Enterprise Legal Project

Separation Agreement & Severance Calculator

The employment legal team's highest-priority automation target. Generates jurisdiction-specific separation agreements across US, EMEA, and Australia. Validates proposed severance against company policy before sign-off.

Status: Deployed
Platform: Glean
Department: Employment Legal
Volume: ~100 agreements/year

Why it was built

The employment legal team handled roughly 100 separation agreements a year across three jurisdictions — US, EMEA, and Australia. Each agreement required manual drafting from templates, jurisdiction-specific language, and severance calculation validated against company policy. A single agreement took 2–4 hours of attorney time. Errors in severance calculation or jurisdiction-specific clauses carried real legal and financial exposure.

This was the team's highest-priority automation target — high volume, high repetition, high consequence for error. Before AI, the only option was more headcount or accepting the risk.

Before
Manual drafting from Word templates. Attorney cross-referenced policy documents, tenure data, and jurisdiction requirements manually. Each agreement: 2–4 hours. Severance calculation errors caught in review — sometimes after sign-off.
2–4 hrs each
After
Agent generates complete jurisdiction-specific agreement from structured inputs. Severance validated against policy automatically. Attorney reviews output rather than drafting from scratch. Consistency across all three jurisdictions.
~30 sec each

What it does

A Glean agent that takes structured inputs — employee tenure, role, location, separation type, salary — and generates a jurisdiction-specific separation agreement with the correct language for US, EMEA, or Australian employment law. A parallel severance calculator validates the proposed severance package against company policy, flagging discrepancies before the agreement reaches the attorney.

The agent's access controls wall employment legal data off from commercial legal (ABA Model Rule 1.6) — the exact confidentiality gap that had blocked adoption of prior contract-management tools. This wasn't a feature add. It was the architecture requirement that made the project possible.

Why this mattered: The confidentiality boundary between employment and commercial legal isn't negotiable. Prior tools couldn't guarantee it, so the team stayed on manual processes. This project proved the boundary could hold while still delivering automation. The access-control architecture became the pattern for every subsequent legal AI project.

Metrics & improvements

~190 hrs/yr
Attorney time saved
100%
Policy-compliant severance
3
Jurisdictions covered
~100
Agreements/year
~$47K
Recovered per year

Improvement: 2–4 hours of manual drafting and cross-referencing → 30 seconds of agent generation plus attorney review. The attorney's role shifted from drafter to reviewer — faster, more consistent, lower risk. The confidentiality architecture unlocked a pattern that scaled to 4 more agents across 3 additional departments within 6 months.

ROI: ~190 attorney-hours/year at a ~$250/hr loaded attorney cost — roughly $47K recovered annually.

← Back to Index