Legal intake software that answers the routine questions and sends anything risky to a lawyer first
Routine questions answered in minutes; anything risky reaches a lawyer first.
See one case, screen by screen ↓Legal intake software that gives the business one front door to Legal. Eight agents read every request from email, Slack, Teams, forms, Salesforce and the ethics line, classify it, answer the routine ones from the legal playbook with the passage cited, and route the rest to the right lawyer or to outside counsel. Anything risky reaches a lawyer first.
How to choose AI solutions for back-office and enterprise work →
Updated
Business teams ask Legal hundreds of questions a week, through the shared legal mailbox, Slack DMs, Teams chats, intranet forms and a button in Salesforce. Most are routine: can I sign this NDA, what is my signing limit, can I accept these tickets, where is the template. Lawyers answer them by hand, one at a time, and because nothing is counted, nobody can say how many arrive, of what type, from which business unit or how long each one waits. Work goes to whoever answers first, so one lawyer is overloaded while another has room.
A few requests are dangerous, and they arrive in the same channels in the same casual words. A cease-and-desist letter forwarded as “can we just tweak the colour?” looks like a marketing question. A whistleblower's report in a shared mailbox breaks confidentiality. A data leak mentioned in a chat can start the 72-hour clock to notify the authority under GDPR Art. 33 without anyone seeing it, and an internal report under the EU Whistleblowing Directive must be acknowledged within 7 days. Outside counsel gets instructed by phone call, without a brief, a budget or a conflict check.
Estimated split of the about 45 minutes a routine answer takes a lawyer by hand (the demo company's own baseline), and the minutes left with the solution, where the agents read, check and answer and legal operations reviews a weekly sample.
Eight agents read, classify, screen, answer, check and route every request; counsel decides everything that is not routine.
Kestrel & Vane Group — about 40,000 people in 30 countries, one legal team of about 40. Since 07:00 on Wednesday 14 October, 38 requests have arrived from six channels. Here is how the people who ask, the agents and the lawyers work them, screen by screen, in the working solution.
Leo messages Legal in Slack: “Can I sign this NDA with Brightwell Analytics? It's the one they sent back from our template.” He attaches the 4-page Word file and fills in no form. At 09:32 the reply lands in the same thread: yes, it is the mutual NDA template v4.2 with all 14 clauses unchanged, and at grade M3 he may sign it — each sentence carries its playbook passage. Later he asks about supplier tickets and gets the gifts limit, cited.
“Yes, you can sign it. It is our mutual NDA template v4.2 and all 14 clauses are unchanged. As a sales manager at grade M3 you may sign mutual NDAs on our template.”
The intake reader takes Leo's grade and business unit from the HR directory. The classifier reads it as NDA on our paper at 0.97. The risk sentinel scans the message and all 1,912 words of the attachment against 7 tripwire families and screens Brightwell Analytics Ltd for sanctions: clear. The playbook answerer compares the document with template NDA-M v4.2 — 14 of 14 clauses match, no tracked changes, no hidden text, 2-year term, Illinois law. The answer checker confirms 3 of 3 sentences, and that grade M3 meets the M2 that SA-3 §2.1 requires.
An answer goes out only when the type is on the self-service list, no tripwire is found, confidence is at least 0.92 and the second agent agrees.
Nadia's front door shows the morning: 14 requests by email, 8 on Slack, 8 on Teams, 5 by form, 2 from Salesforce and 1 on the ethics line. One duplicate was merged and one IT question sent to the service desk. The other 36 sit in four lanes — 15 answered by self-service, 7 drafted for a lawyer to send, 12 routed to a lawyer (4 urgent) and 2 sent to outside counsel — each with the type it was read as, its confidence and the passage it used.
“Every legal request, read, sorted, answered or assigned.”
Brand manager Jonas Lind asks whether marketing can “just tweak the colour on the next print run” and attaches a 3-page letter from Vellum Brands' lawyers. The classifier's first read is Marketing · packaging change at 0.61. The risk sentinel finds tripwire RT-1: “cease and desist” and “trade dress infringement” on page 1, the Lanham Act §43(a) on page 2, a reply demanded by 28 October on page 3. It pulls the request out of self-service, marks it urgent — counsel within the hour — and at 09:35 Jonas gets the approved holding reply.
Holding reply HR-2: “Until she does, please don't reply to the sender, don't change the packaging, and don't discuss the letter outside your team.”
