Setup - AI

Setup - AI

AI Setup

The AI Setup page controls how Safari uses artificial intelligence (AI) to identify, classify, and extract data from served documents. Your company's Owners configure this page to specify whether AI classifies incoming documents during intake, whether AI populates matter fields, and what level of human review is required after AI processing. 

This help page covers the following:
  1. Overview of AI functionality and workflows
  2. AI document identification and classification (in Served Document Manager)
  3. AI data extraction (in Subpoena Manager and Levy Manager)
  4. Safari's AI compliance policies

AI Overview

Safari has a principled approach to the use of AI:  we use it only where it's well suited, such as to safely and effectively support workflow tasks like classification, summarization, and data extraction. Below are the primary ways we use AI in Safari.
  1. AI Document identification, classification, and splitting:  Safari's AI in Served Document Manager reviews an incoming file to determine how many documents it contains, what type each document is, and, following human review (if configured), whether to automatically split the file into separate served documents.  
  2. Info Why this matters?  Many scanned documents come from offices, branches, and mailrooms, where the person scanning may not have legal training to identify individual documents, and multiple documents often get scanned together as one file, and sometimes include non-served documents (NSDs). AI identification, classification, and splitting cuts down on manual work and routes documents to the right team or system faster. It also isolates NSDs from the served document workflow.
    Video:  [link coming soon]
  3. AI Data extraction:  Once a served document is uploaded into a matter, or as part of creating a new matter, Safari's Subpoena Manager and Levy Manager can extract data from the document to populate key matter fields. This works across document types, including subpoenas, levies, garnishments, and summons and complaints.
  4. Info Why this matters?  Manual data entry is slow and prone to error. AI data extraction cuts down on both. Video: AI Data Extraction

Where does each AI feature fit in the workflow?

The flow chart below shows where AI fits into the workflow for served documents:
NSDs (Non-Served Documents): Branch offices and mailrooms often submit NSDs for review with served documents, creating extra work. Served Document Manager identifies and separates these non-served documents to ensure proper routing apart from official legal process documents.
Alert
All AI features are disabled by default. Your company must turn them on in the AI Setup page before they take effect. Only Owners can edit the AI Setup page (under System Setup), and all changes are recorded in audit history.

How do we know that AI processing is accurate?  Can it be reviewed?

Safari's AI is trained to take action only when it has a high degree of confidence. For example, Safari won't populate a matter field unless it's confident the data is accurate.

AI features can also always be configured for human review. The review screens are designed to make it quick to complete reviews and make corrections when needed.  Safari's audit history captures who reviewed the AI and whether the reviewer made any changes.

AI Document Identification, Classification, and Splitting (Served Document Manager™)

Info
VIDEO - Triaging Served Documents: (3 mins) - Triaging Served Documents 
This section controls two things: whether AI document identification and classification is turned on, and when a human must review the AI's output.

Step 1:  Access System Setup
Go to System Setup and select AI Setup under the Intake section. If your company hasn't subscribed to Served Document Manager (SDM), this field displays "Served Document Manager not enabled." To learn more about SDM's AI capabilities, see SDM - AI Document Identification or contact your Customer Success Manager.

Step 2:  Turn on AI Identification & Classification
Set 'Enable Document Identification & Classification' to 'Yes.' This applies only to new Intake Events going forward. It does not apply to Intake Events that already exist.

Step 3: Choose when human review is required
Set 'Human Review of Document Identification & Classification' to one of three options. Each Intake Event file can trigger two separate conditions: the AI may need to split the file into multiple documents, and the AI may detect a non-served document (NSD) in the file. The table below shows when each setting requires review.
When is human review required or bypassed?
Setting
Single served document (no split, no NSD)
Multiple served documents (split needed, no NSD)
Non-served document (NSD) detected
Always Required
Required
Required
Required
Only when splitting or a non-served document is involved
BYPASSED
Required
Required
Only when a non-served document is involved
BYPASSED
BYPASSED
Required

