Kronehoefstraat 85
5612 HL Eindhoven
BTW NL863902509B01
Three Dutch SMEs with One AI Operating System: What It Looks Like in Practice
qw
vc
mshm
.svg)
The three companies looked incomparable at first glance. Different sector, different stack, different teams. Yet all three conversations started the same way: the founder or director was the human connector between four or five systems. Data was scattered, nobody had a clear picture, and the question "how are we doing?" always led to searching rather than an answer.
They also shared one preference: they did not want to replace their existing software. Moneybird, AFAS, Bloxs, SharePoint, their ERP. All in use, all needed. What they wanted was a layer on top, not a new package underneath.
Below, per case: what we built, which layer went live in which phase, and what it delivered.
A Dutch property manager in our portfolio had three critical systems running side by side. Moneybird for accounting and invoicing, AFAS for operations and HR, and Bloxs for the property platform and tenancy agreements. For a single property query, a manager had to work in three applications at once.
The result: simple questions took fifteen minutes instead of one. And with every property question from an investor or tenant, the manager felt the cognitive load of switching between interfaces.
Layer 1 was already there, in the form of Moneybird, AFAS, and Bloxs as knowledge sources. We focused on layer 2: a searchable AI layer on top of those three systems that combines data instead of presenting it separately.
Concrete steps over six weeks:
The manager now types one question instead of opening three interfaces. "Which tenancy agreements for entity X expire this quarter and is anything outstanding in Moneybird?" returns an answer in seconds with the figures and source references attached.
Management has indicated that the time saved per manager is visible in how many property queries they can handle per day, and that the threshold for looking up data has dropped low enough that more data-driven decisions are being made than before.
Primary layer: layer 1 plus layer 2. Layer 3 (chat agents and automations for specific tasks) is the logical next phase.
HOS Safety places safety professionals with large clients. The combination of scheduling, timesheet registration, and invoicing is notoriously complex in staffing: each placement has its own rates, its own surcharges, its own client-specific invoice specifications. At HOS Safety, the information was spread across the scheduling tool and the accounting system, with invoice checks as the weekly bottleneck.
One staff member was consistently spending around 20 hours per week on invoice checks. Does the rate match, do the hours add up, is the billing per client correct, are all surcharges included. That is half a working week of manual side-by-side comparison, for a process that should essentially be a check routine.
HOS Safety was a textbook case where layers 1 and 2 together deliver immediate value the moment you add one control layer (layer 3) on top.
Instead of the staff member going through every invoice line by line, she now receives a list of lines showing deviations, with the source lines from scheduling and timesheets included.
Invoice checks have been reduced from around 20 hours per week to approximately 2 hours per week. A time saving of over 90 percent on one specific process, and capacity that HOS Safety now uses to handle more placements without adding headcount.
HOS Safety's management notes that the effect goes beyond time savings alone. The error rate in invoices falls because discrepancies are flagged consistently, rather than depending on who was doing the checking that week.
Primary layers: layer 1, layer 2, and a targeted module in layer 3. The invoice check is a first AI agent that takes over structural work, built on the searchable data layer below it.
A Dutch telecommunications company had a typical customer service problem. For most customer queries, a service agent needed information from AFAS (customer administration, contracts, invoicing), from project pages on the website (active installations, connection status), and from SharePoint (technical documentation, procedures, escalation paths).
The result: one customer query meant three tabs open, switching between interfaces, and still often forwarding to a second-line colleague. Front-line agents grew frustrated, second-line colleagues became overloaded with escalations that should in principle have been handled at the front line, and customers waited longer than necessary for an answer.
This is the most complete of the three cases: all three layers are live, working together.
An agent receives a customer query, types the question or context into the chat agent, and the agent searches across the three systems for the answer. With source references included, so the agent can verify the information before sharing it with the customer.
Three effects work together.
The side effect: the pressure on senior second-line colleagues decreases. The organisation now uses that capacity to work on structural improvements rather than putting out fires.
Primary layers: all three, working together. This is a complete AIOS implementation, not a first phase.
The pattern is in the pain, not in the software. Property management, staffing, and telecom have completely different stacks. But the underlying complaint is the same everywhere: data is scattered, the founder or senior staff member is the human glue, and it takes too much time to answer a simple question.
Three layers, phased, not in one big bang. At the property manager, layers 1 and 2 are live, layer 3 follows. At HOS Safety, one targeted module in layer 3 is in place, built on the foundation below. At the telecommunications company, the whole system runs. None of these clients switched everything on at once. Everyone started with a working foundation in six weeks, then expanded in steps where the business case was strongest.
Existing software stays. In none of the three cases was an existing system replaced. Moneybird, AFAS, Bloxs, SharePoint, and the website do what they always did. The AI Operating System is the layer on top that makes them work together.
The gain is in time and in decisions. Time savings are measurable (HOS Safety: 18 hours per week on one process). But the second gain is subtler and equally important: because information is within reach, more decisions are made on the basis of data rather than intuition. That is as valuable for a property manager as it is for a telecoms customer service team.
All three cases above started with the same conversation: an AI audit in which we mapped the current stack, identified the biggest time drains, and made concrete what an AI Operating System would deliver for them.
Not a sales conversation. A first look at your systems, your processes, and the three to five connections that will deliver the most time savings in phase 1.
Start with an AI audit. 30 minutes, online or at our office. Within three working days you receive a written summary with the recommended connections, use cases, and indicative investment.
Book an AI audit or read the AI Operating System pillar for Dutch SMEs first.
The three cases cover property management, staffing, and telecom. We also work primarily with professional services firms (accounting practices, consultants, legal, marketing) and construction companies. The architecture is sector-independent. What differs are the connections and the specific questions the system needs to answer. During the AI audit we assess whether your situation is close enough to cases we have handled before, or whether we need to start cautiously with a smaller initial phase.
At the property manager, a working layer 1 plus layer 2 was in place within six weeks. That is the target we work towards for a first phase: 4 to 6 weeks to lay the foundation with the first 3 to 5 systems connected. After that we expand with layer 3 in steps of 2 to 4 weeks per addition. No big-bang implementation, even with larger stacks.
The three cases above use six different systems between them. We also integrate Exact Online, HubSpot, Pipedrive, Salesforce, Google Workspace, Microsoft 365, and project tools such as ClickUp or Asana as standard. For sector-specific software we assess on a case-by-case basis whether an authorised integration path exists. During the AI audit we map your exact stack.
Not every business case starts with a clear number. At the property manager, the gain was initially qualitative: less switching, faster answers, more data-driven decisions. We advise starting with the process that involves the most switching or searching, not the process with the most measurable KPI. Measurability follows naturally.
About the author. Vincent de Vos is founder of The Agentic Group. He has been involved in AI and integration projects for Dutch SMEs since 2018 and is hands-on with every implementation TAG delivers. Previous clients in property management, professional services, construction, staffing, and telecoms.
.svg){kind=icon}