All posts
·5 min read·Pricing

The economics of AI lease abstraction (vs human review)

Human lease abstraction runs $1,200–$2,500 per file. AI extraction starts under $25 per lease and drops below $10 at volume. Here's the cost stack on each side and where the savings actually come from.

Calculator and US dollar bills laid out on a desk
TL

The LeaseBrief team

A typical broker will charge $1,200-$2,500 per lease abstract for a third-party human review. AI extraction at LeaseBrief starts at $25 per lease on the low tier and drops below $10 per lease at volume. The gap isn't because AI is “cheap” — it's because the cost structure is fundamentally different.

The traditional cost stack

A human lease abstract — done by a paralegal, junior associate, or specialized abstracting vendor — has roughly this cost stack:

  • Labor: 4-6 hours × $80-150/hour fully-loaded = $320-900
  • Software / overhead: $50-150 per file
  • QA review: 1-2 hours × senior rate = $200-400
  • Profit margin: 25-50% on top

Net: $1,200-$2,500 per abstract, with 24-72 hour turnaround.

The AI cost stack

For an AI extraction (LeaseBrief uses Anthropic's Claude API), the marginal cost per lease looks like:

  • Tokens: ~$0.50-$1.50 per lease in API costs (varies by document length and model tier)
  • Storage: pennies per lease
  • Compute / queue: pennies per lease
  • QA: $0 in the typical flow — the model grounds every field to a citation, and the customer audits in seconds
  • Fixed costs (engineering, support, infrastructure): amortized across all customers

The math: variable cost is roughly $1-$2 per extraction. The rest of the price covers fixed costs, support, and product development. At our smallest tier ($25/lease), gross margin is tight; at the volume tiers ($10/lease), the same fixed costs spread over more files.

Why the savings exist

AI doesn't replace a single hour of human work — it replaces the entire 4-6 hour read. Specifically:

  • Volume read: Claude reads a 200-page lease in 10-30 seconds. A human reads 25-30 pages an hour.
  • No fatigue: page 187 gets the same attention as page 1.
  • Consistent schema: every abstract uses the same 60+ field structure. Humans drift.
  • Citations are free: the model grounds every field to a section automatically. A human grounding citations doubles the time.
  • Re-extraction is cached: prompt caching makes amendment processing nearly free on the second pass.

Where humans still matter

AI extraction isn't a replacement for legal counsel. The remaining high-leverage work:

  • Lease negotiation and amendments
  • Edge cases — non-standard riders, cross-collateralized rent, ground leases with unusual structure
  • Dispute review — when a CAM audit gets contentious
  • Strategic decisions: should you exercise this option, contest this charge, or restructure

The labor that gets compressed is the bulk-read pass. The high-judgment work stays human and gets more valuable because the bulk pass is no longer the bottleneck.

The hybrid model

Most LeaseBrief power users run a “trust but verify” flow: AI extracts in 60 seconds, a human reviewer scans the high-stakes fields (rent, term, options, defaults) in 5-10 minutes, the whole abstract is on the calendar in under 15 minutes total. Effective cost: $11-25 per lease in compute, plus 15 minutes of an analyst's time. Versus 4-6 hours and $1,500.

Why we publish our pricing

Most lease-abstraction vendors quote per-deal. We publish a flat tier list because:

  • Buyers can model their cost without a sales call
  • The unit economics are public, so customers know what they're paying for
  • It forces us to keep the pricing honest as compute costs drop

Full pricing is on our pricing page. The economics are designed to make the smallest tier viable for solo brokers and the largest tier reasonable for portfolio owners running thousands of leases.