6 AI Projects We Recommend You DON'T Build

Half the AI projects we get asked about should not exist. The technology can do the thing — that is not the question. The question is whether the thing should be done at all, and whether an AI-shaped solution is better than the three simpler options that already work.

At BITBOX.lv we turn down roughly one in three AI project requests. Not because we do not want the work — because shipping the project would make the client's business worse, not better. Here are six patterns we consistently recommend against, and what to do instead.

Six Patterns We Turn Down

1. A chatbot that just replaces your FAQ page

LLMs hallucinate, cost money per query, add latency, and most users do not actually want to chat with a bot for a yes/no answer. A searchable FAQ page or a simple decision-tree bot does the same job cheaper and more reliably. Use AI when the question genuinely requires reasoning over a customer's account data or multiple sources — not when a well-structured page would answer it.

2. AI that writes your brand content

Your brand voice is the thing that separates you from competitors. LLMs produce statistically average prose — the opposite of distinctive. Google is actively detecting AI-generated content, and most of it quietly loses positions. Editors end up spending more time fixing AI drafts than they would have writing from scratch. AI is genuinely useful for research, outlines and first drafts of technical docs — not for the public voice of your company.

3. An autonomous agent that makes critical decisions without humans

If a decision affects money, legal obligations, customer health, or reputation — put a human in the loop. Not because AI cannot decide, but because the cost of a wrong decision is asymmetric. One hallucination on a refund approval costs you €50. One hallucination on a contract clause or a medical triage can end the business. Autonomy is a privilege the system earns after months of human-reviewed operation, not a feature you ship in v1.

4. One “does everything” assistant for your whole company

It sounds efficient: one agent that answers HR questions, reads sales data, reviews code and schedules meetings. In production it is always worse than four focused agents. Context windows fill up, hallucinations rise, quality drops, and nobody trusts the output for anything. Specialised agents with narrow scope beat generalists every single time.

5. A natural-language layer over your database (instead of a dashboard)

“Ask our company AI anything about the business” demos brilliantly. In production: queries are slow, answers are inconsistent, numbers drift between runs, and executives stop trusting the output within a month. For structured data, SQL plus a proper BI dashboard (Metabase, Looker, Tableau) always wins. Use AI here only when questions genuinely span structured and unstructured sources — contracts, emails, spreadsheets at once.

6. AI for hiring — screening, scoring or firing

The EU AI Act classifies employment-related AI as high-risk. Legal liability is real. Bias in training data amplifies at scale. Candidates detect it and walk away. The reputational damage from one candidate publicly sharing a bad experience outweighs any efficiency gain. Hiring is one of the few places where slower and more human is the competitive advantage — not the cost.

What to Build Instead

The pattern behind all six: AI is a hammer, but not everything is a nail. The best AI projects we build have four things in common — the problem already exists without AI, the task is high-volume and boring, the cost of a mistake is low, and there is a clear human escalation path.

If you are not sure whether your idea belongs on this list or the “good ideas” list, tell us about it. We will give you an honest read. Sometimes the answer is “do not build it,” and that is the advice that saves you the most money.

Get in touch and we will tell you what is realistic for your specific business.