The router scores the candidates on expertise, jurisdiction, load and continuity. Amara Osei — Disputes & IP, US, 74% loaded, no conflict — scores 0.91. Sienna Clarke covers the UK and EU only (0.64), Daniel Kwan is at 118% (0.38) and Rahul Mehta is out until Monday 19 October. When Jonas forwards the same letter by email at 09:36, the intake reader merges it into the same request.
Why this lawyer: “Disputes & IP · US · 74% · no conflict · 0.91.”
The risk sentinel has already said the duty to preserve has probably begun. Amara accepts, opens the recommended notice HN-2026-0431, checks the custodians — Jonas Lind, Ana Ruiz from packaging design and the Brand/AeroClean shared folder — and suspends deletion on mail and Teams for the custodians and on the print-vendor proof portal. One click, and it is logged.
The risk sentinel never decides the legal question; counsel does.
Amara asks for outside help. The referral agent drafts the brief: the facts from the letter, what we need, the 28 October deadline, the documents including the preservation notice, and a conflict check that is clear on our list. It proposes Harwood Lyle LLP from the panel — partner $1,150 an hour, 12 of our matters, 92% on budget — with a phase-1 cap of $40,000. Over $25,000, so it waits for Deputy GC Elena Petrova to approve with a reason code.
The referral agent cannot instruct a firm; the Deputy GC approves anything over $25,000.
The urgent view holds what the sentinel escalated. A safety inspector is on site at the Dayton plant. A protected ethics report is visible to Ethics & Compliance only, with acknowledgement due Wed 21 Oct (7 days) and feedback by Tue 12 Jan 2027. A field app showed one customer's technician data to another: notify the authority by Sat 17 Oct 09:12. A distributor quote for re-export to Iran is blocked until cleared. None was answered by self-service.
“Details and reporter visible to Ethics & Compliance only.”
The demand analyst drew 25 of the week's 388 self-service answers at random. Nadia disagrees with item 19: football tickets from a supplier at R$650 each were allowed under the global €150 limit, but the Brazil addendum to the gifts policy sets R$500 per person per event. The analyst proposes pausing gifts self-service for Brazil and drafts a fix to GH-2 for its owner, Sofia Marin. She signs off the week.
Proposed by the demand analyst: “pause GH-2 self-service for Brazil.”
In Q3, the first quarter since go-live: 4,212 requests, 38% answered by self-service in a median of 3 minutes — it was 1.8 days — and a median of 4 minutes to assign the rest. 96.4% inside SLA, and 167 of 167 high-risk requests reached a lawyer first. Demand is broken down by type, business unit and source, with the estimated lawyer time released shown with its working.
The same front door, seen by five of the people who use it — what their week looked like, and what it looks like now.
Eight agents read, classify, screen, answer, check and route every request; counsel decides everything that is not routine.
Reads every request in every channel, with attachments and forwarded threads; finds the requester in the HR directory, the business unit, counterparty, ask and deadline; merges duplicates and redirects requests that are not legal.
Reads each request as one of 24 request types, with its urgency and a confidence score.
Scans every message and attachment for tripwires — threat of litigation, regulator or inspector contact, ethics reports, data incidents, sanctions, competitor contact, court process — pulls the request out of self-service, sets its urgency, starts the clock it implies, takes the protected route and recommends preservation.
Answers the types allowed for self-service from the legal playbook and drafts the rest for a lawyer, citing the passage it used and comparing attachments with the template.
Verifies every sentence of an answer against its passage and the requester's own facts: the attachment clause by clause, their grade and signature authority, the gift value against the limit for their country.
Assigns each request by expertise, jurisdiction, load, absence and conflicts, shows why this lawyer, and proposes a rebalance when someone passes 100%.
Writes the outside-counsel brief, proposes a panel firm with its rates and record, asks for the conflict check and drafts the budget and the instruction.
Counts and reports demand, draws the weekly random sample, and clusters the questions no playbook entry answers, drafting new entries from lawyers' past answers.
Sends advice; legal ops sets the rules. The agents propose; a named person decides.
Legal operations and counsel can ask in plain words. Answers come from the records, cite them, and open the screen behind them.
Why did the Vellum Brands letter skip self-service?
Because the risk sentinel found tripwire RT-1, threat of litigation. The letter says “cease and desist” and “trade dress infringement” on page 1 and cites the Lanham Act §43(a) on page 2. A tripwire is never answered by self-service, whatever the classifier thinks; its first read was Marketing at 0.61. It went to Amara Osei, urgent, due 10:34.
Which request types could we safely add to self-service?
One looks ready: customer logos in case studies. It was asked 23 times in Q3, and lawyers gave the same answer 21 times; Hannah Weiss would own it. Two others are not legal questions: certificates of insurance (31 asks) should go to Risk & Insurance, and export classification (12) to trade compliance. Adding a type needs the owner and Elena Petrova.
