- Client submits on the firm’s site
- Validated on both sides
- Conflict check starts at submit
Form in. Approve. The matter exists.
Built around the firm’s workflow and wired into Clio, Dropbox, and Microsoft 365 — not a platform the firm migrates onto. One retained matter that would have gone to whoever called back first is the payback period.
ClioContacts, matters, files
DropboxClosed files the PMS never saw
Prior prospectsER 1.18 duties, nowhere else
- Clio contact
- Clio matter
- Flat-fee activity
First contact converts
If a flat fee is $X and this converts one extra matter a month, annual return is 12X. One retained file that would have gone to whoever called back first is the payback period.
Non-billable work does not come back
Conflict search, summary, SOL research, Clio setup, filing each email by hand. The automated version is a click. A hundred intakes at 45 minutes each is roughly two working weeks.
The misses that dwarf the savings
A missed SOL is a malpractice claim. A conflict after engagement is disqualification. Silent leakage is a client who phones because the email never arrived.
Marginal cost is effectively zero
Per-intake AI is a fraction of a cent. No per-seat, no per-matter, no usage cliff. Hiring the equivalent capacity is a paralegal salary.
Reference deployment. The firm had hand-filed ~11,200 emails and ~13,000 attachments into a single catch-all matter, one at a time, from Outlook. In a 13-day window, 10 client emails from 4 matters had been lost entirely — discovered only because one client phoned. That is now a routed, held, and surfaced queue.
Intake
A public form on the site the firm already has.
Multi-step, embeddable, rate-limited. The client never hits a mystery rejection — the browser enforces the same caps as the server.
- Multi-step, embeddable on the firm’s existing site
- Client-side limits match the server — no mystery rejections
- Rate-limited, with field errors routed back to the right step
Conflicts
Three sources. Automatic. At submit.
Practice management, the legacy archive, and prior prospective clients — the source nobody else checks. Verdicts stay independent. Missing data is never called clear.
- Clio — contacts, matters, custom fields, document filenames, then AI analysis
- Dropbox — closed files that never made it into the PMS, with fuzzy folder matching
- Prior prospective clients — people who called, were declined, or never engaged. ER 1.18 still applies
- Each source keeps its own verdict. Absence of data is never reported as clear
Confidentiality
Raw client data never leaves the firm.
Privileged notes, contact emails, descriptions — none of it is sent to a model. The conflict engine sees numbered slots. Detail is reattached locally. That is the slide that closes risk-averse partners.
- Privileged notes, emails, and descriptions never leave the firm to a model
- The engine sends numbered slots and derived flags; it gets indices back
- Detail is reattached locally, inside the firm’s systems
The rest of the pipeline
Everything between submit and the Clio activity.
SOL calculated at intake
The model reports facts. Code owns the law — including the public-entity notice trap that shortens the real deadline to 180 days. An attorney acknowledgment clears the countdown and is stored as a bar-defense row.
Triage before a minute is spent
Matter type, urgency, case-value band, and an out-of-scope flag — before an attorney opens the submission.
Evidence captured before it disappears
Posts, listings, and pages are hashed (SHA-256) and stored privately at intake. Not two weeks later. Nothing is publicly archived.
One-click matter in Clio
On approval: contact, matter, and flat-fee activity. The attorney bills from Clio as they always have. Webhooks then drive bill created, sent, and paid — no spreadsheet.
Built in
Correspondence, retrieval, and a place failures go.
Letters that never send themselves
Follow-ups, DMCAs, C&Ds, demands, litigation holds — drafted, queued, sent only after approval. The exact bytes that went out are persisted, including attorney edits.
No hallucinated citations
The model cannot write a citation. It writes fact-specific prose and picks from a vetted library. Every cite is re-verified against CourtListener.
Mail from the firm’s Outlook
Outbound goes from the firm’s M365 mailbox, threads in Sent Items, and routes replies to the right matter. Unmatched mail is held — never dropped.
A review queue, not a inbox pile
Staff star what needs a partner. Founders green-light. The feed shows who flagged what, when. Read state is per person.
A failure surface
Failed submissions, agent errors, undelivered email, held inbound — one page, a badge count, a resolve action. Most firms find out by phone, weeks later.
Read-only assistant and a document library
Ask the firm’s own intakes and files. The model never emits a query the system didn’t define. Retrieval uses page spans, never fabricated page numbers. Visual workflows on intake, approval, decline, and billing events.
One deployment, one firm
Request a custom quoteThe firm’s data. The firm’s keys. The tools it already pays for.
© 2026 Lapt. All rights reserved.
Lapt