Use caseWorkflow automationAI agent
Law firm intake that runs the conflict check before anyone calls the prospect back
Turns web, email and phone inquiries into structured intake, runs conflict searches across parties and their groups, and leaves acceptance to a lawyer.
A blueprint, not a client story. The business described is illustrative; the architecture, integrations and trade-offs are real, and this is how I would build it. By Ergini, .
The short version
An intake workflow for law firms that turns web forms, chats, emails and voicemails into structured intake: matter type, parties, jurisdiction and urgent deadlines. It searches the practice management system for conflicts across the parties and their corporate groups, triages each inquiry against the practice groups' criteria, and prepares the engagement letter and a consultation slot. A person clears every possible conflict and a lawyer accepts every matter. The assistant never gives legal advice and says it is an AI.
- Best for
- Commercial and employment firms of 10 to 80 lawyers that receive inquiries by web, email and phone, often outside office hours.
- Connects to
- Practice management system (RA-MICRO, Clio, LEAP or Actionstep), Outlook and Microsoft 365, Website form and chat, Calendar, iManage, Company data (commercial register, North Data)
- The AI does
- Pulls parties, dates and the matter type out of free-text inquiries in German or English, and drafts the intake summary, acknowledgment and engagement letter.
- People do
- The conflicts person clears every possible hit, a lawyer accepts or declines each matter and sets scope and fees, and nobody discusses the case before conflicts clear.
- Built as
- AI Workflow Automation, usually $15K - $30K
Good inquiries at night, the wrong ones by day
Consider a 30-lawyer commercial and employment firm in Frankfurt, acting mostly for employers and senior executives, in German and English. Inquiries come through the website form, the general inbox and the phone, where evening callers leave voicemails in a shared inbox nobody owns. By day, two intake assistants and the partners' secretaries work through it all.
Inquiries sent after six wait until morning, and by then some of the best have called another firm. Many of the rest are matters the firm does not take: a tenancy dispute, an employee suing a company the firm acts for, a personal injury case. And every booking that goes ahead takes the same half-dozen manual steps: assess the matter, check a diary, open the matter, file documents, take the consultation fee, send the receipt.
The riskier problem is the order of events. Callbacks often happen before anyone has checked who the other side is, so a lawyer can end up holding confidential information from someone the firm cannot act for, and in the worst case the firm is conflicted out of work for an existing client. The Federal Lawyers' Act bars acting for conflicting interests in the same matter (section 43a(4) BRAO), and London and Amsterdam have equivalent rules. Intake has to clear conflicts first and talk second.
An inquiry at 22:40, and what happened behind it
The prospect sees three messages. Behind them: extracted facts, a deadline rule, and a conflict search that finds something only a person can judge.
Website chat, Thursday 17 September, 22:40
Assistant · Website chat
Good evening. I'm the firm's AI intake assistant, not a lawyer, so I can't give legal advice. I'll take a few details so the right lawyer can get back to you; please don't send documents or confidential details yet. Who are the parties involved?
Prospect · Website chat
I'm the HR director of a logistics company in Offenbach, part of a group based in Hamburg. On 9 September we found out that our warehouse manager has been selling stock. We want to dismiss him without notice.
- extract_intake(transcript)employment, employer side, dismissal for cause / parties: prospect's company (Offenbach), parent group (Hamburg), employee / facts known 9 Sep 2026 / English
- deadline_rules(type: "dismissal_for_cause", known_since: "2026-09-09")section 626(2) BGB: two weeks from knowledge / notice must reach the employee by 23 Sep 2026 / works council heard first, if there is one / urgency: high
- alert(duty_lawyer: "employment", urgency: "high")queued for 07:00 Friday / conflict check pending, no contact before clearance
- company_group(name: "prospect's company", source: "company data")parent: a holding SE in Hamburg / 4 sister companies / 2 managing directors
- conflict_search(names: 9, match: "fuzzy")prospect's company: no match / parent holding: possible match 0.91, adverse party in open matter 2025-0412 (commercial dispute) / employee, sister companies, directors: no match
Assistant · Website chat
Thank you. This sounds time-sensitive, so I've marked it urgent. A lawyer's office will contact you tomorrow morning after a short internal check. Please keep your documents until then.
