Data & AI trends

Microsoft has published an analysis of roughly 40,000 agents that organisations have built with Copilot Studio. According to the company, the analysis provides a baseline for what enterprise agent adoption looks like today and a way to measure how it evolves. Microsoft is also introducing a new harness intended to give makers more room to tackle complex business problems with agents. Copilot use is thus moving from isolated experiments towards agents that carry out processes within the organisation. For organisations, this means that arrangements about who builds agents, which data they may access and how they are managed matter at least as much as the technology itself.

According to NRC, around two hundred employees of the Dutch immigration service IND are now working with Robin, an AI assistant the Ministry of Justice and Security built last year as a secure alternative to ChatGPT. The deployment was not disclosed. The chatbot was first tried out by a group of innovation-minded staff and then rolled out step by step. A comparable assistant at the UK immigration service saved considerable working time, yet during its pilot roughly one in ten AI summaries contained errors. For organisations, the case shows that introducing an AI assistant for staff involves more than switching on a tool: it requires clear rules on registration, access to information and checking of outputs.

Consultancy.nl reports that Van Dam Datapartners has helped the Province of North Brabant make data a regular part of everyday work. Within the department for Housing, Work and Living Environment, plenty of data was available, yet decisions still frequently relied on experience. The project therefore focused on putting existing data to use in day to day decision making rather than on collecting more of it. The case illustrates that data availability alone changes little as long as ownership, working processes and habits stay the same. For organisations, this means a data platform only delivers value once its use is embedded in how people actually work.

Writing for TechTarget, Bill Schmarzo argues that many enterprises wrongly treat the AI stack as a new frontier alongside the data platform they already run. In his view the model is only the visible face of AI, while the data architecture does the heavy lifting and decides whether results remain consistent and reliable at scale. A well-built architecture does not lose value over time; with ongoing maintenance and reuse it becomes more useful. Only a few components, he says, are genuinely new and justify separate investment. For organisations this means the condition of their data foundation largely determines how far they can take AI.

InfoWorld examines how Microsoft’s data platform, spanning Azure Databases, Fabric, OneLake, Power BI and its governance tooling, links systems of record and operational applications with analytics and AI. The author frames this as a decision architecture with five layers: context, decision logic, authority, orchestration and execution, the last of which includes a route to review or reverse an action. Trusted data is the precondition; without it, automation is hard to trust at scale. The article argues that architecture should follow workload, data maturity, economics and business value rather than a single destination. For organisations, this means the choice of data foundation largely determines how controllable later automation and AI applications will be.

Security governance firm Syskit released its State of Microsoft 365 Governance Report on 10 September. It finds that 76 percent of surveyed organizations in the UK and the US have deployed or piloted an enterprise AI tool such as Copilot on Microsoft 365 data. Controls trail confidence by a wide margin: 91 percent of respondents say they can see which agents are active and what they can reach, yet only 22 percent have a formal policy defining what AI agents may access, and nearly one in ten lets agents inherit existing permissions. For organizations, this shows that a Copilot rollout is only manageable when permissions and data access have been reviewed and documented beforehand.

A new report on human risk, covered by WINMAG Pro, ranks AI as the second largest human risk in the workplace. Employees have adopted generative AI faster than security teams have been able to regulate its use. The report identifies three areas: unauthorised use of generative AI tools, vibe coding by staff without a software development background, and risks that arise when AI agents act on their own. For organisations, this means that rolling out tools such as Copilot is less a question of adoption than one of rules on data access, permitted use and oversight.

Research by Censuswide, commissioned by Veeam, finds that 70 percent of organisations in EMEA are dealing with so-called shadow agents: AI workflows and agents set up outside the view of IT that interact with sensitive data. Where shadow IT used to mean unauthorised apps, the issue now concerns automated processes that access and handle company data on their own. This makes the question of who may see which data more pressing than it was with conventional software. For organisations, the finding means that rolling out AI assistants stays manageable only when access rights, data classification and logging are settled in advance rather than repaired afterwards.

Consultancy.nl has published an analysis by Mobilee on data quality. Its main point is that organisations that succeed in creating value with data and AI do not start by investing in technology, but in trust in their data. That trust develops when employees know what the data means, how reliable it is and who is accountable for its quality. The author recommends that for every important dataset, organisations name the person who decides about the data and the person who checks in daily practice that it is recorded and used correctly. For organisations, this means data quality is chiefly a matter of ownership and working agreements, and that a technical fix on its own will not resolve it.

In practice, data mesh succeeds mainly as a hybrid model, TechTarget reports. Rather than replacing existing data lakes and warehouses, organisations pair a central platform with distributed ownership of data products. One example is a large logistics provider that, working with consultancy Kearney, kept its data platform, catalogue, quality standards, security and AI governance central, while business units for freight, network and retail owned their own data products and roadmaps. This avoided an expensive rebuild of the cloud platform and sped up analytics. For organisations, the lesson is that choosing a data architecture is less about swapping technology and more about how responsibilities are divided on a shared foundation.

The Dutch municipality of Helmond is working with Argaleo and KPN on smart applications for local government, iBestuur reports. Helmond has used smart traffic lights, sensors and digital models for years, but each individual application tended to arrive with its own dashboard, leaving the underlying data hard to use outside the vendor’s system. The question now is how sensors that are already in place can serve several policy areas at once, starting from a secure single entry point for data. The case shows how stand-alone solutions fragment data over time: only an architecture that opens up sources independently of vendors lets organisations reuse the same data for multiple purposes.

Microsoft has revised its segment reporting structure for fiscal year 2027. The new layout mirrors how executive management evaluates the business and allocates capital, centring on integrated AI infrastructure, adoption of the Copilot platform and cloud productivity. For the first quarter, the company expects Azure to grow by roughly 45 percent in constant currency, while the Productivity and Business Processes segment is projected to keep growing in the mid to high teens, driven by enterprise Copilot adoption. For organisations, this signals that Copilot is becoming a fixed part of the Microsoft environment they already run, which brings questions about access rights, data quality and working agreements forward sooner than many had planned.

Xebia and Data Expo have published the Data & AI Monitor 2026/2027 for the Benelux. Organisations are experimenting widely with AI, but its value often remains unproven. In the public sector and education, 31 percent do not measure the value of AI at all, and healthcare also lags behind in measuring results. According to Mariam Halfhide of Xebia, the figures point to a governance problem rather than a technology problem: experiments succeed, but ownership, reliable data, risk and measuring success are where organisations struggle. The takeaway for organisations is that returns on AI depend largely on how ownership, data quality and measurement are embedded in everyday processes.

At the Deutsche Bank Technology Conference, Microsoft reported that engagement with Microsoft 365 Copilot is now on a par with Teams and Outlook, two of its most widely used products. Many customers are moving from limited pilots to licences covering the whole organisation. Pricing combines a fixed fee per user with usage-based billing for agentic workloads, alongside new tools to keep costs in check. Copilot Cowork is aimed at longer, multi-step tasks. For organisations, the move from pilot to full rollout means that data access, permissions and cost control have to be settled in advance, since Copilot surfaces whatever employees are already allowed to see.

In an opinion piece, ITdaily describes how the rise of AI is shifting the focus of compliance. The question is no longer only which employees can access data, but also which autonomous AI systems may use that information, under what conditions, and with what audit trail. Regulations such as NIS2 and the EU AI Act reinforce this shift. The piece argues that sound data governance is not merely an obligation: organisations that structure their data well are also building the foundation for trustworthy AI applications. For organisations, this means the quality of their data foundation increasingly determines how quickly and responsibly they can deploy AI.