Why we built Predict

For ten years I watched plaintiff attorneys settle cases at half what the jury comps said they were worth. The data was there, buried in Verdict Search and PACER and a thousand county court websites. Carriers had built their AI on top of it. Plaintiff attorneys had built nothing.

If you would rather see the number than read the argument, run your next case through the free predict-your-case calculator. It returns a jurisdiction-tuned value and its confidence band in about two minutes. The essay is here when you want the why.

I came to plaintiff personal injury law the way most attorneys do: by accident. My first job out of law school was at a regional firm in Houston that took whatever cases walked in the door. Three years in, I started keeping a spreadsheet. One column was the case at intake. Another was what we settled for. A third — added later, when I knew how to do it — was what comparable jury verdicts had returned over the prior decade.

The pattern that emerged from that spreadsheet was uncomfortable. Cases we settled at $40K had jury comps in the $90K range. Cases we passed on at intake had, six months later, returned $300K to attorneys who looked at the same fact pattern and saw it differently. The decisions weren't being made by attorneys with better legal minds. They were being made by attorneys with better data — or, more often, by attorneys whose institutional memory happened to contain the right anchor for a given case type.

By 2019 it was clear the asymmetry wasn't just within the plaintiff bar. It was between us and the carriers. Allstate had run claims-side ML on its internal settlement data since 2014. By 2019, every major US property-casualty insurer was setting reserves with a model. The plaintiff side was still running the same Verdict Search lookup workflow it had used in 1995.

The plaintiff bar didn't lose to better lawyers. It lost to better data.

What changed

In 2024 I left private practice to figure out whether the plaintiff side could build its own pricing instrument. The technical problem wasn't the model — gradient-boosted regression has been solved for a decade. The problem was the data: nobody had compiled a plaintiff-side dataset large enough, with enough case-history attributes structured for predictive modeling, to support real jurisdiction-tuned predictions.

For 24 months we built that dataset. 312,000 jury verdicts and reported settlements from 2018 forward, across 50 jurisdictions, structured with the case-history attributes that actually drive settlement value: liability clarity, injury severity, medical specials, property damage, plaintiff-counsel reputation, judge composition, and per-county settlement medians. Then we trained the model on it.

We hold ourselves to the standard we ask of you. Every number above carries a band, including the accuracy figure, and every figure is sourced on the methodology page: 312,000 verdicts, an XGBoost ensemble, a 90% confidence interval via quantile regression.

The result is Predict. It gives a confidence-banded settlement value at intake, jurisdiction-tuned, with the methodology disclosed. The number is the headline; the band is the methodology. Every prediction shows the comparable-verdict cohort. If a prediction misses outside the band, we recalibrate and disclose.

Why confidence bands matter

The instinct, when building a pricing model for attorneys, is to suppress the uncertainty. A clean number is more decisive; a number with a confidence band looks less authoritative. We almost did that. We were talked out of it by attorneys.

Attorneys are trained to evaluate uncertainty. It's the job. They will reject a number without a defense before they accept one. A confidence band is the defense — it says, here is what we know, here is the precision of what we know, and here is what would have to be true for the number to move. That's the difference between an AI tool a plaintiff attorney will bet a case on and one they'll dismiss as a black box.

What we are not

We are not selling to insurance carriers. We never will. The plaintiff-only positioning is not a marketing choice — it is the load-bearing trust contract this brand sits on. The customer class will leave the moment they smell a carrier vendor, and they will be right to leave. Predict's data flows in one direction: toward the attorneys actually fighting for plaintiffs.

We are not the next generation of legaltech AI hype. There are no manifestos here. No "10x your firm" copy. No conference-circuit visual language. The brand is calm, the language is plain, and the data does the work. We will sell by being right, not by being loud.

And we are not building a tool for big firms. The wedge is the solo and small-firm PI attorney, the practitioner who today is making case-selection decisions on instinct, anchored to property damage and a partner's gut. $499 a month. Single tier. No card at signup. Sell without negotiation.

One case anchored $50,000 low, like the $40K settlement against $90K comps that opened this essay, costs more than a year of this. We are not promising that outcome on your cases. We are saying that the math is why a band on every number is worth $499 a month.

What's next

This essay is the one moment I'm the public face of Predict. From here, the brand returns to brand-led. The data speaks. The annual State of PI Case Valuation report ships in December 2026. Predict Day — our first invite-only summit for the data-driven plaintiff bar — happens in Q4 2026 or Q1 2027. The model improves quarterly, with full methodology disclosure.

We are early, so we are not going to borrow testimonials we have not earned. The proof we can offer today is your own case, not someone else's quote.

If you are a plaintiff PI attorney and you've felt the asymmetry I described at the top of this essay, the one that defines a generation of practice, I'd ask you to do one thing. Run the next case in your intake pipeline through the free predict-your-case calculator. Look at the number. Look at the band. See whether the methodology holds up. If it does, the 14-day free trial is there. If it doesn't, write to me and tell me what we missed.

The plaintiff bar deserves the same pricing instruments the carriers have used for a decade. That's what we're building.