Client (Anonymized)
AI Search for Industrial Documentation
The Situation
A Finnish manufacturing company with 300+ employees and decades of technical documentation. Their support team of four handles queries from architects, builders, and construction professionals. Each answer requires manually searching through document management systems for the right PDF or CAD drawing.
The Problem
A single query could take 15 minutes of manual search. Dozens arrive daily. The support team spent most of their time on information retrieval rather than the technical expertise they were hired for. Competitors had moved ahead digitally while this company still handled requests over email and phone.
The Approach
Collaborative workshop using FigJam with nine stages of scope definition
Hybrid search engine: semantic vectors + keyword matching + reciprocal rank fusion
CAD drawing (DWG) viewer with natural language dimension queries
Hallucination detection: every AI claim cross-checked against source material
Trilingual interface: Finnish, Swedish, English with 336+ translation keys
Six iterative demo sessions over three months with the client team
The Magic Moment
At the third demo session, the client's R&D manager said: 'The sun is rising.' Response time went from an estimated 2 days to under 30 seconds. Manual search dropped from 15 minutes to about 30 seconds.
The Outcome
A functional AI search prototype that finds answers across PDFs, CAD drawings, emails, and Excel sheets in three languages. Deployed within the company's existing cloud environment.
Response time: 2 days → 30 seconds
Trilingual search across PDFs, DWG files, emails, and spreadsheets
Hallucination detection with confidence scoring
Customer support panel with AI-suggested responses and ticket routing
What Transferred
Nobody at the company cares about AI. They care about finding the right installation detail without spending 15 minutes searching. The technology is invisible. The result is what matters.