The Front Door of Justice Learns to Sort

Walk into almost any clerk's office in the country and you can still find the residue of the old system: rubber stamps, manila folders, and a counter where a human being decides whether your paperwork is complete enough to join the queue. That counter is the front door of justice, and for most of American history it worked the same way, first in, first served, with a clerk's eye as the only filter. State courts now process the overwhelming majority of the nation's legal disputes, handling roughly 70 million cases filed in 2024 and an estimated 98% of all cases in the United States. At that volume, the question is no longer whether to triage incoming matters, but whether software should help do it, and what it costs when the software gets the sorting wrong.

This piece traces an arc, from the paper-bound past, through a present in which automated intake and priority scoring are spreading faster than courts can govern them, into a near future where the front door becomes an algorithmic triage layer that few defendants will ever see. The technology is genuinely transformative. It is also, in its most consequential form, deeply contested.

The Old Way: A Queue, a Clerk, and a Calendar

The legacy intake model was elegantly simple and chronically slow. A litigant or attorney filed paper; a clerk checked it for completeness, conflicts, and the correct fee; the matter was docketed in roughly the order it arrived; and a judge's calendar absorbed it whenever a slot opened. Prioritization existed, but it lived in human heads and local custom, a speedy-trial clock here, a domestic-violence emergency there, rather than in any consistent, measurable rule.

The weaknesses of that model were structural. Before the pandemic, a National Center for State Courts review found that none of the 130 courts studied across 21 states met the model time standards for resolving cases, 365 days for felonies and 180 days for misdemeanors, according to an American Bar Association analysis of the cost of delayed justice. Delay was expensive in mundane ways too: the NCSC has pegged the cost of a single continuance at a conservative $35, with felony cases averaging three continuances and misdemeanors 2.2, per the same ABA review.

Then COVID-19 turned a chronic problem into an acute one. Court closures and paused jury trials caused state court filings to fall 28% in 2020, with some case types dropping by half, the NCSC's Court Statistics Project reported via the Institute for the Advancement of the American Legal System. But cases did not disappear, they accumulated. A Thomson Reuters study cited by the ABA found state-court backlogs rose 39% and county and municipal backlogs 14% between 2019 and 2021, with a third of courts reporting their backlog "increased greatly." In the 12 states the NCSC examined, nearly 400,000 additional active criminal cases were pending at the end of 2020 than at its start, a particular crisis given speedy-trial deadlines, the Court Statistics Project noted.

The federal courts tell the same story in starker numbers. Civil cases pending in U.S. district courts climbed from 340,925 in 2015 to 633,066 by 2024, an 85.7% increase, even as new filings stayed volatile, according to the Administrative Office of the U.S. Courts. A first-in, first-served queue, it turned out, has no mechanism for catching up. Something had to triage.

The Shift: From Counter Clerk to Triage Layer

The first wave of modernization was not artificial intelligence at all, it was simply digitizing the front door. Electronic filing converted the clerk's completeness check into a software validation step, rejecting incomplete submissions before they ever entered the queue. By 2025, more than 80% of court professionals surveyed said their courts use case management, document management, and e-filing systems, many of which now embed some form of automated assistance, the Thomson Reuters Institute reported. Once filings became structured data, the leap from validation to prioritization became technically trivial, and operationally irresistible.

It became irresistible because the people were running out. More than two-thirds (68%) of judges and court professionals reported staffing shortages in the prior year, and 61% expected them to continue, according to the Consortium's 2025 Survey of State Courts. In the same survey, 45% of courts reported rising caseloads while 77% encountered hearing delays every week, the "perfect storm" of more demand and less capacity that pushes administrators toward automation.

Yet courts are also among the most cautious adopters in the legal world. Only 8% of court respondents said their systems were using or planning to use generative AI, while 60% reported no current plans, the Thomson Reuters State of the Courts Report 2024 found. The result is a telling split: automated intake plumbing, e-filing triage, completeness checks, routing, is now mainstream, while the more visible, judgment-adjacent uses of AI remain stuck in pilots and committee review.

