.7 percent sounds impressive. But the CBS AI monitor 2024 adds important context: companies using AI account for roughly 51 percent of total Dutch business turnover. In other words, it is mainly larger, better-funded organizations that are pushing the adoption figures upward. Small and mid-sized businesses continue to lag behind.

Another important detail: the most common AI use cases in 2024 were marketing and sales (around 36 percent of AI users) and administrative processes (around 30 percent). These are use cases that sit close to existing workflows. For most organizations, using AI for strategic decision-making, predictive models, or operational optimization remains a bridge too far.

Lack of experience is the biggest roadblock

The CBS data on AI adoption considerations is clear: among companies that considered AI but did not move forward, lack of experience was by far the most common reason, at around 74.6 percent. Larger companies also pointed to privacy risks as a barrier.

These are not technical obstacles. They are organizational and human ones. Buying more data or investing in a new AI platform does not solve them.

Why the dropout rate remains so high

The high dropout rate suggests that many organizations treat AI as an experiment without building any long-term foundation underneath it. There is no governance, no data ownership, and no management framework. Once the project lead leaves or the budget is reviewed, the AI initiative disappears with it. Twentynext addresses this through ISO-certified service and management processes, giving clients a framework that continues beyond individual people or project phases.

Try this in your own organization:

• Map out how many active AI applications in your organization have a named owner who is also responsible for continuity.
• Check whether there is a management plan for the data infrastructure behind those applications.
• Ask: are our AI insights connected to a real decision-making process, or are they separate from day-to-day operations?
• If the answer is “no” to more than half of these questions, your AI usage is probably more fragile than it looks.

Data is everywhere, but it doesn’t come together: the Data Engineering problem

Data Engineering is the discipline focused on building reliable data pipelines that turn raw data from multiple sources into clean, integrated, and accessible information for analytics and AI.

The fragmented data landscape

Picture an operations manager at a mid-sized manufacturing company. Production data lives in the ERP system, quality data sits in a separate application, logistics information is held by the carrier, and customer information is stored in the CRM. No one has ever brought all those streams together. Reports are compiled manually each week. Decisions are made using data that is often several days old.

This is not an edge case. It is the norm in organizations that treat data infrastructure as a side effect of buying software rather than as a strategic choice.

What Data Engineering solves

A well-designed data and reporting environment combines sources, aligns definitions (what exactly counts as an “active customer”?), and makes data available in a format that both Business Intelligence tools and AI models can use. Without that foundation, every AI model is built on sand.

The EU AI Act, which came into force on 1 August 2024 and will apply in full from 2 August 2026, makes this more than a best practice for organizations using high-risk AI systems. Article 10 sets strict data quality requirements: training, validation, and test data must be relevant, sufficiently representative, and as error-free as possible. Data lineage must also be demonstrably maintained.

The approach Twentynext uses

Twentynext builds data and reporting environments in which the connections between sources, the data model, and the reporting layer are explicit and documented. That is not just technically sound. It is also a direct step toward compliance with the EU AI Act for organizations planning to use AI in critical processes.

Try this in your own organization:

• List the five most-used data sources behind your core reports.
• For each source, identify: who owns it, how current the data is, and whether it is updated automatically or manually.
• Check whether definitions are consistent: does “revenue” mean the same thing in your CRM as it does in your ERP?
• If more than two sources are being merged manually for a weekly report, Data Engineering is not a nice-to-have. It is a necessity.

When does more data actually lead to better decisions?

The table makes the contrast clear. Initiatives that start with technology or a reporting request usually produce less durable results than those that begin with a concrete business question.

The role of CRISP-DM in structured AI projects

CRISP-DM (Cross-Industry Standard Process for Data Mining) is an iterative six-phase framework for Data Science projects: Business Understanding, Data Understanding, Data Preparation, Modelling, Evaluation, and Deployment. One of its strengths is that the process is not strictly linear. Going back to an earlier phase is not a sign of failure, but a built-in quality mechanism.

Twentynext uses CRISP-DM as a standard approach in Data Science projects. Its real strength lies in the first phase: Business Understanding. Only once it is clear which decision-making process needs to improve, what success looks like, and which stakeholders will use the outcome, does the technical work begin. That may sound obvious, but in practice many organizations skip this phase because of time pressure or excitement about the technology.

What failing AI projects tend to miss

Based on experience across a wide range of client organizations, Twentynext sees three recurring reasons AI projects fail: no clearly defined business question at the start, data that is not ready for modelling because the Data Engineering groundwork is missing, and no plan for adoption and management after delivery. The result is a model that works technically but never gets used operationally.

Try this in your own organization:

• Write the business question your AI project needs to answer in one sentence, without technical jargon.
• Identify who will use the model output and what that looks like in daily practice.
• Check that the required data is available and clean before modelling starts.
• Decide who is responsible for the model after delivery: who monitors performance, and who approves updates?

