"Which AI tool is the best?" We get that question every week. The honest answer is annoying: there isn't one. What does exist is the best tool for your process, your data, your team. And that's not a detail — that's the whole game.
Some numbers first. 67% of Dutch companies now use AI, more than double the figure in 2023 [1]. Sounds great. But only 5% extract measurable value from it [2]. So nine out of ten companies are paying for something that gives them nothing back. That gap is what this article is about. Almost without exception, the cause is tool choice and rollout — not the models themselves.
What you won't find here: a "top 10 AI tools of the quarter" list. Those are outdated before you finish reading them anyway. What you will get: six criteria that actually matter, five categories that are genuinely useful for SMEs, and a sober view of when to buy a standard tool and when custom-built makes sense. By the end you'll be asking better questions. That's worth more than a wrongly chosen subscription.
Why "the best AI tool" is the wrong question
Picture this. A chatbot that works beautifully for a webshop with five hundred standard questions is a disaster for a law firm handling confidential case files. A writing assistant that shines in English stumbles over industry jargon in Dutch. An automation platform that looks cheap at a hundred actions per month destroys your margins at a hundred thousand.
Same piece of kit. Three completely different outcomes.
Dutch SMEs are sitting at a notable point. 84% want to invest more in AI over the next three years [3]. 31% run fully in the cloud, with another 50% on hybrid setups [3]. Good news: the technical barrier to fitting modern AI in is lower here than in most other EU countries. Bad news: that makes the choice more sensitive, not less. Every wrongly chosen tool is a data silo and a licence you'll want to be free of in two years — and untangling those things is never free.
The question to start with isn't "which tool is good." It's "which process currently costs us the most time or money, and what should the outcome look like if it actually worked?" Skip that question and you're buying a solution to a problem that hasn't even been defined yet. Before you go further, also read our guide on how to apply AI in your business. It lays out the framework that should precede any tool decision.
Six things you actually need to look at
Demos and use cases are marketing. The six points below are where the decision is really made:
1. Does the tool fit your specific process? Generic tools are broad but shallow. Sector-specific tools are deep but narrow. It's not unusual to choose wrong on this point and regret it 18 months later.
2. Integrations. Does the tool talk out of the box to your CRM, your accounting package, your ERP, your email? Or does every data flow need a custom connector? A small detail that, in our experience, regularly costs more than the licence itself.
3. Pricing model. Per user. Per conversation. Per action. Per token. Per month. Self-hosted. The wrong model for your volume can multiply your bill by ten the moment you grow. Always ask what happens at 5x your current volume.
4. Data location and GDPR. Where does the data live, who can access it, and will the vendor sign a data processing agreement? A concrete situation: in 2025 the Dutch Data Protection Authority determined that most generative AI models do not meet GDPR requirements and moved from guidance to enforcement [4][5]. On top of that, Cisco reported that 46% of organisations had data leaks via prompts to public AI [6]. Almost always traceable to the wrong tool or no governance. This isn't fear-mongering — it's a bill that's now on the table.
5. Scalability. Does it still work at twice the volume? Or with a second country office? Many cheap tools are a party until you grow and suddenly find yourself paying per action.
6. Learning curve and ownership. How quickly does your team adopt it, and — not unimportant — can you get your data out if you ever want to switch? A tool you can't leave is a gilded cage. Beautiful, expensive, and you're locked in.
Category 1: Generative AI assistants
This is the familiar territory. Chat AI that writes, summarises, brainstorms, translates, generates code. ChatGPT, Claude, Copilot, Gemini — all aimed at the same kind of work. At the basic level the major players deliver roughly the same thing. The difference sits elsewhere.
Where? Integration with your work environment. Data location. Whether your input gets used for training. And whether there's a proper business licence with DPA support behind it. Those are the things that decide on a Thursday morning whether your IT lead sleeps or not.