That shift is measurable where it has been studied. Across municipal courts experimenting with automated case triage, early evaluations suggest the biggest gains come in high-volume, low-complexity dockets, traffic, small claims, and family matters, where systems also improve filing completeness for self-represented litigants, the trade press covering municipal case triage has reported. In jurisdictions facing extreme volume, the upside is dramatic: a simulation using sanitized data from India's national judicial system, where over 50 million cases are pending, showed a 30% reduction in time to clear backlog when automated triage was applied, according to a study in the International Journal of Advanced Research in Computer and Communication Engineering.

Digitizing the front door was the easy part. Teaching it to decide who goes first is where the law gets hard.

What It Looks Like Now: Three Layers of Sorting

In practice, automated intake and scoring operates in three distinct layers, each with a different risk profile. The first is administrative triage, sorting filings by type and completeness. The second is docket prioritization, ranking cases by urgency, age, or statutory deadline so judges and clerks see the most time-sensitive matters first. The third, and most contested, is pretrial risk assessment, scoring an individual defendant's statistical likelihood of failing to appear or being re-arrested, to inform release decisions.

The third layer is the one that has shaped public debate, and it is far more entrenched than the cautious AI-adoption numbers suggest. Pretrial risk assessment instruments predate the current AI wave, many are simple statistical checklists, but they now reach an enormous share of the population. More than 60 distinct risk-assessment tools are in use across federal and state systems, and at least 1,000 counties have embedded some form of pretrial risk tool into their decision-making, according to a policy brief from the University of Michigan's Science, Technology, and Public Policy program. One widely deployed instrument alone, built from 1.5 million records drawn from roughly 300 jurisdictions, reaches a population of 56.3 million people, per data compiled by the Mapping Pretrial Risk project.

The fairness problem nobody has fully solved

Risk scoring's reach is exactly why its fairness debate matters. The defining moment came in 2016, when investigative analysis of one widely used recidivism algorithm in Broward County, Florida, found that Black defendants were nearly twice as likely as white defendants to be wrongly labeled high-risk without going on to re-offend, a false-positive rate of 45% versus 23.5%, even though the tool's overall predictive accuracy was roughly 61% for any subsequent offense and only 20% for predicted violent crime, ProPublica reported. The tool's developer disputed the framing, and subsequent scholarship complicated both sides. Researchers later showed that the disagreement was, in part, mathematically unavoidable: when base rates differ between groups, a tool cannot simultaneously equalize false-positive rates and maintain equal predictive value, as a follow-up ProPublica analysis of competing fairness definitions explained.

That impossibility result reframed the entire conversation. The question is not whether a risk tool can be made perfectly fair, it cannot, because "fair" has multiple incompatible mathematical definitions, but which trade-offs a jurisdiction chooses, and whether it makes that choice transparently. Scholars reviewing the field warn that tools are often "rolled out in many areas before they have been rigorously evaluated," and that defendants can rarely see the basis for the scores assigned to them, as documented by Privacy International. Validation research is genuinely mixed: some studies find that jurisdictions using structured pretrial tools see lower failure-to-appear and re-arrest rates, per a validation report from the Crime and Justice Institute, while critics argue the instruments can launder historical bias into the appearance of objectivity.

The practical upshot is that risk scoring sits at the intersection of efficiency and equity, and courts increasingly treat it as a decision-support input rather than a decision-maker. The same review that catalogued the tools' reach stresses screening risk factors for bias, testing decision thresholds before deployment, and informing jurisdictions about the relative fairness of the factors they use, a governance checklist that the research on bias in pretrial risk assessment argues should precede any rollout.

The Caseload Behind the Algorithm

It helps to remember what these systems are sorting. State court dockets are dominated not by dramatic felonies but by sheer volume: traffic and local-ordinance matters account for nearly half of all incoming cases, while civil filings, driven by contract disputes, have been climbing, the NCSC's Court Statistics Project found in its analysis of 2024 data reported by Detroit Legal News. Total state-court filings reached 64.6 million in 2022, still 18.7 million below the 83.2 million logged in 2019, per the Court Statistics Project's earlier analysis covered by the same publication.

This composition explains why automated triage is spreading fastest where it is least controversial. A traffic citation or a small-claims complaint is high-volume, rule-bound, and low-stakes, the natural first home for completeness checks and routing automation. The reputational and constitutional risk concentrates in the thin slice of criminal matters where a score can influence whether a human being sleeps at home or in jail.

