AI Job Intake — How It Works

For Firms
7 min read
AI Job Intake — How It Works — 4 min Video coming soon. Written guide below covers everything.

1. Document extraction

When you upload service documents (PDF, JPG, PNG, or DOCX) during job creation, Proxiant runs AI extraction automatically. The AI reads the full document and attempts to identify:

Recipient: full name, entity type (individual / company / government)
Service address: street, city, state, zip
Court: court name, jurisdiction, county
Case information: case number, case title, filing date
Job details: document type (summons, subpoena, etc.), client name / law firm, attorney name
Due date: when present in the document (e.g., hearing date or service deadline)

Extraction runs in under 5 seconds for most documents. The results appear in the extraction review popup before any data is saved.

Tip: Upload the summons or main service document first. Supporting pages (exhibits, complaint) can be added too — the AI reads all attachments but gives priority to the first uploaded file.

2. Understanding confidence levels

Every extracted field is assigned a confidence level. These are displayed in the extraction review popup as color-coded labels:

LevelWhat it meansAction required
HIGHAI extracted the field cleanly from clear text in the document with high certainty.Glance to confirm. Pre-accepted by default.
MEDIUMAI found the field but there may be ambiguity — multiple possible values, partial text, or inferred from context.Review and explicitly confirm each one before saving.
LOWAI could not extract this field reliably. Value is either blank or a low-confidence guess.Fill in manually. Do not rely on the suggested value.
Important: MEDIUM fields are highlighted in amber and require your explicit confirmation — they will not auto-save without your review. Always check MEDIUM fields carefully; these are the most common source of data entry errors.

3. Client disambiguation

When the AI finds a client name in the document (e.g., "Smith & Associates Law Firm"), it looks up matching records in your client database. If it finds exactly one match, it pre-selects that client as HIGH confidence.

If it finds multiple possible matches (e.g., you have three clients with "Smith" in the name), a disambiguation card appears before the rest of the extraction review. Each possible match is shown with the client name, contact, and recent job count — pick the correct one before continuing.

Note: If no match is found, the client field is left blank and flagged LOW. You can add the client inline during job creation — click "Add new client" in the client field dropdown.

4. Email intake setup

Proxiant can create jobs automatically from inbound emails. Clients or staff email a service request (with attachments) to your firm's intake address, and Proxiant creates a draft job for review.

To configure email intake:

1. Go to Settings → Email Intake.
2. Your intake address is pre-configured as service@yourfirm.proxiant.co.
3. Optionally, configure a forwarding rule from your own domain (e.g., service@yourfirm.com) to your Proxiant intake address.
4. Set the default review state: Auto-create and notify (job created immediately, staff notified) or Hold for review (job goes to an intake queue for manual confirmation).

Tip: Set up per-client intake email addresses to improve client matching. A client's email domain can be mapped to their client record in Settings → Clients.

5. Email threading

For follow-up emails about an existing job, use the job-specific address:

service+JOB0147@yourfirm.proxiant.co

Replace JOB0147 with your actual job number (no spaces, no special characters). Emails sent to this address are attached to the correct job as notes, including any file attachments.

This is useful when clients send additional documents after a job is created, or when you forward an email conversation into a job's timeline for record-keeping.


6. Tips for better extraction results

Use text-based PDFs, not scanned images. Digitally generated PDFs (from Word, court e-filing systems) produce near-perfect extraction. Scanned documents work but may produce more MEDIUM/LOW fields — especially for addresses and case numbers.

Avoid heavily redacted documents. Redacted fields will obviously not extract, but heavy redaction can also confuse nearby fields.

First page matters most. The AI reads the full document but weights the first page heavily. If the summons header is on page 1, extraction is typically excellent.

Common issue: Handwritten addresses in the "To be served" line often extract as LOW confidence. If your clients commonly use these forms, plan to manually verify the address field even on otherwise clean extractions.

7. What AI never does

🔒 Trust guarantee: Nothing is written to the database until you explicitly confirm the extraction and click Save. The AI is advisory — every field can be overridden, and no job record is created without human sign-off. AI extraction results are not logged as permanent data; only your confirmed values are saved.

The AI also never:

• Contacts clients directly
• Sends notifications before job confirmation
• Assigns servers without your instruction
• Makes billing or invoicing decisions
• Accesses documents outside your firm's account