Last year, I spent weeks watching people look for 🍆 online.

I was leading a team designing and building the next generation of “AI powered” grocery search for a big retailer, and we discovered a fascinating disconnect: most online grocery experiences fundamentally misunderstand how humans think.

Throughout our user research, we saw the same thing over again:

⌨️ “purple vegetable” → 0 results 😕 ⌨️ “purple vegetables” → 0 results 😠 ⌨️ “aubergine” / “eggplant” → … 🤔

I call it the purple vegetable test. Try it yourself - search any grocery site/app for “purple vegetable”. Most fail spectacularly.

The problem

Most online grocery experiences are:

  • Built around rigid product catalogues, not flexible human thinking
  • Interfaces designed for precision, not the messy reality of how we think
  • Technology that demands we adapt to it, rather than it adapting to us

The solution

What we built was pretty cool - a new type of grocery search that works how people actually think, with a “human-centred AI layer” that bridges natural language with the retailer’s catalog architecture:

→ When you type “summer bbq”, it understands you’re planning an outdoor gathering and suggests complementary items → It creates custom categories that never existed before your search - like “Quick marinades with pantry ingredients” or “Crowd-pleasing sides that travel well” → It distinguishes between a casual weekend cookout and a holiday celebration, adapting suggestions appropriately → It knows when to show time-saving techniques alongside products because it understands that sometimes the struggle isn’t what to cook but how to cook it faster.

The approach

The startup mindset would say “disrupt everything” - tear it down, rebuild from scratch. But that ignores the reality of enterprises. We recognised that decades of catalog development contained pricing rules, inventory management, and merchandising wisdom - the backbone of the entire business. We didn’t replace the system; we built an intelligent bridge between human language and machine architecture that amplifies the strengths of both.

The impact

Massive: less time searching, less frustration, higher conversion rates. But the beautiful part was that to users it didn’t feel revolutionary - it felt natural. To me, this exemplifies AI’s greatest promise: dramatic improvements that feel like they should have always been this way. The magic wasn’t the LLMs or semantic search. The magic was that we built a grocery experience that feels familiar while being fundamentally more human.

This is the unsexy reality of effective AI: not chatbots and image generators, but invisible intelligence solving real problems while balancing technical innovation with business realities.

I think the “purple vegetable test” can be applied to any digital experience. Does your technology understand what humans actually mean, or does it force people to think like machines?

The companies that get this right will win the next decade.

-

This post was adapted from my original LinkedIn post published in May 2025.