AI in Your Company: Five Mistakes Most SMBs Make

Zsigmond Máriás, who founded Loginet in 2008, has spent 17 years building enterprise software, proprietary products and outsourced development for startups. Lately, more and more small and medium-sized business clients have been asking him for guidance on AI — companies that know they need to act, but aren't sure how. Drawing on Loginet's own AI implementation guide, five mistakes come up again and again.

Start With a Problem That Hurts

Most companies begin AI adoption with ambitious strategies and long planning cycles. Máriás recommends the opposite: "Pick a problem that drives you crazy, and solve that one." A typical example is a B2B e-commerce operation receiving orders through several channels and entering them by hand. Rather than automating everything at once, build something that turns a typical order into a spreadsheet — prove it, get the team to trust it, then expand into the CRM or webshop. One internal success story makes every AI project after it far easier to approve than an abstract, multi-month strategy ever could.

Messy Processes Plus AI Equals Faster Mess

A process with gaps and unreliable data will produce errors faster once AI is layered on top. "Feed it a bad prompt, ask for the wrong thing, and it'll give you the wrong answer fast," Máriás notes. Fix the process first — trace where the data comes from, find the weak points, repair the broken connections — before adding an AI layer. Input quality determines output quality, which makes this an organizational problem before it's a technical one. A related mistake: inventing a novel problem for AI to solve instead of pointing it at the tedious, error-prone work employees already dislike.

People Stay. Busywork Goes.

Companies that cut developers and blame AI are usually doing what Máriás calls "AI washing" — using AI as cover for headcount decisions that were made for other reasons. Well-run organizations instead redirect people's time from busywork to work that matters. At Loginet, landing pages that used to need a developer, a copywriter and a coordinator — and rarely got finished — now ship in a single morning, done by one person. Workshop notes that used to be skipped because writing them up took too long now get turned into summaries by AI processing the recording. In both cases, work that used to be abandoned is now actually getting done.

Fix Your Knowledge Base Before You Deploy AI

Disorganized internal documentation makes AI answer with generic, internet-average responses. Getting good answers out of it requires clean data, current documentation and knowledge that's actually accessible. Most clients say their documentation is solid and their specs are thorough — in practice, the documentation is out of date, the real knowledge lives in people's heads, and the specs are thousands of pages of inconsistencies. AI can make documentation affordable in a way it never was before, but only if the company also commits to keeping the knowledge base current, not treating it as a one-off cleanup.

Keep the AI Engine Swappable

The AI landscape moves fast — today's best model is yesterday's news. As Máriás puts it, the industry standard "has been the standard since yesterday, or for the last twenty minutes." That makes vendor independence essential: build on a specific AI platform, but "build it so you can swap it out." The model-agnostic approach means building replaceable components, so that when a better model appears from any provider, the whole system doesn't need to be rebuilt. Loginet applies the same philosophy to its own tools, designing them so any one component can be swapped out on its own.

Recommended Next Steps

Find one problem area that hurts, bring in expert help — internal or external — to assess whether AI can solve it and at what cost, and pilot it at a small scale. Once it succeeds, showcase it internally and move on to the next one. One internal success story makes the next project far easier to get approved.