AI for law firms, shipped
— not pitched.

An `ai for law firms` engagement with us ships production AI workflows on your firm's stack — not slide decks, not pilots that die at month 4. The day-to-day: scope which workflow moves a P&L line (most often contract review, e-discovery re-ranking, matter intake + conflict-check automation, legal research assist, deposition summary, or KM Q&A), assign each workflow to a privilege ring (per §6), build the integration against your Relativity / Everlaw / iManage / NetDocuments / Clio / Ironclad stack, pick the right model per workflow (Sonnet 4.6 for clause reasoning + cite-grounded research, Haiku 4.5 for high-volume classification, GPT-5.4-mini for structured extraction), deploy in a privilege-aware topology (Anthropic Bedrock with customer-managed keys + retention=0 on rings 1 and 2, standard commercial tier on ring 3 with citation logging), ship behind a partner / attorney sign-off flag, then operate the workflow long enough to prove cost-of-ownership before scaling. We don't sell a product — Harvey, CoCounsel, Lexis+ AI, Westlaw Precision, Clio Duo, and Ironclad are the benchmark products you should compare us against. We're the right answer when the workflow needs to be shaped to your matter intake, your KM corpus, and your privilege topology rather than a product team's roadmap.

Neither — we're a development partner, not a reseller, and we integrate with the legal AI products you already run rather than replace them. Harvey is strong for AmLaw 100 generalist legal work; CoCounsel (Thomson Reuters) is strong for research + drafting integrated with Westlaw Precision; Lexis+ AI is the LexisNexis-stack equivalent; Clio Duo is the practice-management-embedded option for solo + small firms; Ironclad is the leading CLM. Each is the right answer for some firms. We build when the firm needs a workflow that those products don't ship — your firm's own clause library as the playbook, your firm's KM corpus as the retrieval source, your firm's privilege-ring topology rather than a vendor's default, your firm's DMS (iManage / NetDocuments) as the source-of-truth, or a workflow surface (matter intake from a specific channel, a custom e-discovery re-rank against a specific Relativity field schema) the product roadmap doesn't cover. We'll say in the audit if a packaged product is the better answer for your firm; we've recommended Harvey + Ironclad to firms whose scope wasn't worth a custom build.

Privilege is a content classification, not a vendor checkbox — and we ship four deployment patterns mapped to four privilege rings (see §6). Ring 1 (attorney work-product · client communications): on-prem or zero-data-retention tenancy only, no third-party logs, citation chain auditable. Ring 2 (matter documents · privileged): vendor BYOK encryption with Anthropic Claude on AWS Bedrock + customer-managed KMS keys + retention=0, or Azure OpenAI with retention disabled. Ring 3 (public-record · case law · regulatory filings): standard commercial APIs OK on Anthropic / OpenAI tiers, with citation logging required per ABA Op. 512. Ring 4 (firm marketing · KM articles · public web): any model, any vendor, normal commercial terms. The FRE 502(b) inadvertent-waiver doctrine looks at whether the firm took reasonable steps to prevent disclosure — routing ring 1 or ring 2 content through a shared-fleet API with vendor-side prompt retention is not a defensible posture, and we won't ship a workflow with that topology. The honest constraint: privilege-aware deployment costs more per inference than a consumer-API workflow. We'll quantify it in the audit before any build commits.

Every workflow we ship sits on one side of the UPL line, and the rule we use to draw it is verbatim: AI does drafting, summarization, research, classification, and intake — never client-facing legal advice. ABA Formal Opinion 512 (July 2024) on the GenAI use of duties names supervision, competence, and confidentiality as the operative obligations under Model Rules 1.1, 1.6, and 5.3. The California State Bar Practical Guidance (Nov 2023) covers the same ground under California's professional-conduct rules, and the New York State Bar Task Force Report (Apr 2024) addresses the practitioner-facing duties under New York's rules. None of these soften the UPL statutes — they reinforce that AI-assisted work is the lawyer's work product, with the lawyer accountable for competence and supervision. In practice this means: a partner or responsible attorney reviews and signs every output that has client-facing consequence, the model never sends to a client or files autonomously, and the firm's supervision policy is documented in the workflow's audit log. We won't ship a workflow whose ROI depends on removing the attorney from the sign-off step.