Who has capacity for a sales contract review this week?
Grace Okafor is at 52% and can do first passes on playbook fallbacks, with Daniel Kwan sending. Chloe Brandt is at 71% and is cross-trained on sales terms. Daniel himself is at 118%. The router has proposed a rebalance that brings him to 96%.
Draft the Q3 demand summary for the GC
Q3: 4,212 requests; 38% answered by self-service in a median of 3 minutes; 167 high-risk requests, all with a lawyer first; NDAs (812) and contract reviews (1,059) were the biggest load. The full summary is open for you to send.
13 screens from the working solution, on its sample data. Pick one to see it large.
Legal operations sets the front door's rules. Every change is logged, and changes to self-service need the Deputy GC's co-sign.
| Setting | Default | Choose from |
|---|---|---|
| Minimum confidence for self-service | 0.92 | Below this, a request is never answered without a lawyer |
| Second agent must agree | Always on | Locked: the answer checker verifies every sentence |
| Contract value ceiling | US$250,000 | Contracts above this always go to a lawyer |
| Gift limits by country | €150 per person per event; Brazil R$500 | From the gifts policy and its country addenda |
| Any tripwire | Always on | Locked: 7 tripwire families, phrases you can add, read in every message and attachment |
| Routing weights | Expertise 40 · load 25 · jurisdiction 20 · continuity 15 | Backups named per lawyer; protected routes fixed |
| Response targets | Urgent 1 hour · high 4 hours · standard 1 business day | Contract review 3 business days; set by legal operations, logged |
| Types and countries allowed | 13 playbook entries, up to 18 countries | Switched per entry and per country in the playbook, with co-sign |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample data · “target” = the design goal, measured in the live solution · “estimated” = our estimate, from the demo company's own baseline of about 45 minutes a routine answer · “typical” = published figures (CLOC 2025 State of the Industry survey via Above the Law — 83% expect demand to rise, 34% plan to automate routine tasks, 33% plan legal intake technology; EU Directive 2019/1937 Art. 9 on the 7-day acknowledgement; GDPR Art. 33 on 72-hour breach notification). People, companies, letters and requests named on this page are fictional — characters and sample data in the working solution.
Legal intake software that gives the business one front door to Legal. Eight agents read every request from email, Slack, Teams, forms, Salesforce and the ethics line, classify it, answer the routine ones from the legal playbook with the passage cited, and route the rest to the right lawyer or to outside counsel. Anything risky reaches a lawyer first.
People ask in their own words, in the channel they already use, and agents read the message and its attachments. A second agent checks every self-service answer against the passage and the requester's facts before it goes out, a risk agent can pull any request out of self-service, and the router explains every assignment.
The risk sentinel reads every message and attachment for tripwires such as a threat of litigation, an inspector on site, an ethics report, a data incident, sanctions, competitor contact or court process. A hit pulls the request out of self-service, marks its urgency, starts the clock it implies and sends it to counsel. The tripwire rule cannot be switched off for any type.
Only the request types legal operations puts on the self-service list, per country — in the demo, things like NDAs on our paper, signing limits, gifts and hospitality, templates, entity information and policy questions. Adding a type needs its owning lawyer and the Deputy GC, and a weekly random sample decides what stays.
They take a protected route to Ethics & Compliance only, and the reporter's identity is hidden from everyone else, including legal operations. The acknowledgement and feedback clocks are shown on the request — 7 days and 3 months in the demo, following the EU Whistleblowing Directive.
Yes. The router ranks lawyers by expertise, jurisdiction, load and continuity, skips anyone who is out, and shows why it chose each one. When a lawyer passes 100% of capacity it proposes a rebalance that respects authority, which legal operations applies.
Yes. Only counsel sends advice or accepts routed work, the Deputy GC approves referrals over $25,000 and every change to self-service, and the Head of Legal Operations signs off the weekly sample. The agents never instruct a firm, answer anything a tripwire touched or change a rule.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your playbook, templates, signature authority and gift limits, and a sample of past requests — and it goes live once every quality gate has passed. We will walk you through it on your own requests first.
Every deadline met, and every page traced to the request it answers.
1-min demoPublic Records Request ManagementEvery withheld line cites its law, and every release goes out before the clock runs.
1-min demoLegal Hold ManagementDisposal stopped before it happens, and a Rule 37(e) record verified every day.
1-min demoWe'll run Legal Intake & Triage on a sample of your own legal mailbox and your playbook.