Why most teams make AI-features that don't add value

Over the past years, I've witnessed numerous AI projects fail. The pattern is often the same: a team gets tasked with "building something with AI," develops a feature, and nobody uses it. This happens most frequently with feature teams - teams that build features without measuring results.

For any software product, adding real value is challenging without a product team approach. Many software features see minimal usage - the result of teams focusing on launching features rather than solving problems. While you might get away with a feature team approach in traditional software development, it's particularly risky with AI applications.

AI presents three unique challenges:

1. The question isn't whether your AI application works, but how well it performs. Your first version likely adds minimal value

2. Performance often degrades due to changes in user behavior or data

3. AI models can produce unexpected results that you'll only discover in practice

To address these challenges, you need a systematic approach to measurement, feedback collection, and iteration. The faster you can execute this cycle, the better your product becomes.

These challenges make a product team approach essential. It's no coincidence that organizations leading in AI - like OpenAI, Anthropic, and Meta - work according to product team principles:

• Setting measurable goals

• Monitoring results

• Allowing room for iteration until features add value

If you know upfront you can't implement this approach, be honest: without it, an AI feature is unlikely to succeed. Spend your time and energy on other projects without AI components.

The transition from feature team to product team is crucial for successful AI implementation. It requires a new mindset: shifting from 'delivering features' to 'solving problems.' You must be willing to experiment and sometimes admit when an approach isn't working. This approach not only creates better AI features but improves the quality of your entire product.