Discovery AI uses predefined mappings to extract provisions from uploaded documents. To enable this process, you must define the types of documents the system will process, like master service agreement (MSA), statement of work (SOW), or non-disclosure Agreement (NDA). For each document type, you must set up the data extraction mapping that Discovery AI uses to identify and extract provisions. Data extraction setup is process of mapping the built-in and custom provisions with the CLM schema so that extracted information like fields, clause, tables and obligations can be published on the contract record.
If your organization uses custom provision models and has enabled X-Author on Salesforce, a data sync containing standard provisions and seed data (ID: 10ce17c4-ee2e-4351-8d1a-27c4e9e7877a) is already present for onboarding. In such cases, make sure the custom provision models are correctly mapped and included above the data sync configuration.
Click the Data Extraction Setup tab in the left navigation bar to raise the Data Extraction Setup window.
Click the New button.
This raises the New Document Extraction Mapping pop-up.
Give the new mapping a name, enter a brief description, and choose a record type from the Choose database record pull-down menu.
The database record value corresponds to a particular contract or document type (such as MSA or SOW) to be imported by Discovery AI. For example, in a CLM deployment, this shows agreement records.
CAUTION:
Once a data extraction setup is created, you cannot modify the Database Record field to change from one record type to another, nor is the clone feature available. A mapping error may result in a complete rework with a new data extraction setup.
In the Conga platform, database records, also called record types, are used to categorize contracts by type and to define which fields, clauses, and extraction logic to apply. Because each contract type has unique clauses and data fields, mapping each extraction to one record type ensures that extractions are performed based on the contract type selected.
You can create only one data extraction setup for each database record type. To support different extractions of the same database record type (for example, "NDA-America" vs. "NDA-EMEA"), use multiple worksheets under the same data extraction mapping.
CAUTION:
Do not map a provision model to more than one field or clause. If a reviewer modifies a provision model mapped to multiple fields or clauses, there will be conflicts that affect the provision model's performance.
Select review levels in the Choose review level pull-down menu.
To deactivate the requirement to review, select 0: Skip Review. Document review will be skipped and no contract summary will be generated or saved to CLM.
Note:
When a skipped document is successfully extracted, the Reviewed By field remains blank and the file is assigned a Completed status. If any field's extracted value cannot be normalized to match its data type, the file is assigned a Ready for Review status and must be reviewed. On review, the document's Reviewed By field is populated by the user name of the document's reviewer.
To use a single-stage review process, select 1: One level - Review and Complete from the Review Level pull-down menu. You must then select a reviewer from the resulting Reviewer pull-down.
For a more formal process, select 2: Two levels - Review and Approve from the Review Level pull-down menu. You must then select a reviewer from the Reviewer pull-down menu and an approver from the Approvers menu.
To let users to check out batches of documents for review, slide the Allow reviewers to get next batch toggle on and then set the batch size. This enables users to check out a block of records of the designated number with a single "get batch" action. Users cannot request a new batch until they review or reassign all documents currently assigned to them.
Under Import Options, select extraction options to make available to your users. These are:
OCR to import the document as searchable text
Automated Extraction to submit the document for AI-only extraction, or
Verified Extraction to submit the document for AI extraction with attorney review.
These choices are not mutually exclusive.
Depending on whether this is a purchase or vendor agreement, set the Organization Role? toggle to either Buyer or Seller.
This setting contextualizes the generative AI prompt for Insights. Depending on this setting, the customer or user is presented to the AI engine as a buyer or seller. The default position is Seller.
If you want reviewers to be able to extend the worksheet with additional detected provisions as described in Adding Unmapped Fields and Clauses, slide the Allow Worksheet Extension toggle on.
Note: Adding an unmapped value links the identified passage to the appropriate provision map, associating it with the corresponding fields, clauses, or values defined in the CLM schema. Future document imports will recognize and extract these new provisions, but this update does not retroactively apply to previously imported documents.
Click SAVE.
You are returned to the Data Extraction Setup window. The mapping you just created is listed.You must populate the new mapping, defining its basic details and organizing provision models into worksheets, as described in Editing a Data Extraction Mapping.
Click the Data Extraction Setup tab in the left navigation bar to raise the Data Extraction Setup window.
Click the hypertext name of the map that you will edit in the Name column.
