cdn.sanity.io

Lawyers and Doctors Adopted AI Faster Than Engineers — The Regulated Industry Paradox

Regulated industries AI, Harvey, Abridge, legal AI, healthcare AI, ROI paradoxBusiness

Where Enterprises are Actually Adopting AI

Harvey Raises at $11 Billion Valuation to Scale Agents Across Law Firms and Enterprises

In just 4 months, AI medical scribe Abridge doubles valuation to $5.3B

"Legal and healthcare are too conservative for AI." That's what everyone said. Then a16z pulled actual revenue data and found the opposite — legal is #2 and healthcare is #3, right after tech. Lawyers are paying for AI faster than the engineers building it.

3-Second Summary

"Regulated = slow" myth → Real data says opposite → Regulation makes ROI obvious → Harvey, Abridge, LBOX pattern

Why is this counterintuitive?

Look at the traditional SaaS sales cycle. Legal and healthcare have always been the graveyard for software companies. Long procurement cycles, painful compliance reviews, low tech literacy among buyers. Healthcare especially — Epic and other EHRs locked the market down so tight that no new SaaS could squeeze in.

Then AI arrived and flipped the script. a16z analyzed actual contract and revenue data from Fortune 500 and Global 2000 enterprises. Right behind tech, the two industries with the most revenue momentum are legal and healthcare. This isn't a sentiment survey. It's real money flowing into these vendors.

$200M

Harvey ARR (3 years from founding)

$11B

Harvey valuation (Mar 2026)

$5.3B

Abridge valuation (doubled in 4 months)

150+

U.S. health systems on Abridge

Harvey now powers 100,000+ lawyers across 1,300 organizations in 60+ countries — including most of AmLaw 100. Abridge's AI scribe is used by 150+ of the largest U.S. health systems, hitting $117M in contracted ARR by Q1 2026.

Why is regulation a booster, not a burden?

Here's the counterintuitive insight. The conventional wisdom is wrong because regulated industries are actually the easiest place to prove AI ROI. Generic SaaS has to sell "this makes things slightly easier." Regulated industries flip that completely.

Dimension	Generic Industry	Regulated (Legal / Healthcare)
Core work	Workflows, meetings	Dense text analysis, reasoning, drafting
Hourly billing rate	Tens of dollars	Hundreds to thousands of dollars
Document volume	Dozens per week	Hundreds to thousands of pages per case
Savings per task	Modest	Thousands per case
ROI proof	"It feels faster"	"3 hours → 30 min, instantly quantifiable"

A lawyer reviewing a 1,000-page court ruling typically takes days. Korea's LBOX AI does it in 2 minutes. When you compress something that costs $500/hour into minutes, ROI math becomes trivially easy.

Regulation itself produces structured data

Here's another reason regulated industries are perfect for AI: regulation forces everything into text. Court rulings, contracts, medical records, insurance claims — all formalized text corpora. The most ideal training and reasoning data possible. Unregulated industries have messy data; regulation, ironically, cleans it up.

What's the common pattern across Harvey, Abridge, LBOX?

Looking at U.S. and Korean cases together, a clear pattern emerges.

Don't replace existing systems — bolt on next to them
Abridge grew fast because it didn't try to replace Epic. It just turned doctor-patient conversations into clinical notes — one specific task. Harvey doesn't replace law firms' case management systems either. It sits next to lawyers, helping with document analysis and drafting.
Automate prep work, not judgment
AI doesn't diagnose patients or argue cases. That's still human work. But the 70-80% of work that's collecting and organizing the information needed for that judgment? AI compresses that. Liability stays clean.
Domain-specific data is the moat
Korea's LBOX has 4,500+ practicing lawyers as users and a massive Korean case-law corpus. Casenote has 300,000+ rulings. Generic ChatGPT can't accurately cite a 2023 Korean Supreme Court ruling. Domain data is the durable advantage.
Incentives align with revenue
Law firms bill by the hour. AI lets them handle more cases in the same time. Eve, an AI for U.S. plaintiff lawyers, hit unicorn status by targeting plaintiff law specifically — where "more cases = more revenue" maps directly to AI value. AI directly generates revenue, not just savings.

How to apply this

Practical takeaways for operators, CIOs, and founders.

