Every personal injury firm in Los Angeles runs the same funnel: calls come in, somebody decides which ones are worth a file, and the rest are declined. For most of the last decade that first decision was made by an intake coordinator working from a one-page script.
Over the past eighteen months, a recognisable pattern has emerged among firms that replaced that script with a software-assisted first pass: an early version that made things worse, a rebuild, and a quieter set of gains than the vendor pitch promised.
What did intake look like before you changed anything?
Two coordinators, a phone queue and a spreadsheet. The script asked about the date, the injury, whether there was a police report and whether they had seen a doctor. If the answers looked reasonable, it went to an attorney for a second look — usually within two days, sometimes within a week if we were busy.
The problem was never the obvious cases. It was the middle. A call that sounded thin at 4pm on a Friday got a different answer than the same facts on a Tuesday morning.
What went wrong with the first version you rolled out?
We pointed it at three years of closed files and asked it to predict which new calls resembled the ones we had settled well. That sounds sensible and it is completely backwards. It learned to reproduce our old declines, including the bad ones.
Two months in, we noticed the decline rate for calls that came in Spanish was running several points higher than the English queue, with no difference in the underlying facts. That was our own historic pattern being handed back to us with a confidence score attached.
How did you fix it?
We stopped asking it to predict outcomes and started asking it to extract facts. It now pulls the same twelve data points every time — date of loss, treatment gap, reported liability, policy limits if known — and flags what is missing. It does not score the case. A person still decides.
That sounds like a downgrade. In practice it was the whole benefit. The coordinators stopped re-typing, and the attorney reviewing the file sees the same twelve fields in the same order on every single matter.
What are you actually measuring now?
Time to first attorney contact. That is the number on the wall. It was 46 hours across the six firms when we started asking; it is 9 hours now at the firms that finished the rebuild.
We also audit declines monthly — a random sample of twenty, re-reviewed blind by an attorney who did not make the original call. That audit is what caught the language problem, and it is the part nobody wants to pay for.
What would you tell a five-person firm considering this?
Do the boring half first. Write down what your twelve fields are and get every person answering the phone to capture them the same way. If you cannot do that on paper, software will not fix it — it will just make the inconsistency faster and harder to see.
It learned to reproduce our old declines, including the bad ones. The score just made them look objective.Illustrative quote, composite managing-partner persona
What we took away
- Firms that asked the software to extract facts reported gains; firms that asked it to score cases reproduced their own historic bias.
- Median time to first attorney contact fell from 46 hours to 9 across the firms that completed a rebuild.
- Every firm that caught a problem caught it through a blind monthly audit of declines — not through the vendor dashboard.
- Standardising the intake questions on paper came before, and mattered more than, any tooling.
Method
This piece is an editorial composite illustrating a pattern of intake automation reported among Los Angeles County personal injury firms ranging from roughly 5 to 34 attorneys — it is not a transcript of interviews with six named firms. Figures such as the 46-to-9-hour change are illustrative and have not been independently audited.