This raises the worksheet mapping window.
Click the WORKSHEETS tab.
Select the worksheet you will edit from the Choose Worksheet pull-down menu.
Select the Fields, Clauses, Tables, Insights, or Obligations tab, then click MANAGE PROVISIONS or NEW to raise the corresponding mapping pop-up.
For fields, click MANAGE PROVISIONS and continue to the next step.
For clauses, an Automatic Clause Comparison toggle appears. Activating this enables automatic clause comparison for all documents during import as described in Reviewing Imported Clauses and Fields. If this is turned off, the reviewer does not see the comparison summary, but it remains available on reviewer request. Click MANAGE PROVISIONS and skip to step 10.
For tables, a Default Orientation pull-down appears. Configure the anticipated table orientation as described in Horizontal and Vertical Tables. Click NEW and skip to step 8.
For insights, click NEW and follow the process described in Insights Setup.
Click Select Field(s) to Map in the Map Fields window.
Select one or more fields by clicking the adjacent check box in the Select Field(s) to Map pop-up.
For each field you select, choose whether it is to be automatically or manually extracted.
If the extraction is manual, no further configuration is required for this field: It will not be extracted automatically, and the reviewer will be responsible for selecting the text.
CAUTION:
Selecting Manual Extraction resets any automatic extraction configurations entered.
Indicate each required field by checking the Make this Required Field check box.
Some field types offer a Permit New Record Addition option. Check this box to enable reviewers to add new records. If you pick this, you must add the types of records a reviewer can add.
Click Add/Edit Provision Mapping to select the provision model or models to associate with the field you are defining.
Tip:
If you choose multiple similar provision models, your mapping will be more likely to find a field, but also more likely to find an incorrect one. Your reviewers must assign the result to one provision model or another, so a more selective approach is usually indicated.
Click DONE. You are returned to the WORKSHEETS tab view.
Click Select Clause(s) to Map or Create a Clause to Map.
You will add clauses to your mapping either by selecting existing ones or by creating your own.
If you choose to select clauses, choose the pertinent clause from the Add Clauses from Library pop-up. If you elect to create a clause to map, you are taken directly to the Map Clauses window to configure your clause mapping.
If you are creating a clause, you must name it. If you have selected one, it already has a name, which you cannot edit.
For each clause you select or create, choose whether it is to be automatically or manually extracted.
If the review is manual, no further configuration is required for this clause: It will not be extracted automatically, and the reviewer will be responsible for selecting the text.
CAUTION: Selecting Manual Extraction resets any automatic extraction configurations entered.
Indicate each required clause by checking the Make this Required Field check box.
Click the Extract Complete Paragraph check box to extract the provision's full context when it is detected.
Click Add/Edit Provision Mapping to select the provision model or models to associate with the clause you are defining.
Tip: If you choose multiple similar provision models, your mapping will be more likely to find a clause, but also more likely to find an incorrect one. Your reviewers must assign the result to one provision model or another, so a more selective approach is usually indicated.
Click DONE. You are returned to the WORKSHEETS tab view.
For obligation extractions, clicking the Obligations tab raises a grid view of existing obligations. Click the obligation you will edit, then the More icon (), then select Edit to raise the obligation edit pop-up.
You can choose either Auto-Extraction or Manual Extraction.
Note:
Obligations are fully defined in CLM. This setting controls the one feature that Discovery AI introduces. For more on configuring obligation objects from CLM, see Obligation Management in the CLM user guide, which contains sections on Creating an Obligation and Obligation Fulfillment.
You can configure at most ten obligations in one mapping.
You can introduce at most five fields per obligation.
Select the table provision and choose either the automatic or manual extraction method. Click SAVE to commit the table provision to the mapping.
Selecting Manual Extraction ends the configuration. Reviewers must extract the table manually, and no further configuration is possible.
Note: You can only map one CLM object per table mapping. You cannot map two tables with a common agreement line item in one mapping.You can configure at most ten tables in one mapping. You can configure at most ten columns per table.
Click the Data Extraction Setup tab in the left navigation bar to raise the Data Extraction Setup window.
Click the hypertext name of the map that you will edit in the Name column.
This raises the worksheet mapping window.
Click the WORKSHEETS tab.