- 07:30 Friday: the conflicts assistant opens the possible match first. The firm acts against the Hamburg holding in an unrelated dispute, so should it act for a subsidiary in an employment matter? Section 43a(4) BRAO covers the same matter only; the relationship with the existing client is a partner's call, not the model's.
Conflicts first, conversation second
The model handles language: extracting, summarizing, drafting. Deadlines come from rule tables, the conflict search is code, and every decision that commits the firm is a person's.
01 Trigger · Microsoft Graph, form webhook
An inquiry arrives by form, chat, email or voicemail
Intake emails arrive through Microsoft Graph, voicemails as transcripts, and the website form and chat post directly, all into one queue with source and time.
02 Plain code
Filter spam and known scams
Spam first. Then rules catch the scams aimed at law firms: a new client keen to pay money in before anything is agreed, an overpayment to refund, pressure to move funds through the client account. Those go to the compliance partner, not intake.
03 AI model · Structured output
Extract the intake record
A model fills a fixed schema: matter type, which side, parties and roles, jurisdiction, key dates, what the prospect wants, language. It records what was said, marks inferences as inferences, and never invents a party.
04 Plain code
Flag deadlines from the rule table
Key dates run through the practice groups' table: three weeks to challenge a dismissal after the notice arrives (section 4 KSchG), two weeks to dismiss for cause after learning the facts (section 626(2) BGB), and so on. A hit marks the inquiry urgent and alerts the duty lawyer; it is never shown to the prospect as advice.
05 System · Practice management API, company data
Run the conflict search
Every party, plus parents, subsidiaries and managing directors from company data, is searched against clients, contacts, adverse parties and closed matters. Matching is deliberately fuzzy: legal forms stripped (GmbH, AG, SE, B.V., Ltd), umlauts and spellings normalized, former names included.
06 Decision
What did the search find?
Thresholds set with the firm's risk partner; the model plays no part.
- No match above the review threshold then listed as clear on the conflicts person's daily sign-off
- A possible match on a party or a related company then held for the conflicts person, matched records side by side
- A match to an adverse party in an open matter then held, and the responsible partner told before any contact
07 Plain code
Triage against practice-group rules
Each group's rules decide fit: which side the firm acts for, minimum values, jurisdictions. A misfit gets a polite referral drafted, for example to the bar association's lawyer search. The model's only input is the matter type it extracted.
08 Person
A lawyer accepts or declines
The lawyer sees one page: summary, deadline flag, conflict result and who cleared it, triage suggestion. Accepting books a consultation from their calendar and prepares the engagement letter; declining sends the referral they approved. Only a lawyer tells a prospect the firm will act.
09 Result · Practice management API, iManage
The matter is opened and filed
The client and matter are created in the practice management system with the conflict record attached, the draft engagement letter goes to the iManage workspace for the lawyer to finish, and identity checks start where the matter type requires them.
Triage rules the practice groups own
The model classifies the matter type and pulls out dates. Everything below is written, versioned and signed off by the lawyers, so any triage decision traces back to the rule that made it.
| Practice group | Accepts | Refers elsewhere | Urgency flags |
|---|---|---|---|
| Employment | Employers of any size; managing directors and senior executives | Employees below management level | Notice of dismissal received (three weeks to sue); misconduct discovered (two weeks to dismiss for cause) |
| Commercial contracts | Companies with drafting work or disputes above the group's minimum value | Consumer matters, tenancy, small debt collection | A contractual termination or notice date within 30 days |
| Corporate | Shareholder disputes, GmbH and AG governance, transactions | Company formations below the group's minimum fee | A shareholder meeting or filing date within 14 days |
The hard parts of legal intake
A conflict hidden in a group structure
The prospect names their company, but the firm's client is its parent, or a sister company, or the other side's managing director sits on its board. The search covers related companies from company data, fuzzy matching produces false positives on purpose, and a person clears every possible hit, with who cleared it and why kept on the matter.
Prospects who tell you everything
People send their whole story, and employment matters often include health data, a special category under GDPR. The form and chat ask only for parties and dates and say not to send details yet. Anything sent anyway sits in a restricted record only the conflicts person can open until the matter is accepted, and it is deleted on the retention date if the firm declines.