For Dutch SMEs (and most international SMEs), the core question isn't "which model is smartest." It's: is my team already using it on the side? A Gartner study found that 69% of organisations suspect employees use unauthorised public AI tools — shadow AI [7]. Banning doesn't help. What helps is offering a controlled version quickly. Enterprise licence. Data processing agreement. Clear rules about what's allowed in.
Honestly: this is low-hanging fruit. The OECD studied it and found 39% of SMEs with a skills shortage say generative AI helps close that gap [8]. But you have to roll it out company-wide, not as scattered subscriptions held by seven different people that nobody manages centrally.
Category 2: AI in customer service and chatbots
Want to see results fast? Start here. Companies see an average return of $3.50 for every dollar invested in AI customer service. Leaders hit 8x [9]. A peer-reviewed study followed a Romanian micro-enterprise — fewer than 10 employees — for six months. Response time dropped. Support workload dropped. Both measurable [10]. So it really works, even at the smallest end of the market.
Three flavours in this category:
- SaaS chatbots with pre-trained models. Quick to deploy, great for FAQ volume. But: hard to connect to your internal knowledge. Personalisation hits a wall fast.
- AI layer on top of an existing service desk or CRM platform. Strong if you're already deep in such a platform. The agent uses your real tickets, your real knowledge base. One major CRM vendor reported that their customer agents resolve 25% more tickets, work 15% faster, and handle an average of 65% of conversations autonomously [11]. Impressive — but remember that's their own number on their own platform.
- Custom chatbots on your own data and APIs. Here you choose the language model yourself, host it in an EU cloud, and connect it to your inventory system or case management software. For industries with confidential data or unique workflows, this is the only route that scales without compromise. More expensive upfront. Cheaper later.
A German e-commerce business deployed a multilingual AI chatbot and was running customer service in six new languages within two to four weeks. No local hires required [12]. That kind of outcome doesn't come from a free plug-in. For a deeper look, see our guide on AI customer service.
Category 3: Workflow and process automation
This category has exploded over the past two years. 7 in 10 companies say AI agents are their most important automation tool in 2025, and 2 in 3 already report productivity gains [13]. The major platforms — n8n, Make, Zapier, Power Automate — now offer visual flow builders, AI nodes, and direct LLM integration [14]. All of them.
The question for SMEs isn't "which platform is technically the strongest." It's more like:
| Profile | Usually fits |
|---|---|
| Simple integrations, low volume, no IT team | Standard SaaS platform with visual builder |
| High monthly action counts, complex logic, IT-aware | Self-hosted or hybrid platform |
| Industry-specific processes, sensitive data, legacy integration | Custom automation with expert implementation |
The treacherous part of automation platforms is exactly what I mentioned earlier: the pricing model. What looks cheap at a thousand actions per month can become unaffordable at a hundred thousand. On the other hand, a self-hosted solution with no one maintaining it is a ticking time bomb within a year. For our take on this landscape, see automating your business and reducing manual work and workflow automation.
Category 4: Document and invoice processing
Invoices, contracts, delivery notes, passport copies. Anywhere PDFs and scans flow in, there's now massive value to extract. Modern AI achieves 93–99% accuracy on invoice fields, against around 80% for classic OCR [15]. A typical AP automation pays back within 12 to 18 months and can cut matching time by up to 90% [16].
A pause here, because this is a category where it's easy to buy badly. A beautiful demo on a clean invoice is a very different thing from a hundred different supplier formats in your reality. Always ask for a test on your own documents, not the vendor's.
Three options:
- Generic document processing SaaS. Good for standard formats and average volumes.
- Industry-specific processing (logistics, healthcare, legal, finance). Higher accuracy, but you're more locked into that single vendor universe.
- Custom pipelines with OCR plus a dedicated language model. Predictable costs. Room to deviate from standard fields — industry-specific rules, multi-step approval paths, that kind of thing. One case study showed a self-hosted pipeline reaching 93% extraction accuracy across highly varied invoice formats while removing the variable per-document cost [15].