Click WORKSHEET ACTIONS and select NEW from the drop-down menu or select CLONE and edit the cloned mapping.
Tip:
If you are starting from scratch, look for an existing worksheet to clone. Many orgs are deployed with pre-configured sample worksheets that you can use as a starting point.
In the New Worksheet pop-up window, enter the new worksheet's name and a brief description. Click the Default Worksheet check-box to make this worksheet the default when a document of this type is imported.
Click the tab for the type of worksheet you will create: Fields, Clauses, Tables, Insights, or Obligations, and follow the linked documentation to complete your worksheet.
The Admin can create a new worksheet under the same data extraction setup or edit, delete, or clone an existing worksheet from the Worksheet Actions menu.
You have Data Extraction Setup open to a mapping with the WORKSHEETS tab selected.
Select the Fields tab, then click MANAGE PROVISIONS to raise the corresponding mapping pop-up.
Select either Auto Extraction or Manual Extraction.
Provisions marked for auto-extraction are automatically extracted by the system and available for review.
Provisions marked for manual extraction require the end user to map and approve the field value manually on the review screen. Manual extractions require no further provision configuration.
CAUTION:
If you select Manual Extraction, you will lose any configuration information for this mapping.
Select Make this Required Field to make the extraction required.
If the extraction is required, the reviewer cannot close the review without a value for the provision.
Select the provision models to apply from the Available Provisions list and click the right-arrow icon to move them to the "Mapped Provisions:" list.
For workflows set to Skip Review (review level 0), this field must have a value.
Note:
When a default value is assigned to a provision, it appears in the reviewer's Provisions Found section, clearly marked as a default value. A default value does not fail validation checks for required fields.
If the reviewer modifies the default value or marks it as reviewed, the updated value is saved to the agreement record.
If the user does not modify the default value or mark it as reviewed, the default value is saved to the agreement record.
Default values are not sent to the AI engine for training.
Select the provision models to apply from the Available Provisions list and click the right-arrow icon to move them to the "Mapped Provisions:" list.
Tip:
You can narrow your provision model choices dramatically by entering one or more words in the search bar.
Hovering over a provision model type in the Available Provisions column raises explanatory text describing the model.
For fields associated with lookup values (values already defined in CLM), an additional control, Allow to add new record appears. Select this to enable reviewers who encounter lookup values to add lookup values to CLM from the extraction. If this is selected, you must select the associated lookup fields from the object's Account Values drop-down menu. The reviewer will have to enter these fields manually to generate a new record.
Note:
You can configure at most 70 fields and clauses in one mapping.
Click DONE. You are returned to the WORKSHEETS tab view.
You have Data Extraction Setup open to a mapping with the WORKSHEETS tab selected.
Select the Clauses tab, then click MANAGE PROVISIONS to raise the corresponding mapping pop-up.
Select either Auto Extraction or Manual Extraction.
Provisions marked for auto-extraction are automatically extracted by the system and available for review.
Provisions marked for manual extraction require the end user to map and approve the field value manually on the review screen. Manual extractions require no further provision configuration.
CAUTION:
If you select Manual Extraction, you will lose any configuration information for this mapping.
Select Make this Required Field to make the extraction required.
If the extraction is required, the reviewer cannot close the review without a value for the provision.
Select the provision models to apply from the Available Provisions list and click the right-arrow icon to move them to the "Mapped Provisions:" list.
For workflows set to Skip Review (review level 0), this field must have a value.
Note:
When a default value is assigned to a provision, it appears in the reviewer's Provisions Found section, clearly marked as a default value. A default value does not fail validation checks for required fields.
If the reviewer modifies the default value or marks it as reviewed, the updated value is saved to the agreement record.
If the user does not modify the default value or mark it as reviewed, the default value is saved to the agreement record.
Default values are not sent to the AI engine for training.
Select the provision models to apply from the Available Provisions list and click the right-arrow icon to move them to the "Mapped Provisions:" list.
Tip:
You can narrow your provision model choices dramatically by entering one or more words in the search bar.
Hovering over a provision model type in the Available Provisions column raises explanatory text describing the model.
Activate the "Automatic Clause Comparison" toggle to enable automatic clause comparison for all documents with extractable clauses during import.
