What this automation does
Every contract request enters the same way, lands with the right person and is visible from request to signed copy.
- One front door. Requesters use a short form, a Slack shortcut, or email to a contracts@ address with the document attached.
- Classify and extract. A model reads the request and document, identifies the type (NDA, MSA, SOW, order form, DPA, vendor agreement, amendment), whether it is your paper or theirs, and extracts counterparty, value, term, renewal terms and whether personal data is involved.
- Fill the gaps. If required fields are missing, the workflow asks the requester specific follow-up questions instead of a lawyer chasing them.
- Enrich. It looks up the counterparty in Salesforce or HubSpot for sales deals, or in the vendor list for purchases, and attaches the related opportunity or vendor record.
- Route by matrix. A written routing table assigns the request. For example, an NDA on your template goes to self-service generation, a sales contract over a value threshold goes to a named lawyer, and anything involving personal data also goes to security or privacy.
- Track. A ticket is created in Jira or your legal tracker with a turnaround target based on type, and the requester gets status updates in Slack.
- Sign and file. On approval, a DocuSign envelope goes out with signers in order. The executed copy is filed in SharePoint with a consistent name, and key dates go into the contract register.
When it is worth building
The signs are familiar: requests arrive by email, chat, hallway and calendar invite; nobody can say how many contracts are in flight; sales asks “where is my contract?” daily; and an auto-renewal slipped by last year because no one tracked the notice date. At around 20 requests a month, a single legal person loses meaningful time just triaging.
If you get a few requests a month, a shared inbox and a spreadsheet are enough. If you already have a CLM with an intake module, configure it rather than building beside it.
Architecture
| Layer | What we use | Why |
|---|---|---|
| Front door | Jotform or a Slack shortcut, plus a contracts@ mailbox | Requesters choose the channel; every path ends in the same queue |
| Orchestration | n8n | Built-in Slack, Salesforce, HubSpot and Jira nodes, DocuSign’s eSignature REST API through the HTTP Request node, and long-running waits for signature |
| Classification and extraction | Claude or OpenAI’s GPT models with a fixed JSON schema | Handles every format of contract and request without templates |
| Routing rules | A table owned by legal, evaluated in code in priority order | Lawyers can change routing without a developer, and every assignment is explainable |
| Tracker | Jira, or your matter system | Turnaround targets, ownership and history in one place |
| Signature | DocuSign | Envelope status updates close the loop automatically |
| Repository | SharePoint document library with metadata columns | Executed copies are findable by counterparty, type and dates |
How to build it
- List agreement types and owners. Agree with legal on the types you handle, who reviews each one, and the turnaround target for each.
- Write the routing matrix. Define rules by type, paper, value, data involvement and business unit, with an explicit priority so two matching rules never conflict.
- Design the intake form. Ask only what the model cannot reliably extract, such as business purpose, deadline and the internal owner.
- Build classification and extraction. Test against 50 past requests. Measure type accuracy and field accuracy separately.
- Add follow-up questions. Generate specific questions for missing fields and send them to the requester, pausing the request until answered.
- Connect tracking and signature. Create the ticket, post updates on status changes, and create DocuSign envelopes from approved final versions.
- File and register. Save executed copies with metadata and put confirmed notice dates on a reminder schedule.
The downloadable starter template below covers the Slack and email intake, model classification and extraction, the routing table lookup and Jira ticket creation.
What breaks in production
- Requests without the document. “Can legal look at the Acme deal?” arrives with nothing attached. The workflow should ask for the draft before creating a review ticket.
- Wrong version attached. Requesters attach an older draft from their downloads folder. Ask them to confirm it is the latest version from the counterparty, and record the file hash.
- Duplicate requests. Two people request the same vendor contract. Match on counterparty and type within an open window and merge them.
- Ambiguous routing. A request matches several rules. Without priority order, it lands with the wrong person or with two people at once.
- Notice period misreads. “60 days prior to the end of the then-current term” is easy to calculate wrongly after a renewal. A person confirms extracted dates.
- Signature status lag. DocuSign status events arrive late or out of order. Poll as a backstop before telling someone their contract is stuck.
Timeline and cost
Most legal teams are live in two to four weeks. Lumeor builds this in the range of $5,000 to $11,000, depending on the number of intake channels, CRM and tracker integrations, and whether signature and filing are included. Add AI first-pass review later as a separate project; see how to review contracts with AI. For the simplest agreements, automated NDA generation can remove them from the queue entirely.
How to measure success
- Median time from request to assigned reviewer
- Share of requests complete on first submission, with no follow-up needed
- Cycle time from request to signature, by agreement type
- Turnaround target breaches per month
- Missed renewal or notice deadlines, which should be zero
- Requests still arriving outside the front door, tracked by lawyers forwarding them in