Opportunity space: where LawVu Assistant meets the inbox

A huge share of in-house legal work still arrives as email. Today's add-ins do the mechanical plumbing of capture well; every step of understanding and acting is still manual.

Reach · 1-10How many users it touches and how often. Daily-for-everyone scores high.
Impact · 1-10How much value it delivers when it fires: time saved, risk reduced, work unlocked.
When?Now = current Assistant abilities · Next = near-term build · Horizon = aspirational.
Filter

Spotlight 01 · Opportunity 1.3

Schema-aware intake auto-population

The problem today. Intake forms for matters and contracts are often long and demanding: matter type, counterparties, jurisdiction, value, risk level, business unit, key dates, plus a layer of org-specific custom fields and constrained option lists. Populating one well takes real time and real judgement, and it's the exact friction that makes people skip LawVu and leave work in their inbox. Worse, a half-completed form produces a weak record that undermines every downstream report and AI feature.

Why this is a step-change. An incoming email and its attachments already contain most of the answers - the parties, the ask, the deadline, the contract value, the business context - just in unstructured prose. The agentic move is to close the gap between that prose and the structured schema:

The mechanism

  • Pick the right type. Infer the correct matter/contract type from the email so the right form (and the right custom fields) is the one being filled.
  • Read the schema first. Pull the live creation schema for the relevant matter/contract type, so the Assistant knows exactly which fields exist, which are required, and which option-list values are valid - then map extracted data onto those fields rather than inventing its own.
  • Extract, map, and select. Populate free-text fields from the thread and attachments, and resolve constrained fields to the correct option-list value (e.g. "we're buying" becomes Procurement / Inbound).
  • Close the gaps intelligently. For any notable gaps, offer to draft a short reply to the requester instead of the lawyer chasing it manually.
AssistantDrafted a new Procurement matter from this email. 9 of 11 fields populated (2 inferred, marked for your review). Missing: budget owner. Draft a reply asking the requester?

The compounding payoff

Complete, well-structured records are what LawVu's reporting, search, and other AI features feed on, so this saves minutes at creation and raises data quality across the entire platform. It directly attacks the "leave it in the inbox" behaviour the add-ins exist to fix. Feasible now in its core form; custom-field and option-list mapping is where the depth and differentiation live.

Spotlight 02 · Opportunity 3.5, with 1.6 and 3.6

Contract version intelligence

A new version of a contract landing in the inbox mid-negotiation is one of the most common - and most cognitively expensive - moments in an in-house lawyer's day. The job-to-be-done is blunt: "What changed, and do I care?" Mechanical diffing already exists (Word compare, Litera/Workshare, LawVu's own version comparison). The agentic opportunity is everything above the diff - reading the substance and legal effect of the changes - anchored by the fact that LawVu already holds the prior version, so the comparison has a trustworthy baseline rather than whatever file the user happens to have open.

The flow: attachment-aware versioning (1.6) recognises the inbound file as a new version of an existing contract, compares it against the currently stored version, and offers to run a full change analysis - ready in the contract record and summarised back in the inbox panel.

