Small businesses face a different AI challenge than enterprises: limited budget, limited technical staff, and limited time to experiment. The right approach is not to start with infrastructure or custom models - it is to find the three or four points where AI saves meaningful time with off-the-shelf tools, prove the value, then decide whether deeper investment makes sense.

Quick Wins With Existing Tools

Before building anything, look at what the tools you already use can do:

Email - Gmail and Outlook have built-in AI assistance. Beyond the built-in tools, connecting your email to Claude or ChatGPT via their consumer interfaces costs 20-25 EUR/month per user. Use it for drafting client emails, summarizing long email threads, and researching topics you are writing about.

Document processing - If you handle invoices, contracts, or forms, you likely have manual re-keying work that AI can eliminate. AWS Textract costs fractions of a cent per page for structured data extraction. For a small business processing 500 invoices per month, the AWS cost is under 10 EUR. The integration work is the main investment.

Customer support - If you receive 50+ similar inquiries per month, an FAQ chatbot handles them without human intervention. Tools like Intercom and Zendesk have AI answer-bots built in. Custom implementations require more investment but handle more complex cases.

Meeting summaries - Tools like Otter.ai, Fireflies, or Notion AI transcribe and summarize meetings automatically. For businesses with frequent client calls, this saves 20-30 minutes per meeting in note-taking.

The 3-Step Evaluation

Before spending money on AI, answer three questions:

  1. What takes the most repetitive time? - Pick the task where someone spends the most hours doing work that follows a consistent pattern. Pattern = automatable.

  2. What is that time worth? - If automating saves 5 hours/week at 50 EUR/hour, that is 13,000 EUR/year. Most small-business AI tools cost under 2,000 EUR/year. The math is usually simple.

  3. What does failure look like? - For customer-facing automation, a bad response has a real cost. For internal processes, the risk is lower. Match the automation level to the risk tolerance.

When Off-the-Shelf Tools Hit Their Limits

Generic AI tools stop working well when:

  • Your terminology is specialized and the AI does not know it
  • Your process has specific rules or exceptions that generic prompts cannot capture
  • You need to integrate with proprietary internal systems
  • Volume is high enough that per-call API costs become significant

At that point, a custom solution makes sense. “Custom” does not mean building a model from scratch - it means building a workflow that uses AI APIs (Bedrock, OpenAI) with your own prompts, your own data, and your own integration logic. A focused custom implementation for one specific process typically costs 15,000-40,000 EUR and pays back in under a year at meaningful automation rates.

Common Mistakes

Starting with infrastructure - Small businesses do not need a data lake or ML platform. They need one working automation. Start with one process.

Automating broken processes - If a process is chaotic or inconsistent, automation makes it fast and chaotic. Fix the process first.

Forgetting about humans in the loop - The first version of any automation should have a human review step. Remove human review gradually as confidence in the AI output builds. Removing it on day one leads to problems.

Evaluating AI by demo, not by production - Demos are cherry-picked. Test with your actual data, including messy and edge-case examples, before committing.