Yes — all six, and the integration patterns differ enough to flag upfront. Relativity: kCura API + Relativity One stored fields are the workhorses; GenAI re-rank typically runs on the candidate set surfaced by Relativity's TAR engine, with the re-ranker's confidence score written back to a custom field. Everlaw: more open API surface but smaller schema flexibility — re-rank workflows ship faster, custom field schemas need more upfront mapping. iManage: cloud + on-prem variants matter; cloud (iManage Cloud) ships in 2–4 weeks for ring-2 workflows, on-prem (Work 10 Server) adds 2–4 weeks for VPN + storage-bucket setup. NetDocuments: similar shape, with the ndOffice add-in often the right surface for the attorney-facing UI. Clio: Clio Manage API + Clio Duo coexistence is the most common scope — we build adjacent to Clio Duo rather than replacing it. Ironclad: strong API for clause-library extraction; common scope is enriching Ironclad's CLM with your firm's own playbook + a custom redline model rather than replacing Ironclad. We do not certify against every vendor's full schema; we'll show you the integration scope in the audit before the pilot.

An `ai contract review` workflow runs in two cost buckets. Model layer on Sonnet 4.6 with prompt caching warm: roughly $0.12–$0.45 per NDA (3–8K tokens), $0.30–$1.20 per MSA (8–20K tokens), $0.40–$2.00 per SOW (10–25K tokens) depending on length + playbook size. Infrastructure layer (DMS integration, audit-log writes, privilege-ring routing, eval-suite hits): typically $0.08–$0.25 per contract. Total: roughly $0.20–$2.25 per contract depending on type and depth of clause-library comparison. For an in-house team reviewing 200 contracts a month (mix of NDAs + MSAs), model + infra typically sits in the $300–$900/month range; counsel time returned is the economic story, not the per-contract spend. Honest cost ceiling is engineering: getting the firm's house playbook loaded right, the redline format aligned to the firm's house style, and the counsel-facing UI to a place counsel actually trusts. That work is the engagement, not the per-contract token spend.

No — AI sits on top of TAR, not instead of it. TAR (Technical-Assisted Review, predictive coding) is well-suited to the bulk relevance pass on a high-volume responsive doc set — see §5 quadrant Q2 — and its $0.30–$1.20/doc baseline is hard to beat at scale. GenAI's job is the boundary band: the docs TAR ranks 0.55–0.75 where the classifier is uncertain, and where the nuance (sarcasm, code-named projects, context implied across an email thread) is exactly what a transformer model is better at than a logistic-regression classifier. We re-rank that band, add $0.05–$0.20/doc on top of the TAR baseline, and typically see 40–55% reviewer-time reduction on the boundary cases that drive most of the actual review spend. Privilege-log triage runs on the same set — flagging inconsistent privilege calls across reviewers, drafting log entries from the underlying content, surfacing rationale mismatches before production. Relativity and Everlaw are the platforms we ship into; Reveal is workable but less common in our engagements. The wrong answer is GenAI on the full corpus — see §5 quadrant Q3 (junk / dedupe / NIST), where pure ML wins on cost.

Three tiers, pricing-locked across the cluster. (1) Legal AI audit: $3K fixed, 1–2 weeks. We shadow partner / GC / e-discovery / compliance, score candidate workflows by attorney + paralegal hours returned × time-to-ship × privilege-ring exposure, deliver a 90-day roadmap with per-workflow cost bands + privilege-ring assignment + an honest "these won't pay back yet" list. (2) Pilot to production: $10–25K fixed, 4–8 weeks. One workflow shipped end-to-end on your stack, privilege-ring-aware, citation-chain-logged, partner-in-loop, with a walk-away point at week 6 — if the metric won't move, we stop before production hardening and you don't pay phase 2. (3) Continuous legal AI team: from $5K/month, no annual contract. Embedded PM + AI engineer + legal-ops analyst shipping the next workflow on your roadmap, with per-workflow monthly cost-of-ownership reporting and privilege-ring review before each new workflow ships. Most `legal ai company` engagements we run start with the audit, ship the first workflow on the pilot, then move to monthly for workflows two through five. Cost-per-decision reported monthly on every shipped workflow — no per-decision number, no engagement.