For administration and finance we've gone deeper in AI invoicing.
Category 5: Sales and marketing AI
Marketing is past the experimental phase. 75% of marketing teams report clear ROI from AI initiatives and 66% of marketers use AI daily [17]. One major CRM vendor's own marketing team saw reporting time drop from 90 minutes per person to 20. Conversions up 82%. Click-through up 50% in demand gen [17]. That's their own house, so take it with a pinch of salt. But we regularly see this same order of magnitude in our own projects with teams that take it seriously.
Three sub-types within this category:
- AI inside CRM platforms. Deepest impact if you already live in that house. Scoring, content suggestions, agent functionality on your real data.
- Stand-alone sales AI. For outbound and email personalisation. Useful in isolation. The real benefit only appears once it's connected to your CRM and marketing automation.
- Custom integrations between your tools. For mid-sized companies, this is often where the biggest leverage sits. Not so much buying a new tool as making the tools you already have work together intelligently through an AI layer that understands your business rules.
Standard tool or custom: the real decision point
Everyone hits this question sooner or later. The 2025 data is clear, and surprisingly counter-intuitive:
- Companies that build everything from scratch themselves succeed in about 33% of cases. Companies that buy or partner succeed in about 67% [18].
- 42% of companies pulled back AI initiatives in 2024 because building from scratch turned out to be heavier than expected [19].
- Companies that build strategic digital assets aligned with their core activity achieve 20–30% higher profit margins [20].
Looks contradictory. It isn't. The winning strategy is hybrid. Buy the heavy standard functionality. Build only what truly sets you apart. And use AI to assemble the integration layer between those worlds quickly [21]. It's not "build or buy." It's "own or orchestrate."
In practice:
- Standard SaaS is fine for processes that work the same in every business — email, time tracking, basic reporting. Trying to be unique here wastes energy.
- Industry-specific tools are the right call when your sector has its own rules and formats (healthcare, finance, logistics, legal).
- Custom is the choice where your process sets you apart. Where you have an edge over competitors and need to maximise that edge in software no one else has.
The mistake most companies in the 95-to-5 gap make is almost always the same: they buy ten separate SaaS tools, find each of them okay individually, and discover after a year that their data is fragmented, their processes don't fit together, and no one has an overview. The answer isn't more tools. The answer is a deliberate mix. With an implementation partner who keeps the architecture coherent. For the financial side, see also what AI implementation costs.
What the winners do differently
The 2026 numbers are genuinely good news for those who play this smart. Globally, $37 billion flowed from companies into generative AI in 2025. A 3.2x increase over 2024. The largest share — $19 billion — went to user-facing AI software, not infrastructure [22]. Companies are buying outcomes. Not toolboxes.
What stands out about that 5% with measurable value? In our projects we keep seeing the same three things:
- They start with one well-chosen process. Customer service, invoice processing, a specific marketing function. Only when that demonstrably works does the next thing get tackled. No big bang. No "AI strategy 2027."
- They mix standard with custom. No pure cowboy builds. Also no pure SaaS stack with no coherence. Instead a core of solid standard tools, with custom integration and bespoke logic at the hinge points.
- They choose on the right criteria. The shiniest demo doesn't win. The tool that works in their specific setup today, and still works two years from now, does.
For SME owners — Dutch and beyond — this is the moment. The infrastructure is ready, the models are mature, the cases are there. Your competitors are still busy experimenting. Whoever gets a few well-chosen processes in order now will, in twelve months, hold a lead that doesn't sit in tools but in how efficiently and how personally their business runs.
No prophecy. Just how it looks right now.
Stop choosing on gut feel — choose based on your actual processes
We help SMEs test, integrate, and roll out their AI shortlist — with full attention to GDPR, scalability, and ownership of your data. No hype, just results. Get in touch →
Sources
[1] Searchlab, "AI in Nederland Statistieken 2026", https://searchlab.nl/statistieken/ai-in-nederland-statistieken-2026
[2] First AI Movers, "95% AI Adoption, 5% Value Creation: How Dutch SMEs Can Close the Gap in 2026", https://www.firstaimovers.com/p/ai-adoption-netherlands-sme-2026
[3] Wolters Kluwer, "Dutch SMEs are leading the way in Europe in AI ambitions and cloud infrastructure", https://www.wolterskluwer.com/en/news/dutch-smes-are-leading-the-way-in-europe-in-terms-of-ai-ambitions-and-cloud-infrastructure
[4] PPC Land, "Dutch data authority sets GDPR preconditions for AI models", https://ppc.land/dutch-data-authority-sets-gdpr-preconditions-for-ai-models/
[5] Law & More, "Using AI In Your Dutch Business: GDPR And Compliance Risks Explained", https://lawandmore.eu/using-ai-in-your-dutch-business-gdpr-and-compliance-risks-explained-2/
[6] Invicti, "Shadow AI: Risks, Challenges, and Solutions in 2025", https://www.invicti.com/blog/web-security/shadow-ai-risks-challenges-solutions-for-2025
[7] Olakai, "Shadow AI: The Enterprise Risk You Cannot Afford to Ignore", https://olakai.ai/blog/shadow-ai-risk/
[8] OECD, "AI adoption by small and medium-sized enterprises (2025)", https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/12/ai-adoption-by-small-and-medium-sized-enterprises_9c48eae6/426399c1-en.pdf
[9] Freshworks, "How AI is unlocking ROI in customer service", https://www.freshworks.com/How-AI-is-unlocking-ROI-in-customer-service/
[10] MDPI, "Implementing AI Chatbots in Customer Service Optimization — A Case Study in Micro-Enterprise (2025)", https://www.mdpi.com/2078-2489/16/12/1078
[11] HubSpot Investor Relations, "HubSpot Launches New and Enhanced AI Agents Plus Over 200 Updates", https://ir.hubspot.com/news-releases/news-release-details/hubspot-launches-new-and-enhanced-ai-agents-plus-over-200
[12] Qualimero, "Chatbot Customer Service Case Data", https://qualimero.com/en/blog/chatbot-customer-service
[13] Arcade, "Workflow Automation Trends & Enterprise ROI Insights", https://www.arcade.dev/blog/ai-workflow-automation-metrics/
[14] Zapier, "n8n vs Make: Which is best? (2026)", https://zapier.com/blog/n8n-vs-make/
[15] Madhi, "AI invoice processing with 93% accuracy using OCR + SLM", https://www.madhi.ai/customer-stories/ai-invoice-processing-with-ocr-and-slm
[16] Artsyl, "OCR for Invoice Processing (2025-2026)", https://www.artsyltech.com/OCR-for-invoice-processing
[17] HubSpot, "AI Trends for Marketers Report 2025", https://blog.hubspot.com/marketing/state-of-ai-report
[18] MarkTechPost, "Build vs Buy for Enterprise AI (2025) — citing MIT enterprise AI research", https://www.marktechpost.com/2025/08/24/build-vs-buy-for-enterprise-ai-2025-a-u-s-market-decision-framework-for-vps-of-ai-product/
[19] Webchain, "Build vs Buy Software — How AI Changed the Decision in 2026", https://webchain.ro/build-vs-buy-software-how-ai-changed-the-decision-in-2026/
[20] TechAhead, "Enterprise AI Build vs Buy vs Partner Decision Framework for 2026", https://www.techaheadcorp.com/blog/enterprise-ai-build-vs-buy-vs-partner/
[21] CIO.com, "Build vs buy: A CIO's journey through the software decision maze", https://www.cio.com/article/4056428/build-vs-buy-a-cios-journey-through-the-software-decision-maze.html
[22] Menlo Ventures, "2025: The State of Generative AI in the Enterprise", https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/
