Endex ditches RAG for OpenAI reasoning, claims 70% analyst approval
Curated by the Inblix editorial team
Most AI tools aimed at finance lean hard on retrieval-augmented generation — essentially smarter search. Endex is betting that’s not enough. The AI platform is building what it calls an AI Analyst, and rather than just surfacing documents, it’s using OpenAI’s o-series reasoning models to actually think through financial data. CEO Tarun Amasa frames the distinction bluntly: “Finance professionals don’t just need search results; they need structured thinking and deep analysis.”
The approach means Endex agents can autonomously chew on tasks that normally eat up junior analysts’ nights and weekends — prepping investment committee memos, summarizing earnings, or running due diligence in a data room. The output isn’t just text either; the system can deliver findings as Excel models, slide decks, or formal emails with full source traceability. That audit trail is non-negotiable in finance, where a missed adjustment in an EBITDA reconciliation or an overlooked change-in-control clause can crater a deal thesis.
Endex’s work with OpenAI spans several models, but the company reports a real leap when it moved from chaining together complex prompts to relying on o1’s native reasoning. In blind tests, financial experts preferred responses from the o1 model 70% of the time over non-reasoning counterparts. Separately, the newer o3-mini model slashed latency to one-third of previous levels per turn, which means the system can now grind through dense confidential information packages and automates financial model reconciliation at a speed that starts to feel practical for live deal work.
Co-founder Pratham Soni points to the collaboration with OpenAI as key to tailoring model behavior to a professional’s expectations. The engineering team built a testing framework that lets them watch reasoning depth and response latency in real time. That kind of observability matters because in finance, you don’t just need the right answer — you need to see the work. If Endex can consistently deliver that, it won’t just augment analysts; it’ll reset what firms expect from their first-year associates.
💡 Key Takeaways
- Endex is explicitly rejecting simple RAG architectures in favor of OpenAI’s reasoning models, aiming to replicate how a human analyst interrogates financial data rather than just retrieving it.
- In blind user tests, finance professionals preferred outputs from OpenAI’s o1 reasoning model 70% of the time over non-reasoning models, signaling a real quality gap in typical enterprise AI search.
- Switching to the o3-mini model cut per-turn latency by roughly two-thirds, making complex, multi-step analyses like automated financial model reconciliation viable for time-sensitive deal workflows.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.