A deadline nobody noticed
In employment work the clock often starts before the prospect calls. The rule table flags dates in the inquiry, alerts the duty lawyer the same day and escalates if nobody picks it up by noon. Once a matter is open, its deadlines move into court deadline docketing.
The assistant drifting into advice
'Do I have a case?' is what every prospect asks. The assistant only collects facts, books and explains the process, and gives one fixed answer to anything asking what someone should do. That answer is tested against a growing set of real questions, including indirect ones about whether the firm would take the case or whether the other side acted lawfully.
Spam, fake clients and the client account
Forms attract spam and sales pitches, which filters handle. The more serious pattern is the fake client who pays money in early, then asks for a refund to a different account. Those rules route to the compliance partner, and the assistant never discusses payments or bank details.
Confidentiality and the model provider
German lawyers may involve service providers only under section 43e BRAO, which requires a contract in text form binding the provider to confidentiality. The build uses an EU-processing model provider under such a contract, keeps intake data out of training, logs every AI action per matter, and can run on a self-hosted model in Europe instead.
Lawyer, conflicts person, code and model
The model prepares, code checks, people decide, as in my human-in-the-loop patterns.
The AI model
Extracting parties, dates and the matter type
Inquiries are free text in German and English, and turning them into fields is language work.
Drafting the engagement letter and the acknowledgment
Filled from the intake record and the firm's templates, then finished by the lawyer.
The out-of-hours conversation with the prospect
Collects facts and offers a callback time, says it is an AI, and gives no advice.
Plain code
Pulling parents, subsidiaries and directors from company data
A lookup, logged, so the search can be repeated exactly.
Deadline flags
Dates are arithmetic against a table the lawyers maintain, not interpretation.
Fuzzy conflict search
Repeatable, with thresholds the risk partner set and can audit.
A person
Clearing possible conflicts
Weighing a group relationship against the firm's duties is professional judgment.
Accepting or declining the matter, scope and fees
Taking on a client commits the firm, professionally and commercially.
What the assistant may do alone, and what it never will
Only the out-of-hours conversation ever runs without a person watching, and it earns that in stages.
| Stage | What runs without a person | Earned by |
|---|---|---|
| Shadow | Nothing. Intake records and conflict reports run beside the current process for staff to compare. | Parties and dates match staff notes across a few weeks of real inquiries. |
| Draft | Still nothing. Staff send the drafted acknowledgments and referrals, and the conflict report replaces manual searching. | Most drafts go out unchanged, and the conflicts person confirms the search finds everything their manual search found. |
| Out of hours | The assistant replies by itself outside office hours: acknowledgment, intake questions, a callback time. | A month of reviewed transcripts with no advice given and every urgent flag caught. |
| Never | Clearing conflicts, accepting matters, scope and fees, and anything said about the merits of a case. | Stays with people permanently. |
Clio Grow, Lawmatics, Intaker or a build?
If you are a US or UK firm on Clio, start with Clio Grow, Clio's intake and client-relationship product, connected to the matters in Clio Manage. Lawmatics covers similar ground with more marketing automation; Intaker focuses on website chat and intake forms; Eve's AI intake is built for US plaintiff firms, with integrations such as Clio Grow and Litify. For one practice area and a simple conflict check, one of these is probably enough, and cheaper than a build.
A build wins where they stop. Their integrations center on US and UK practice management systems, so a German firm on RA-MICRO or Advoware finds little to connect to. Firms with several practice groups, each with its own rules and deadline table, outgrow a single intake form quickly. And ask each vendor whether its conflict search covers related companies and records who cleared each hit: that is where I most often see a build justified.
Often the answer is a hybrid: keep Clio Grow or your practice management system as the home of intake, and add what is specific to your firm around it: the bilingual assistant, the group-aware conflict search and the deadline rules. Inquiries that are really sales leads, such as a company asking for a retainer proposal, can follow the route of inbound lead qualification.
How you would know it is working
A blueprint has no results to report, so here is what I would measure from the first week instead, on your own data.
- Time to first human contact
- From inquiry to a person's reply or callback, split by office hours and out of hours, and by urgent and normal inquiries.