What the AI layer adds on top of a raw diff

  • Semantic change summary. Translate mechanical edits into meaning and effect: "Liability narrowed to direct damages only; unlimited IP-infringement carve-out deleted; a 30-day cure period added to termination." One paragraph a lawyer can read in fifteen seconds.
  • Materiality triage. Separate the handful of changes that matter from the typos, renumbering, and formatting - and rank the substantive ones by risk.
  • Playbook / fallback check. Compare each change against the org's playbook and acceptable fallbacks - drawing on the clause libraries and playbooks already curated in LawVu Draft - and flag where the counterparty has pushed out of bounds: "Cap is now 6 months' fees - your floor is 12."
  • Sneaky-change detection. Surface language you previously struck that's been quietly reinstated, or your edits silently dropped - the classic risk in round-trip redlining (feeds the cross-round ledger, 3.6).
  • Obligation & key-date delta. Detect whether the edits moved payment terms, renewal or termination dates, or obligations, and update the contract's extracted metadata accordingly - so the structured record stays true to the latest version.
  • Risk re-scoring + re-approval trigger. Re-evaluate overall risk posture; if a change crosses a defined threshold, flag that re-approval is required and route it.
  • Two audiences, two summaries. A precise legal summary for the lawyer, and a plain-English commercial one for the business owner.
  • Drafted response. Offer a reply that accepts the benign changes and pushes back on the out-of-policy ones using playbook fallback language - turning analysis straight into action (links to 5.1).
  • Narrated version history. Every stored version gets an AI-generated summary, so the timeline reads as a clear narrative of the negotiation rather than a stack of undifferentiated files.
AssistantThis draft changes 3 substantive terms. Liability cap lowered to 6 months' fees, below your playbook floor of 12. A 30-day cure period was added. Auto-renewal removed.

Feasibility

The single-version "what changed and what it means" summary is Next and high-impact - it leans on abilities the Assistant already has (file analysis, extraction, summarisation) plus a reliable anchor version. The cross-round ledger, automated re-approval routing, and playbook-grounded counter-drafting are Horizon, maturing alongside trust and deeper Draft/Lens integration.

Quick wins · high reach, feasible now

Everyday-for-everyone capabilities that ride on current Assistant strengths. Do first.

  • Natural-language search from the inbox - retrieval the Assistant already does, surfaced where the user actually reads mail; an obvious, low-cost first step.
  • Smart matter/contract suggestion and schema-aware intake auto-population - make the existing core workflow intelligent.
  • Thread summarisation, open-email matter catch-up - the highest-demand LLM behaviours.
  • Task extraction, key-date capture, metadata enrichment - turn email prose into operational records: work that's visible, assignable, and reportable across the team rather than stuck in one inbox.

High-impact bets · build deliberately

Lower reach or longer horizon, but step-changes in value.

  • Substantive version-change analysis - "what changed and what it means" the moment a new contract version arrives, anchored to the version LawVu already holds.
  • Context-grounded reply drafting and knowledge-grounded answers - shifts the Assistant from organising work to doing it.
  • Shared-mailbox triage agent - unlocks an intake channel that's structurally unavailable today.

Aspirational frontier · sequence later

Depend on monitoring and autonomy maturing, and on trust built by the items above.

  • Confidence-gated auto-filing - high value, but gated to later: mis-filing risks exposing privileged or confidential content to the wrong matter audience, so it should only ship once suggestion accuracy (1.1) is proven in production and the consequences of a wrong file are contained.
  • Follow-up tracker, inbox digest, deadline guardian.

A natural narrative arc

Capture it (1)+ Search it (2) Understand it (3) Structure it (4) Respond to it (5) Let the agent run it (6)

Search sits early on purpose: the Assistant already has these retrieval abilities, so surfacing them in a mail-based chat is the most natural place to start.

Cross-cutting design principles

Keep the approval gate

The Assistant's existing model - propose, human approves, audit-log everything - maps perfectly to email, where mistakes (mis-filing, mis-sending) are costly.

Confidence-tiered autonomy

Auto-act on the obvious, surface the ambiguous, never silently guess on the sensitive.

Meet the user in the panel

The add-in already pins open beside the inbox; the Assistant should live there as a conversational surface, not a separate destination.

Grounded, cited, in-region

Inbox answers should inherit LawVu's existing posture - grounded in the customer's own data, cited where it matters, data staying in-region, prompts not used for training.

Open questions

Should Custom Skills (admin-defined Assistant instructions) be exposed in the add-in so each org can tune inbox triage to its own matter types and naming conventions?
How do we treat non-legal email noise - does the Assistant stay silent unless an email looks legal, or offer a lightweight "is this legal work?" affordance?