Question the "we're too conservative" excuse
Conservatism doesn't slow adoption. Conservative industries actually mean "high-cost-per-hour humans doing precise work" — which is exactly the cost AI cuts. If your industry hasn't moved yet, that's an opportunity, not a warning.
Automate prep, not the decision
Leave judgment, decisions, and liability with humans. Have AI handle the upstream work — collecting info, organizing, drafting. Lowest adoption resistance, cleanest liability.
Don't touch core systems
Abridge sits next to Epic instead of replacing it. If you try to replace ERP or CRM, your sales cycle becomes 1-2 years. Bolt on. Compress one workflow.
Start with the highest-billing-rate roles
Legal, medical, accounting, consulting — high hourly rates mean fast AI ROI. List the most expensive text-based repetitive work in your org. Start there.
Domain data ownership is the moat (for builders)
If you're building, the question is: who can own the domain-specific data that generic LLMs can't access? Localized professional data — Korean legal, Japanese medical, etc. — is exactly where global Big Tech can't enter. That's the opening.

Korea has its own variable: professional guild regulation

The "LawTalk" controversy showed that Korean legal and medical markets have an extra variable — professional guilds (Bar Association, Medical Association) actively regulating tech. Both Super Lawyer and LBOX position as "tools that licensed professionals use," sidestepping the issue. If you're building regulated AI in Korea, design for guild conflict from day one.

Want to dig deeper?

Where Enterprises are Actually Adopting AI The full a16z analysis — revenue momentum overlaid with GDPval benchmarks a16z.com

Harvey Raises at $11B Valuation Harvey's announcement — AmLaw 100 penetration, 100K+ lawyers, 1,300 organizations harvey.ai

Abridge doubles valuation to $5.3B in 4 months TechCrunch — Abridge's ARR and Epic integration strategy techcrunch.com

Korean Legal AI Race: LBOX, LawCompany, Casenote Hankyung — Korea's legal tech competition between LBOX, Super Lawyer, and Casenote hankyung.com

AI Automation for Regulated Industries Domino — Governance, audit, and reproducibility requirements for regulated AI domino.ai

LBOX-Casenote Consolidation Reshapes Korean LegalTech Law School Times — LBOX's 4,500+ lawyer userbase and the Casenote merger lawschooltimes.com

FAQ

Did highly regulated industries really adopt AI faster than tech?

Yes. a16z used actual contract and revenue data — not self-reported sentiment surveys — and found legal as #2 and healthcare as #3, right after tech. Harvey hit $200M ARR in 3 years. Abridge doubled in valuation in just 4 months. Regulated industries flip the curve because their ROI is so easy to prove.

My company isn't in a regulated industry. How does this apply?

Two ways. First, find the highest-billing-rate, most text-heavy repetitive work in your org and start there — that's where AI ROI is fastest. Second, if you're thinking about a startup, 'conservative markets nobody enters' are actually the most attractive ones. Conservative = expensive humans = big AI value.

Are legal and medical AI startups viable in Korea?

Already happening. LBOX has 4,500+ lawyer users, LawCompany's Super Lawyer is in production, Casenote was just consolidated, and on the medical side Lunit and Vuno are pushing globally. The Korean variable is the professional guilds — Bar Association, Medical Association — so positioning as 'a tool licensed professionals use directly' is the current playbook.

Will AI actually replace lawyers and doctors?

a16z's data and the Harvey/Abridge cases suggest amplification, not replacement. AI doesn't take over the judgment, decision, or liability — that stays human. It compresses the upstream prep work (typically 70-80% of total work). One lawyer who used to handle 5 cases now handles 50. Capacity expands; the role doesn't disappear.

As a builder, where should I be looking?

a16z called out specific zones: domains where GDPval scores jumped 20+ points in 4 months but no breakout AI startup exists yet — accounting, audit, detective work. Companies that built infrastructure and customer relationships before model capabilities crossed the threshold ended up owning those spaces. In a Korean context, tax, labor compliance, and insurance claims are similar candidates.

Written by Rush

Tracking where business meets AI.

Did you find this reference helpful?

Get curated references delivered to your inbox weekly

Share this reference

Top 20% of Companies Capture 74% of AI's Economic Value — PwC's 1,217-Executive Study Reveals the Real Gap

PwC's 2026 AI Performance Study shows that 74% of AI's economic value is captured by just 20% of companies. Here's what AI leaders do differently and how to close the gap.

Explore more AI workflow guides on similar topics