How to Start with AI in Your Business: A Practical Guide for SMEs

Key Takeaways: Only 23% of Dutch SMEs actively use AI, while 68% say they do not know where to start. This article provides a concrete seven-step plan to go from zero to a working AI pilot project. We cover process analysis, data readiness, automation candidates, the choice between off-the-shelf and custom, running a four-week pilot, and the subsidies that cover your initial investment. The average SME can have a working AI application running within four months for less than EUR 15,000 out of pocket.

Forget the hype, start with your business processes

AI does not begin with technology but with understanding your own business processes. The companies that deploy AI most successfully do not start with "what can AI do?" but with "where are we losing time and money?" That inverted approach prevents you from investing in technology that looks impressive but solves nothing.

The media present AI as a technology that upends entire industries. That is true for some sectors, but for the average SME with 10 to 100 employees the reality is more sober and, precisely because of that, more valuable. AI is currently most effective as an accelerator of existing processes, not as a replacement for entire departments.

Research from Statistics Netherlands (CBS) shows that Dutch companies implementing AI achieve an average 12-18% productivity gain on the automated processes. For a company with 25 employees, that translates to the equivalent output of 3 to 4.5 additional staff, without extra salary costs. That gain, however, only materialises in the right applications applied to the right processes.

Start therefore with a process audit. Walk through your business and note where employees do repetitive work, where errors arise from manual data entry, where decisions rely on gut feeling rather than data, and where waiting times slow the process. This does not need to be a formal consultancy engagement. An afternoon with your team leads and a whiteboard typically yields a list of ten to fifteen improvement opportunities.

Audit your data readiness before buying any tool

Your data is the fuel of every AI application. Without structured, reliable and accessible data, even the best AI solution has nothing to work with. Invest first in getting your data foundation in order before allocating budget for AI tools.

Data readiness means three things in practice. First, is your data digitally available and structured? A company that stores receipts in a shoebox and keeps customer data in five different spreadsheets is not yet ready for AI. That need not be a problem, but it is essential to assess it realistically.

Second, is your data reliable? Research from Gartner shows that organisations lose an average of USD 12.9 million per year due to poor data quality. For SMEs the amounts are smaller, but the principle is identical. If your CRM is full of outdated contacts, duplicates and incomplete records, an AI model trained on it will produce unreliable results.

Third, is your data accessible? Many SMEs work with systems that communicate poorly with each other. Your accounting software does not talk to your CRM, your webshop is disconnected from your inventory system, and your project management runs on yet another platform. API integrations and data-integration tools can break down these silos, but that requires deliberate investment.

A realistic data-readiness assessment costs EUR 2,000 to 5,000 with an external specialist and takes two to three weeks. The result is a clear picture of where you stand, what needs to be fixed first, and how much time and budget that requires. Often it turns out that 60-70% of the necessary data is already available and that targeted improvements to data quality and integration are sufficient to get started.

Identify your top 3 automation candidates

The best AI candidates are processes that are high-volume, rule-based and data-intensive. Not every process lends itself to automation, and the art lies in focusing on the three applications with the highest impact and the lowest implementation threshold.

Use a simple scoring method. Assess each process from your process audit on three criteria: time spent per week in hours, error-proneness on a scale of 1 to 5, and data availability on the same scale. The processes with the highest total score are your best candidates.

In practice, we see the same top candidates at SMEs time and again. Invoice checking and administrative processing almost always scores high: it costs an average of 8 to 15 hours per week at a company with 30 employees, is strongly rule-based, and the data already sits in your accounting system. Customer service routing and standard responses is a second common candidate, with typically 40-60% of queries that can be fully automated. Inventory optimisation and purchasing planning based on historical sales data is the third classic.

Select a maximum of three candidates for further exploration. The temptation is to tackle five or ten processes at once, but that spreads your focus and budget too thin. Start with the candidate that scores highest and triggers the least organisational resistance. Success on the first project creates internal support for the next steps.

Choose between off-the-shelf tools and custom solutions

For 80% of SME use cases, an existing tool with configuration is sufficient. Custom development is only necessary when your process or data is so unique that no standard solution fits, or when integration with existing systems is too complex for a plug-and-play approach.

Off-the-shelf AI tools have grown explosively in quality and affordability over the past two years. Platforms such as Microsoft Copilot, Google Gemini for Workspace and industry-specific tools offer AI functionality that is immediately deployable. Costs typically range from EUR 10 to 50 per user per month, with implementation costs of EUR 2,000 to 10,000 for configuration and training.

Custom solutions become relevant when your business process deviates from the standard, when you work with sensitive data that cannot be sent to external cloud services, or when the desired integration with your existing systems is not possible via standard connectors. A custom project typically costs EUR 15,000 to 75,000 and takes two to four months.