In short: 'Always Required' reviews everything. The splitting option reviews everything except a single, clean served document. The NSD-only option reviews only when a non-served document is detected, even if the file also needed to be split.
Notes
RECOMMENDATION
Which setting fits your company depends on how your processors already review served documents at the matter level, and whether you're comfortable relying on that review to catch a split the AI missed.
  1. If your processors review the full served document when they work a matter, consider 'Only when a non-served document is involved.' A misclassified NSD is always caught before it's finalized, since every setting requires review whenever the AI detects one. The remaining risk with this setting is a missed split, since it's the only one that bypasses human review in that scenario. Choose it if you're confident your processors would catch an incorrectly split document during normal matter handling.
  2. If you want split decisions reviewed every time, without relying on matter-level review to catch a mistake, consider 'Only when splitting or a non-served document is involved.' This requires review any time the AI needs to split a file, in addition to any time it detects an NSD, so it doesn't depend on a downstream catch for either kind of error.
  3. If your company has zero tolerance for classification errors, or processor practices vary from person to person, we recommend 'Always Required.' This puts a human in the loop for every classification and split decision, rather than relying on matter-level review at all.

AI Data Extraction (Subpoena  Manager™ and Levy  Manager™)

Info
VIDEO - Reviewing Extracted Data: (2 mins) - Reviewing AI-Extracted Matter Fields
This section controls two things: whether AI data extraction is turned on, and when a human must review the AI's output.

Step 1:  Access System Setup
Go to System Setup and select AI Setup under the Intake section. 

Step 2:  Turn on AI Data Extraction
Set 'Enable AI Data Extraction' to 'Yes.' This applies only to new Intake Events going forward. It does not apply to Intake Events that already exist.

Step 3: Choose which Subtypes are Enabled
The toggle options are as follows:
  1. All. AI Data Extraction applies to every Type and Subtype
  2. Selected. AI Data Extraction applies only to the Types and Subtypes you select.
Step 4: Choose whether and when human review is required
When AI populates a field, a 'sparkle'  indicator marks it, which is what we call an "AI data alert." This setting controls whether and how users must address that icon before changing a matter's status. Choose the level of review that fits your company's risk tolerance:
  1. Review Required - Users must validate the AI-populated data before changing status
  2. Review Optional - Users can clear the AI data alert with one click, but are not required to validate to change status
  3. Alert Hidden – Users never see the AI data alert.
You can override this default for individual Subtypes, to do so:  
  1. Go to Subtype Setup and open the Subtype
  2. Locate the AI Data Extraction Review field
  3. Toggle from Use Default to Override
  4. Select the applicable review mode for that Subtype
Notes
Actions Tracked in Audit History & Reports
When a user verifies the AI data extraction and clears the alert, the person and time cleared are tracked in the audit history.  This information is also available in reports.

Safari's AI compliance policies

Safari SOP (“Safari”) applies artificial intelligence (“AI”) within a tightly controlled, auditable framework to enhance automation in the processing of legal orders such as government levies, garnishments, subpoenas, and related materials.  To review a copy of our statement of AI Governance and Data Privacy, please download the document at the bottom of this help page.

Is there a question we haven't answered? Please send it to us at support@safarilaw.com


    • Related Articles

    • Workflow - AI Data Extraction

      AI Data Extraction Overview When AI Data Extraction is enabled, Safari AI reads each served document and fills in the matching matter fields, so you spend less time manually keying data. Depending on your company's AI Data Extraction Review settings ...
    • SDM - AI Document Identification

      Served Document Manager™ - AI Document Identification SDM's AI finds the document boundaries inside an Intake Event and classifies each document by Type and Subtype, reducing the manual work of splitting and sorting documents during intake. Your ...
    • Setup - Subtypes

      Overview NOTE: Only system Owners can edit Subtypes. For more information see Setup - Users. Safari requires a Subtype field in every matter. You can use and configure Subtypes for several purposes, including matter assignment, workflows, transfers ...
    • Setup / Intake: Intake Email

      Safari Intake Emails INTAKE EMAIL FEATURE: IMPORTANT SECURITY CONSIDERATIONS Understanding the Security of the Intake Email Feature. While our Intake Email feature is designed to use Transport Layer Security (TLS) to encrypt emails during transit, ...
    • Setup - Users

      This help page for system Owners and Admins covers the following topics: Viewing List of Users Creating/Editing Users Includes information about adding outside counsel as users System and Access Permissions Deactivating Users Individual users can ...