Search: Intent, Not Inventory

Redirect

Canonical note: Search Intent Not Inventory

Source: https://medium.com/@dtunkelang/search-intent-not-inventory-289386f28a21
Author: Daniel Tunkelang

Summary

One of Daniel Tunkelang’s most cited pieces. The central argument: search systems that optimize for finding documents that match the query (“inventory”) fail users who need their underlying information goal satisfied (“intent”).

The Inventory Model (What Most Systems Do)

Most search systems operate as “inventory systems”:

  1. User submits query
  2. System finds documents containing query terms (or semantically similar)
  3. System ranks documents by relevance to query
  4. User presented with documents

This works when the user’s intent = “find documents about X.”

But most search intents are not “find documents about X.”

The Intent Model (What Users Actually Want)

Users search to accomplish goals:

QueryInventory ViewIntent View
”flu symptoms”Documents about flu symptomsDetermine if I have flu / what to do
”Paris restaurants”Restaurant review pagesMake a reservation tonight
”how to tie a tie”Tutorial pagesLearn to tie my tie right now
”competitor product X”Review pagesDecide whether to buy vs. competitor

The inventory model returns the same results for all users with “flu symptoms.” The intent model might return different results for “concerned parent” vs. “person who has been sick for a week.”

Why This Matters for Search Design

Implication 1: Retrieval is not the end The search system’s job isn’t to return relevant documents — it’s to help the user achieve their goal. A relevant document that requires 10 minutes of reading is less useful than a direct answer snippet.

Implication 2: Features beyond text relevance matter For “Paris restaurants” intent: recency, hours, distance, price range matter as much as text relevance.

Implication 3: Zero-result is sometimes correct If no document satisfies the intent, returning poor matches is worse than returning nothing (with a helpful alternative suggestion).

Implication 4: Conversational search is natural Intent-based search naturally leads to dialogue: “What’s your occasion? What budget? Dietary restrictions?” Inventory-based search is one-shot; intent-based is conversational.

Agentic Search is the architectural realization of intent-based search:

  • Agent understands intent (not just query)
  • Agent plans retrieval strategy around intent
  • Agent verifies if retrieved content satisfies intent
  • Agent reformulates if intent is not yet satisfied

People