Why this memo.
iManage Insight+ is a real product and the team behind it ships value at AmLaw 100 firms. Mid-market firms with 20 to 150 attorneys are usually not the right ICP for it. The 47-attorney litigation firm with iManage Work, no Innovation Partner, and a $1.4M annual leakage problem needs custom AI built on top of iManage, not the AmLaw-segment product.
This memo is what we'd commission for that firm. Below: the iManage surface we touch, the three workflows we ship in the first 6 weeks, and how the permissions and ethical-wall architecture stays intact.
The iManage surface area we touch.
iManage Work exposes the iManage REST APIs, which give authenticated read/write access to documents, folders, matters, workspaces, and metadata. Auth is OAuth2 with the firm's identity provider. Permissions and ethical walls are enforced at the iManage server level; the AI layer queries with the user's actual permissions, not a service-account super-user.
This matters more in legal than in any other vertical we work in. A retrieval system that bypasses ethical walls is a malpractice incident waiting to happen. Our query architecture preserves them.
Workflow I: Matter-aware retrieval and deal-precedent search.
The expensive workflow. Year-one associate starts a research task; spends three hours combing the matter archive for prior memos on the same issue, the same judge, the same opposing counsel. Often misses the most relevant precedent because the archive's metadata is inconsistent.
The custom-AI version: associate types the research question in plain English. The retrieval layer queries iManage with the associate's actual permissions, reads matter metadata + document content, returns the top 8-12 most relevant prior matters with the partner who handled each, the outcome, and the specific paragraphs that match. Associate cites and adapts; partner reviews; firm bills full hours instead of writing off ramp time.
Recovery: 6-12 months of effective ramp time per associate at audited firms.
Workflow II: Billable-hour reconstruction on Friday afternoons.
The leakage workflow. Partner writes time descriptions Friday afternoon for work done Monday-Wednesday. Reconstructions are vague, conservative, or skipped. Captured-but-uninvoiced is half the problem; unwritten is the other half.
The custom-AI version: reads the partner's iManage activity (documents opened, edited, emailed), Outlook calendar, Microsoft Teams call history, and matter context; drafts time entries with descriptions matter-mapped and ready for partner edit. Partner reviews and signs in 5 minutes per day instead of 90 minutes per week.
Recovery: $400K-$2M annual unbilled-time recovery at audited firms in this size band.
Workflow III: Knowledge ingestion with permissions intact.
The retiring partner workflow. Twenty years of memos, briefs, deal precedents in the partner's iManage personal folders. When the partner leaves, the firm loses the institutional value unless someone ingests it before the door closes.
The custom-AI version: bulk-ingest into a permissions-preserving retrieval index, with the original iManage ACLs maintained. The next lateral hire has access to twenty years of the firm's intellectual capital, gated by the same ethical-wall logic that gated the original documents.
What we don't build.
We do not replace iManage Work. We do not bypass ethical walls. We do not build "Harvey for your firm" because Harvey already exists and serves a slightly different segment well. If the firm's needs are document-review-and-redlining at scale, Harvey or Spellbook is probably the right answer. If the firm's needs are firm-specific workflow automation on top of iManage, this memo describes what we ship.
Risk & governance.
Three things every legal AI commission must do, and that we do by default: (1) preserve client confidentiality through actual permissions enforcement, not policy memos; (2) provide an audit trail that survives subpoena; (3) avoid hallucination on dispositive citations through retrieval-grounded generation, not unconstrained generation. Architecture decisions follow from these constraints, not the other way around.