- Missed conflicts in shadow mode
- Possible conflicts the manual search found and the automated search did not. It has to be zero before manual searching stops.
- Urgent flags caught
- Inquiries a lawyer later confirmed had a deadline, and whether the flag fired, reviewed monthly with each practice group.
- Lawyer time on declined matters
- Calls and consultations spent on inquiries the triage rules should have referred elsewhere. It shows where the rules need work.
- Inquiry-to-matter conversion by channel
- Which channels and practice groups turn into accepted matters, now that evening inquiries get an answer.
What a build like this costs
This is built as AI Workflow Automation, which runs $3.5K - $60K overall. A build like this one usually lands in the multi-step workflow tier: $15K - $30K, 3-5 weeks. The first working version runs on your real data well before the end of that window.
What it costs to run
Model costs are cents per inquiry, because each one is extracted once and the drafts are short. Company data providers charge per lookup or by subscription, and voicemail transcription is billed by the minute. Hosting in the EU, or a self-hosted model where the firm wants one, is the larger fixed line.
What moves the price
- The practice management system: Clio's API is documented, while desktop systems such as RA-MICRO usually need a connector on the firm's server or an interface agreed with the vendor
- How many practice groups, each with its own triage rules and deadline table
- Conflict search depth: names only, or group structures from company data with a sign-off trail
- Channels and languages: form and email only, or also chat and transcribed voicemail, in German and English
Who this is for
- Commercial and employment firms of 10 to 80 lawyers whose inquiries arrive by web, email and phone, often after hours
- German firms on RA-MICRO or similar systems that US intake tools do not connect to
- Firms whose conflict check is a name search by a secretary, with no record of related companies or of who cleared a hit
- Employment practices where intake has to spot a two-week or three-week deadline on the day the inquiry arrives
Questions people ask about this
How do I automate my law firm's intake process?
Automate the collecting and checking, not the deciding. Capture every inquiry into one queue, extract parties, dates and the matter type, run the conflict search across the parties and their corporate groups, and flag deadlines from a rule table. The conflicts person and the accepting lawyer each get one page; booking, the engagement letter and opening the matter follow their decision.
Can AI run conflict checks for a law firm?
AI can prepare a conflict check, but a person should clear it. Code searches every party, plus parents, subsidiaries and managing directors, against clients, adverse parties and closed matters, with fuzzy matching for spellings and legal forms. The model only extracts names from the inquiry. Possible hits reach the conflicts person with the matched records side by side, and each clearance is logged.
Does a law firm's AI intake assistant have to say it is an AI?
Yes, in the EU. Since 2 August 2026, Article 50 of the AI Act has required that people know when they are dealing with an AI system, unless it is already obvious. This assistant opens every conversation by saying it is the firm's AI intake assistant, not a lawyer, and that it cannot give legal advice. My guide to Article 50 covers the rest.
Will this work with RA-MICRO, Clio or LEAP?
Yes, by different routes. Clio has a documented API with webhooks, so contacts, matters and the conflict search connect directly. LEAP and Actionstep have integration routes to check against your version and plan. RA-MICRO and other German desktop systems usually need a connector on the firm's own server or an interface agreed with the vendor, which I scope in the first week.
How do we get past confidentiality concerns about using AI for intake?
Control where the data goes and keep a record of what the AI did. Inquiries go to a model provider under a data processing agreement with EU processing and no training on your data, plus, in Germany, the confidentiality contract section 43e BRAO requires. Details beyond parties and dates stay restricted until conflicts clear, and a per-matter log shows every AI action.
What does law firm intake automation cost?
Intake like this usually prices in the multi-step tier of my workflow automation range, because the effort is in the conflict search, the practice-group rules and the practice management integration, not the model. A desktop system without an API, several practice groups or group-structure searches push it toward the top of that tier. Running costs are mostly hosting and company data lookups.
Sources
- BRAO section 43a (conflicting interests)
- BRAO section 43e (use of service providers)
- KSchG section 4 (three weeks to challenge a dismissal)
- BGB section 626 (two weeks to dismiss for cause)
- EU AI Act implementation timeline (Article 50)
- Eve launches an AI intake specialist for plaintiffs' law firms (LawNext)
- Eve: case intake and evaluation