If the "Automatic Clause Comparison" toggle is turned off, the reviewer does not see a summary of clause similarities and differences for every imported document they review. However, this comparison feature remains available on reviewer request. See Reviewing Imported Clauses and Fields for more.
Note:
You can configure at most 70 fields and clauses in one mapping.
Click DONE. You are returned to the WORKSHEETS tab view.
You have Data Extraction Setup open to a mapping with the WORKSHEETS tab selected.
You have defined one or more risk scales as described in Risk Definition.
Insights are AI-driven evaluations that interpret extracted contract data to identify what is important, risky, or actionable within an agreement. They transform extracted provisions into meaningful intelligence—helping users quickly understand contract implications and make informed decisions. Powered by Conga's Discovery AI, Insights analyze contract language, terms, and clause structures, even when phrasing varies across third-party or legacy contracts. By leveraging predefined standards, risk scales, and mapping logic, the AI assesses provisions, assigns risk scores, and provides guidance to mitigate potential exposure.
For example, if a company standard specifies a payment term of Net 30 days but the extracted contract includes Net 60 days, the AI identifies the deviation, flags it as high financial risk (if the defined threshold is over Net 45), and may recommend negotiating the term to Net-45 or less.
If all fields and clauses are mapped for this worksheet, click Insights. If provisions remain to be mapped, map them before defining Insights.
Click the New button to map a new risk or click the check box for an existing risk definition, then click the More () icon and select Edit from the pop-up menu.
Select the insight type. Your choices are Risk and Question.
Choosing Risk indicates you want to assign information to a risk class, based either on pre-configured values or on the assessment of the AI. Selecting Question pre-defines prompts for the AI to interpret, with the answer appearing in the reviewer's results. For example, "What is the total contract value?" or "Which state laws govern this contract?" might yield "$150000" or "Delaware" in the Answer field.
You can program the tool to identify a risk using a Boolean assessment of comparative logic operations. Each condition compares the extracted data to pre-configured values, and acts when any, all, or custom conditions are met. For custom conditions you can apply Boolean AND or OR tests of multiple conditions: (1 AND ( 2 OR 3 ) ), for example.
For a question, name the question in Question Title and enter a natural-language question in the Question field.
You can limit demands on the AI by selecting a provision model from the Found in Provision field.
Tip:
You can use natural language to select non-standard operations. For example, "One month after the contract goes into effect" or "Does not contain an indemnification clause."
You can configure at most ten questions in one insights mapping.
No more than five comparative logic operation criteria are allowed for each insights mapping.
You have Data Extraction Setup open to a mapping with the WORKSHEETS tab selected.
Select the Obligations tab, then click NEW.
This raises the corresponding mapping pop-up.
Select the obligation you will edit from the Search Obligation Provision pull-down menu, pick either the Manual or Auto-Extraction method, and click SAVE.
Note:
You can configure at most ten obligations in one mapping.
You can introduce at most five fields per obligation.
Thresholds and Best Practices
When configuring and using the Discovery AI tool, keep the following thresholds and best practices in mind.
Uploads
Upload Property
Limit / Notes
Number of documents for OCR
2,000 per day per org/instance (for all users)
Document size
50 MB per document
Number of pages
200 pages per document
Files in bulk upload
No more than 100 files
Files for extraction per day
1,000 files
Supported file types
DOCX, PDF
Upload sources
Conga CLM
Local file system
MS SharePoint
Folder/bulk upload requirement
All files must be of the same document type (MSA, NDA, SOW etc.)
Documents imported from Conga CLM
200 documents
Limitations
Item
Threshold / Limit
Fields or clauses per document
50
Tables per worksheet
10
Note:
Vertically or horizontally-aligned tables are supported.
Hybrid table structures are supported.
Columns per table
10
Fields or clauses per mapping
70
Obligations per mapping
10
Fields per obligation
5
Criteria per insight
5
Questions per insight mapping
10
Forms layout
Not supported
Best Practices
Category
Best Practice
Document quality
Use high-resolution scans.
Avoid borders or decorative elements that interfere with content.
Table scan quality
Ensure the table is clearly visible and not distorted.
Layout requirements
Table should have a well-defined layout with clear headers and visible borders.
Header guidelines
No special characters or symbols in header rows or column names.
), then select Edit to raise the obligation edit pop-up.
) icon and select Edit from the pop-up menu.
