Methodology
How we build each case
Lumiary documents AI that real companies run in production, so you can learn from what worked and what did not. Each case answers the same questions: what problem they had, what they built, which models and partners they used, how they rolled it out, and what results they reported.
What qualifies
- A named company has deployed AI with real users and reported an outcome.
- At least one primary source: the company itself, or a vendor case study that names it.
- Plans, pilots without results, and announcements of intent do not qualify.
- Reversals and failures do qualify. We record what went wrong.
Evidence levels
Each case carries one label, so you can see at a glance who stands behind the claims.
- Independently reported
- An independent news outlet, an audited filing or a court record confirms the deployment or its figures.
- Company reported
- The deploying company states it, in a blog post, press release, earnings call or talk. Not independently confirmed.
- Vendor reported
- A vendor's case study names the customer. Useful, but the vendor has a reason to present results well.
How we handle numbers
- Every figure links to the source that states it.
- We copy figures exactly. We do not round, combine or compute new ones.
- If a model is not disclosed, we say so. We never guess it from a vendor relationship.
Summarize and link
We write every case in our own words and link to the original sources. We quote at most one short sentence per case, with attribution. To read the full story, follow the sources.
Corrections
If a case contains an error, or a company has published new results, tell us at corrections@lumiary.ai with a link to the source. We update the case and note the date.