Checklist: best practices for Data Science, AI, and Business Intelligence

Best-practice checklist for Data Science, AI consulting, and Business Intelligence:

• Define the business question before choosing the technology: A clear problem statement prevents an AI model from being technically impressive but organizationally useless.
• Invest in Data Engineering as the foundation: Reliable data pipelines are a prerequisite for both BI reporting and AI models. Without them, insights are inherently unreliable.
• Use CRISP-DM or a similar iterative framework: A structured project approach increases the odds of reusable, transferable results. Twentynext applies this systematically.
• Put data governance and ownership in place: Assign responsibility for data quality to a designated owner so reports and models do not depend on one individual.
• Plan adoption and training alongside technical delivery: Technology without user adoption creates no decision value. Invest in training and awareness at team level.
• Use ISO-certified or equivalent management processes: Especially in business-critical data environments, continuity and quality assurance should be documented contractually and procedurally.
• Monitor AI models actively after go-live: Models degrade as data patterns shift. Set a review cadence in place (typically every quarter or after a significant process change).
• Prepare early for the EU AI Act: High-risk AI systems require demonstrable data lineage, technical documentation, and automatic logging. Start now, not just before the 2 August 2026 deadline.

What organizations should avoid when implementing AI

Starting with technology instead of the problem

The most common mistake in AI projects is getting the order backwards: choosing a platform first, then figuring out what question it is supposed to answer. That leads to expensive implementations that never move beyond the pilot stage. Twentynext starts with the client’s business challenge, not the technology.

Relying on standalone Excel files for decision-making

Decision-making based on manually assembled Excel files has a fundamental flaw: it is not scalable, not auditable, and not repeatable. The moment the person who creates the file is off sick or leaves the company, the information flow breaks down. An integrated data and reporting environment solves this by providing one consistent source of truth for everyone involved.

Treating adoption as an afterthought

An AI model that is not used by the people it was built for has no value. Yet in many projects, adoption is bolted on at the end as a series of demonstrations. A more effective approach is to build adoption into the very first CRISP-DM phase: involve users in defining the problem, include them in testing, and let them help define what a usable outcome looks like.

Try this in your own organization:

• Check whether the three most recent BI reports in your organization are used to support a measurable decision or are simply informative.
• Ask the five most involved users of an AI application how they interpret the result: inconsistent interpretations usually point to an adoption issue.
• Review the management plan for your most critical data environment: if there is no deliberate review cycle, then management is effectively reactive.

Frequently asked questions

Why doesn’t more data automatically lead to better decisions?

Data without context does not create decision value. The real issue is that in many organizations, data, AI models, and decision-making processes operate independently: data is collected, models are built, but the link to the actual decision point is missing. Only when Data Engineering, Data Science, and Business Intelligence work together around a clearly defined business question does more data turn into better decisions. Organizations that achieve this integration typically see measurable improvements in decision speed and reliability within six to eighteen months.

How does Twentynext help organizations use AI sustainably?

Twentynext brings together Data Engineering, Data Science, Business Intelligence, and AI in one integrated service offering, supporting clients across the Netherlands from its Eindhoven base. The approach always starts with the business challenge, not the technology choice. By using the CRISP-DM methodology and ISO-certified service and management processes, Twentynext ensures that AI applications do not end after the pilot phase but become a structural part of business operations. You can read more about this integrated approach at Twentynext’s services.

What is CRISP-DM and why is it relevant for AI projects?

CRISP-DM stands for Cross-Industry Standard Process for Data Mining and is a six-phase iterative framework that has shaped Data Science projects for decades. The phases run from Business Understanding through Data Preparation and Modelling to Deployment, with built-in feedback loops between stages. For AI projects, the first phase, Business Understanding, is the most critical: if the problem statement is vague here, the outcome may be technically sound but organizationally irrelevant. Twentynext uses CRISP-DM systematically to ensure transparency and repeatability in projects.

What does the EU AI Act mean for Dutch organizations?

The EU AI Act came into force on 1 August 2024 and will fully apply from 2 August 2026. For high-risk AI systems, it requires strict data quality standards, demonstrable data lineage, and the retention of technical documentation for at least ten years. Organizations that start putting structured Data Engineering and governance in place now are also laying the groundwork for compliance. Waiting until 2026 increases the risk of rushed, expensive, and error-prone changes.

How do you choose the right partner for a data and AI project?

The right partner combines strategic advice, technical delivery, and long-term management, so your organization does not have to rely on multiple niche providers for different parts of the chain. Practical criteria include: does the partner start with the business question or the technology, do they have experience in both Data Engineering and AI modelling, and is there a certified management structure in place after delivery? For a first look at what that kind of approach involves, see Twentynext’s consulting services.

Conclusion

The paradox of data overload and weak decision-making can be solved, but not by adding more technology alone. The answer lies in connection: between the business question and the data model, between Data Engineering and AI, and between the model and the people expected to use it in practice.

CBS figures show that the AI dropout rate in the Netherlands remains structurally high. That is not a reason for pessimism, but a signal that the approach needs to change. Organizations that invest in a solid data foundation, a structured project process, and a manageable long-term operating model are the ones that create sustainable value from data and AI.

For organizations in the Netherlands ready to take that step, Twentynext offers the combination of expertise, methodology, and guidance needed to make it happen. Not as a technology vendor, but as a partner that starts with the question that actually matters. See how Twentynext helps organizations grow through data-driven decision-making for a first introduction.