The backlog is receding, the share of courts reporting growing backlogs fell from 44% in 2022 to 25% in 2024, and delay reports fell from 45% to 27%, the Thomson Reuters State of the Courts Report 2024 found. But it is receding against a backdrop of acute staffing shortages, which means the relief is being driven as much by automation and case-management discipline as by hiring.

The Next Few Years: An Invisible Triage Desk

Over the next three to seven years, expect the three sorting layers to converge into a single, largely invisible intake pipeline. Filings will arrive through structured digital portals; automated systems will validate, classify, deduplicate, and run conflict and eligibility checks instantly; and a priority tag will follow each matter through its lifecycle. For the vast majority of routine cases, no human will touch the file until an exception is flagged. The clerk's counter will not disappear, but it will become a place for handling what the software could not.

Three forces will shape whether that future is trusted. The first is governance. Because courts are slow, cautious adopters, recall that 60% had no generative-AI plans as of 2024, per Thomson Reuters, the near-term expansion will likely come through procurement standards, model-validation requirements, and bias audits rather than through bold deployments. The fairness debate has already produced a template: validate before deploying, document the chosen fairness trade-off, and give individuals a way to understand and contest a score.

The second force is the human-in-the-loop principle. The durable lesson of the risk-scoring controversy is that automated outputs should inform, not replace, judicial discretion, and that a score presented without context can quietly become a verdict. Courts that succeed will treat intake and risk systems as triage nurses, not diagnosing physicians.

The third force is transparency. The single most persistent criticism of pretrial tools is opacity: defendants rarely learn how their scores were produced, as Privacy International documented. The next generation of court technology will be judged less on accuracy and more on explainability, whether a clerk, a judge, and a defendant can all see why the front door sorted a case the way it did.

Conclusion

The courthouse queue was never neutral. It rewarded those who could afford to wait, hired counsel to navigate the paperwork, and showed up early. Automated intake and risk scoring promise something better: a front door that catches errors before they cause delay, that surfaces genuinely urgent matters first, and that frees scarce clerks and judges to focus on judgment rather than data entry. The data shows the demand is real, 70 million state filings, a doubled federal civil backlog, and two-thirds of courts short-staffed. The data also shows the peril, in a 45%-versus-23.5% gap that no clever engineering has fully closed. The courts that thrive will be the ones that automate the sorting of paper aggressively and the sorting of people humbly, and that never confuse a faster front door for a fairer one.

Sources

1. National Center for State Courts, Caseload headline statistics (70M cases filed, 98% of all cases in state courts, 2024)
2. Administrative Office of the U.S. Courts, Judicial Caseload Indicators, Federal Judicial Caseload Statistics 2024
3. Thomson Reuters Institute, Courts grapple with AI revolution amid staffing crisis (2025 Survey of State Courts)
4. Thomson Reuters Institute, State of the Courts Report 2024: What do courts think of Gen AI?
5. Thomson Reuters, State of the Courts Report 2024 (full PDF)
6. Thomson Reuters, Responsible AI in courts: problems to solve, questions to ask
7. American Bar Association, The Cost of Delayed Justice: Making Courts More Productive to Reduce Case Backlogs (2022)
8. IAALS (University of Denver), The Pandemic's Impact on State Court Filings
9. National Center for State Courts, State court caseloads during and after COVID-19
10. Detroit Legal News, Caseload data helps courts with decision-making (NCSC Court Statistics Project, 2024 data)
11. Detroit Legal News, Caseload trends revealed in state court data analysis (2022 data)
12. University of Michigan STPP, The Risks of Pretrial Risk Assessment Tools: Policy Considerations (2023)
13. Mapping Pretrial Risk, How many jurisdictions use each tool
14. ProPublica, Machine Bias: Risk Assessments in Criminal Sentencing (2016)
15. ProPublica, Bias in Criminal Risk Scores Is Mathematically Inevitable, Researchers Say
16. Privacy International, ProPublica analysis finds bias in criminal-justice risk scoring
17. Crime and Justice Institute, Validation of a Pretrial Risk Assessment Tool
18. Balancing Promise and Caution in Pretrial Risk Assessment, bias screening recommendations
19. CityGov, Case Triage and Workflow Prioritization Using AI
20. IJARCCE, AI, Cloud Integration for Scalable Judicial Data Processing (India backlog triage simulation)