Applied AI research, built for real work.
We study how AI performs inside real businesses: on real systems, with real data, measured in time and money. What works becomes software.
What we study.
Four questions drive our work right now. Each one comes from a product we are building.
- Agents on real office systemsHow an AI agent can read email, calendars and CRMs safely, with read-only access, and still find something useful. Related: Spotlight
- Measuring lost timeHow to find where a team loses time, and put a number on it in hours and NZ$, without a consultant in the room. Related: Spotlight
- Tested automationHow to prove a fix works before anyone installs it, so a small office can trust automation without a technical team. Related: Spotlight
- PrioritisationHow to rank a day's work so the most valuable task comes first, and learn from each outcome. Related: Shortlist
Benchmarks don't run offices.
A lot of AI is scored on test sets. Ours is scored in hours and dollars, on real work.
How we research.
Three rules we hold every project to.
- 01
Real data, with permission
We test on real systems, with read-only access and the owner's consent.
- 02
Measured in hours and dollars
A result counts when it saves time or money someone can see, not when it scores well on a benchmark.
- 03
Ship what works
Research that holds up becomes software. Research that doesn't gets written up, so we don't repeat it.
Write-ups.
What we learn, written up in plain language.
Nothing published yet.
Our first write-ups will appear on this page. To hear when they're out, email hello@flaught.ai.
Research with us.
Bring us a problem from your business. We study it with your team and your data, then build what works.