June 2026 - You're solving the wrong problem with AI Search

What you'll learn

Misunderstood failure patterns holding back your AI search

Most search failures are misdiagnosed. Learn to name what you're experiencing before you reach for the wrong solution.
Decades of wisdom from the front-lines of top search teams

Lessons learned across diverse clients of companies like at Qdrant, Voyager Search, Lucidworks, OpenSource Connections.
10 patterns. Every search system fits one.

A diagnostic map across ecommerce, media, legal, finance, geospatial and more, with the signature failure mode for each.
Agentic retrieval inherits your broken archetype

It's not a new problem. It's your existing problem, now with an agent on top.
Why this topic matters

Many teams waste time implementing AI Search solutions that solve the wrong problem. They keep rebuilding relevance when they have a catalog problem. They add vector search when there's a schema problem. They tune ranking, but have no signal. Every search failure belongs to a recognizable pattern, and this session gives you the diagnostic layer top search consultants learn, but most teams skip.

Transcript

1. You're solving the wrong problem with AI Search! Atita Arora
3. How many of you have shipped an AI feature in the last 12 months?
4. How many of you diagnosed “the problem” before you picked the tool?
5. That gap is what this session is about! "We fail more often because we solve the wrong problem than because we get the wrong solution to the right problem." Russell L. Ackoff, Redesigning the Future, 1974
6. "Before you fix your search, know what's actually broken." unscriptedai.substack.com Earlier this month I started a publication , specifically because this gap bothered me enough to write about it ! Nearly two decades in Search / IR AI just made this mistake significantly more expensive. Problem diagnosed vs problem solved
7. The symptom looks the same but the problem never is ! The Diagnosis Problem !
8. Not zero results. Confidently wrong results. Search isn’t aware that it is same as Paracetamol Someone did the synonym mapping
9. The moment you name the pattern, the fix becomes obvious !
10. 10 patterns. Every search system fits one. 1. The Uniqueness Problem 2. The Complexity Machine 3. The Precision Mandate 4. The Firehose 5. The Extraction Pattern 6. The Media Vault 7. The Knowledge Graph 8. Geospatial 9. Q-commerce 10. Code Search
11. #1 The Uniqueness Problem E.g. Marketplaces , C2C , Pre-loved , Pre-owned Issues here : - Keyword Stuffing - Incomplete Metadata - Inconsistent Vocabulary - Missing Canonical form - Ephemeral Inventory You can't rank what you can't retrieve!
12. What actually fixes it : - Attribute extraction at ingestion time - Seller assisted structuring - Visual Similar retrieval - Query expansion from purchase signals Fix is at ingestion. Ranking can't save a broken catalog. #1 The Uniqueness Problem
13. #2 The Complexity Machine or #4 The Firehose E.g. Retail , Ecommerce or News , Live events Catalog is usually fine , the war is in ranking Issues here : - Search optimizes relevance , - Most relevant docs - 3yrs old Business optimises revenue Most recent docs - noise
14. #2 The Complexity Machine or #4 The Firehose What actually fixes it : - Ranking as a business decision - not as a relevance problem - Freshness as a first class ranking signal - not as a filter Better metadata, more attributes were never the actual problem.
15. #7 Knowledge Graph : Neither works - Synonym/Metadata helped you with Acetaminophen -> Paracetamol - Ranking can help you get the most sold variant (promoted brand) on the top - Question like "Acetaminophen with liver disease" or "Acetaminophen with Warfarin" Drug -> interacts with -> Warfarin / Alcohol / ? -> Dose / Impact -> metabolised by -> Liver -> threshold Traverses the intersection of relationships. Gets you to the answer
16. #7 Knowledge Graph E.g. Research , B2B Supply Issues here : - Taxonomy gaps gets mistaken for ranking failures - Entities exist - Relationships between them don’t - Wrong retrieval unit - documents vs nodes You can’t rank your way to a relationship
17. - Each pattern breaks differently. - Each pattern needs a different fix at a different layer. - Now imagine you've misdiagnosed your pattern, and you put an agent on top of it
18. How do you know which archetype you're in? I go deeper in forthcoming writeups unscriptedai.substack.com
19. Agents are a consumption pattern They inherit whatever sits underneath - correctly diagnosed or not! A smarter interface on a misdiagnosed system is still a misdiagnosed system!
20. Diagnose first. Then Build Full Diagnostic map - unscriptedai.substack.com Follow for what comes next !