Brief, mostly signal
Lexifina says a firm-wide knowledge graph cut agent time 73%
The vendor reports equal accuracy on five questions, with 92% fewer input tokens and 66% fewer model calls. The numbers are unverified vendor claims.
What the research found
Lexifina reports that a firm-wide knowledge graph let an agent answer five firm-wide questions with equal accuracy, 73% less time, 92% fewer input tokens and 66% fewer model calls, with little model spend to build it.
Claims and how well the evidence supports them
- Agent with a firm-wide knowledge graph answered five firm-wide questions with 73% less time 98%
- The graph approach used 92% fewer input tokens read 98%
- The graph approach used 66% fewer model calls 98%
- Building the graph took almost no model tokens 99%
Evidence
- lexifina.com https://lexifina.com/blog/building-full-firm-knowledge-graphs-to-dramatically-speed-up-agents
- news.ycombinator.com https://news.ycombinator.com/item?id=50167523
Actionability rated low.