See the case: Every accident, without searching by hand
This was a proof of concept with a client and it isn’t in operation. Open the sheet to see what was built and what was tested.
Every accident, without searching by hand
A proof of concept so the firm learns about each accident without searching for it by hand.
A law firm’s investigators checked news sites and portals by hand every 2 hours, source by source, and emailed each story to the attorney to decide.
A system checks news, search engines and local media every 30 minutes to 2 hours, drops what isn’t an accident, pulls out the details and leaves each case in a dashboard; the clearest ones arrive by email right away.
- 01An investigator checks news and portals every 2 hours
- 02Searches keyword by keyword, source by source
- 03Emails each story to the attorney
- 04The attorney replies whether to proceed
- 01The system checks the sources every 30 minutes to 2 hours
- 02Drops duplicates and anything that isn’t an accident
- 03Pulls out the type, place, date and people mentioned
- 04Clear cases arrive by email; all of them land in the dashboard
- If the system has low confidence, the story doesn’t enter the dashboard
- With medium confidence it enters the dashboard, with no email alert
- Only the clearest cases send an email right away
The attorney sees each detected story with its summary, the original link, the type of accident and how confident the system is. From there they mark it to proceed or dismiss it, assign it to an investigator and add notes. Keywords and sources are edited in the same dashboard.
Simplified screen of the proof-of-concept dashboard. The keywords are the firm’s own; the rows are illustrative.
- Mar 13, 2026Discovery meeting with the firm
- Mar 16–19We build the proof of concept
- Mar 25–26Tests with real news: the first two sources save accidents to the dashboard
- Jun 2026Code and documentation ready to continue it
- PendingTest the remaining sources and launch it
The same setup could watch other topics in news and publications: mentions of your brand, your competitors or your industry. In the call we look at which sources your team checks today, how often, and what it does with what it finds.
- It’s a proof of concept: it isn’t in operation and the firm doesn’t use it today.
- Two of the sources were tested with real news and saved accidents to the dashboard; the rest weren’t fully tested.
- With many local media sources active at once, the checks became unstable; it worked well with about 15.
- We didn’t measure how much sooner the firm learns about an accident, or how many stories it catches compared with a manual search.
- The scope was detecting and sorting the stories; investigating each case is still the firm’s work.