AI won't automatically make legal services cheaper
AI can speed up legal work, but it won't, by default, lower the total cost of getting a client to a good outcome. Three bottlenecks block the path from capability to access to justice: regulation, adversarial equilibria, and human decision speed. If the industry doesn't address them, we'll get cheaper documents, not better results.
Bottleneck 1: Regulation limits access and experimentation
Unauthorized practice of law rules and entity-based restrictions keep many AI tools out of consumer hands and limit new delivery models. Without room to test, measure, and iterate, we get pilots that look good on slides but stall in the wild. Regulatory sandboxes offer a path to controlled experimentation with clear guardrails and audits.
- Explore sandbox-style pilots that pair technologists with licensed oversight.
- Push for outcome-based reporting: complaint rates, resolution times, and equity of access.
- Tie approvals to measurable public benefit with sunset reviews.
See Utah's judicial sandbox for a working example of structured regulatory experimentation.
Bottleneck 2: Adversarial equilibria raise the bar, not the savings
In litigation, both sides adopt productivity tools, and the "par" shifts up. E-discovery showed this clearly: digitization lowered unit costs, then volume exploded, and parties used that volume to impose burdens. Proportionality rules help, but they rely on active case management and disciplined ESI protocols.
- Lock ESI scope early with sampling, tiered custodians, and date limits.
- Use negotiated search terms and defensible workflows to avoid volume games.
- Leverage cost-shifting and proportionality under FRCP 26 to check excess.
Bottleneck 3: Human decision speed caps throughput
Even if AI drafts in seconds, judges still resolve disputes at human speed, and lawyers still need to understand and own their filings and deals. Oversight can't be skipped; it just moves. That puts a ceiling on acceleration unless we change how matters are managed and decided.
- Shift time from drafting to review with structured checklists and issue matrices.
- Batch decisions with standardized formats to reduce cognitive switching costs.
- Adopt page, time, and hearing limits that force signal over volume.
What this means for your practice
- Firm leaders: Fund AI where it moves outcomes-cycle time to decision, likelihood of settlement, error rates-not where it just produces more paper. Set review protocols, approval thresholds, and escalation triggers before rollout.
- In-house counsel: Build vendor governance and usage policies into MSAs. Specify discovery cooperation, formats, sampling, and cost-shifting. Consider arbitration with procedural limits when disputes are document-heavy.
- Litigators: Use early case assessment to bound claims, custodians, and theories. Agree on structured exchanges (issues lists, stipulations, narrow RFPs) to avoid AI-fueled arms races.
- Transactional teams: Point AI at clause extraction, risk flagging, and playbook alignment. Keep final decisions with humans and log rationale for each deviation.
- Legal aid and clinics: Deploy AI for triage and document prep inside approved oversight models. Track client outcomes and complaint rates to support sandbox approvals.
- Courts and bar leaders: Pilot active case management, standardized orders, and presumptive limits on discovery. Require early proportionality conferences with data, not adjectives.
Practical guardrails for AI deployment
- Use retrieval-augmented workflows that cite sources and expose confidence levels.
- Mandate human review for factual assertions, citations, and client-impacting decisions.
- Log prompts, outputs, and approvals; retain samples for audits and sanctions defense.
- Strip or mask PII; confine sensitive data to enterprise instances with clear DPAs.
- Benchmark on your own matters; compare error types, not just speed.
Metrics that actually matter
- Time to resolution, time to term sheet, or time to first meaningful offer.
- Motion win rates, defect rates in filings, and rework hours per matter.
- Discovery spend as a share of total matter cost and documents reviewed per decision.
- Settlement variance from expected value and client satisfaction by stage.
Where reform can move the needle
Regulatory sandboxes can widen access while protecting consumers with audits and outcome reporting. Judicial innovations-proportionality with teeth, early narrowing, standard orders-can break the tech arms race. Expanding arbitration options with clear procedural boundaries can contain volume and speed up decisions where parties agree.
A cautious forecast
AI will make producing legal work cheaper. Client-relevant outcomes will improve only if rules, incentives, and delivery models evolve with it. The firms and courts that win will align technology with procedure, governance, and measurement-not hype.
If your team needs structured training to evaluate and supervise AI in legal workflows, see curated options by role at Complete AI Training.
Your membership also unlocks:
