AI Won’t Save a Bad Product

AI is everywhere in product pitches right now.

Add an AI assistant.

Add predictive intelligence.

Add computer vision.

Add GenAI.

Suddenly the product sounds more advanced.

But here’s the uncomfortable question:

Does the user actually need it?

Because AI can make a good product better.

It can also make a bad product more expensive.

The first question: What problem does AI solve?

Don’t start with:

“Where can we add AI?”

Start with:

“What problem does the user have?”

For hardware, AI can make sense when there is a meaningful relationship between data and decision-making.

For example:

• predictive maintenance
• computer vision
• sensor-data interpretation
• anomaly detection
• intelligent classification
• natural-language interfaces

But “AI-powered” isn’t itself a user benefit.

The second question: Can we validate it without building everything?

One of our favourite approaches is to test the experience before building the entire technical system.

This could mean:

• manually processing data
• using a Wizard-of-Oz workflow
• using an existing model
• testing the user interface first
• validating whether users actually act on the output

If users don’t care about the output when it’s manually produced, building a sophisticated AI infrastructure probably won’t fix that.

The third question: What does it cost at scale?

A prototype can make almost anything look affordable.

Production is different.

Think about:

• cloud compute
• API costs
• data storage
• inference
• latency
• connectivity
• device power consumption
• maintenance

An AI feature isn’t free simply because the model already exists.

AI should solve a problem, not decorate a pitch deck

Before recommending AI, we ask:

1. Does this solve a real user problem?

2. Can we validate the value before building the full system?

3. Does the economics work at scale?

If the answers aren’t clear, we don’t force AI into Version 1.

Sometimes the smartest AI decision is:

Not yet.

The goal isn’t to make a product sound intelligent.

The goal is to make the product useful.

And sometimes the most intelligent engineering decision is knowing where AI doesn’t belong.

Building a connected product and wondering whether AI belongs in Version 1? Start with the user problem, not the model.