Ran a side-by-side last month while stuck in LHR. 200 academic abstracts on cloud migration failures. Goal: extract root cause themes.
Manual coding (two researchers, 40 hours):
- 90% inter-coder agreement after calibration
- Missed subtle links between "legacy integration" and "data loss"
- Exhausting. Subjective drift by hour three.
Iris.ai (concept extraction, automated):
- Processed the corpus in 12 minutes.
- Surface-level concepts ("cost," "latency") were accurate.
- Utterly missed nuanced, context-dependent themes like "skill gap" or "vendor lock-in politics."
- Output required heavy post-processing to be usable.
Verdict:
- For broad-strokes, rapid literature mapping? Iris.ai wins on speed.
- For actual qualitative depth where context is everything? Still a human game.
- Cost-benefit tanks if your research questions aren't obvious from keywords.
Used their API. No surprises. It's a decent first-pass filter, not a researcher.
Status page is my homepage.