The middle ground is gaining popularity: a standard tool with custom configuration. Think of a chatbot platform trained on your specific product documentation, or a document-processing tool configured for your invoice formats. Costs fall between EUR 5,000 and 20,000, with the advantage that you benefit from the platform's ongoing development.

Run a four-week pilot

A well-structured four-week pilot gives you enough data to make a substantiated go/no-go decision. The pilot must be small enough to remain manageable but large enough to deliver statistically relevant results.

Week 1 is preparation: define exactly which KPIs you measure, which baseline you use and what scope the pilot covers. If you are automating invoice processing, measure current processing time per invoice, error rate and lead time from receipt to booking. Record these baseline measurements in writing.

Weeks 2 and 3 are execution: the AI solution runs in parallel with the existing process. Employees still perform the work manually, but the AI processes the same input. This allows you to compare output directly without risk to your operations. During this phase, measure carefully: processing time, error rate, exceptions the AI cannot handle, and user experience.

Week 4 is evaluation and decision-making. Analyse the results: does the AI meet the predefined KPIs? Typically we see 70-85% accuracy on structured tasks after four weeks, rising to 90-95% after six to eight weeks of fine-tuning. The decision then is not whether the AI is perfect, but whether the improvement trajectory justifies the investment.

A four-week pilot typically costs EUR 3,000 to 8,000 for an off-the-shelf tool, or EUR 10,000 to 25,000 for custom development. Budget an additional 15-20 hours of internal time for project management and evaluation.

Measure results and make a data-driven decision

After the pilot, translate results into a concrete business case with payback period, so the investment decision rests on facts rather than enthusiasm. The most convincing metric for management is payback time in months.

Calculate savings at three levels. Direct time savings: how many hours per week does the AI application save, multiplied by the average hourly cost including employer charges of approximately EUR 45 to 65 for SME staff? Error reduction: what do errors currently cost you in rework, complaint handling or missed revenue? Scalability: what growth can you absorb without additional headcount?

A concrete example: a wholesale company processing 50 purchase invoices daily saves an average of 22 hours per week with AI document processing. At an hourly cost of EUR 52 including employer charges, that is EUR 1,144 per week, or nearly EUR 60,000 per year. With implementation costs of EUR 15,000 and monthly costs of EUR 400, the payback period is less than four months.

Be honest about limitations. Not every pilot delivers positive results. In roughly 30% of pilots, we find the business case is insufficient to proceed, often because data quality is too low or the process has too many exceptions. That is not a failure but a valuable conclusion that protects you from a larger misguided investment.

Subsidies that fund your first AI steps

The Dutch government allocates hundreds of millions of euros annually to innovation subsidies that are directly applicable to AI implementations in SMEs. The three most important schemes together cover up to 50-60% of your total project costs, significantly reducing your own contribution.

The WBSO (R&D Tax Credit) is the most accessible and generous subsidy for AI projects. This scheme reduces your labour costs for R&D activities with a payroll tax reduction of 32% on the first EUR 350,000 in R&D wages, and 16% above that. For SMEs with fewer than 250 employees, an increased starter rate of 40% applies. If you develop your own AI model or adapt existing models to your business process, that qualifies under the WBSO. In 2025, the scheme allocated EUR 1.7 billion to over 22,000 applicants.

The SLIM scheme (SME Learning and Development Incentive) reimburses up to 80% of the costs of training your employees to work with AI tools and data-driven processes. This includes training in using BI dashboards, interpreting AI output or managing automated processes. The maximum subsidy amount is EUR 25,000 per application for individual SMEs.

The MIT scheme (SME Innovation Incentive for Regional and Top Sectors) supports feasibility studies and collaborative R&D projects. A feasibility project receives up to 35% subsidy with a maximum of EUR 20,000, ideal for the exploratory phase in which you investigate which AI application is most suitable for your company.

By combining these schemes, you can reduce an AI project costing EUR 30,000 to an own investment of EUR 12,000 to 15,000. The application procedure for the WBSO is relatively straightforward and can be submitted quarterly. The SLIM scheme works with application windows. A specialised subsidy adviser or an implementation partner such as Stratalytic can handle the application for you, so you can focus on the substance of the project.

From first step to structural AI transformation

The first successful AI project is the starting shot, not the finish line. Companies that structurally benefit from AI build a systematic approach in which each project builds on the data infrastructure and organisational knowledge of the previous one. The step from first pilot to second and third project typically proceeds faster and more cheaply because the foundations are already in place.

After a successful pilot, the natural next step is scaling the proven application to the full organisation while simultaneously starting a second pilot for the next candidate from your top 3. Most SMEs we guide run two to three AI applications in production after twelve months, with a combined annual saving of EUR 80,000 to 200,000.

The most important thing: start. The perfect preparation does not exist, and waiting until AI is "mature enough" is the recipe for permanent delay. Your competitors who start today are building not only technological advantage but also organisational learning capacity that becomes increasingly difficult to catch up with. Four weeks and a modest investment is all you need to discover whether AI adds value to your business.