AI Autofill From Mapped Responses
In Analyst Evaluation projects, analysts map Responder responses to a structured report template, creating a curated set of question-and-answer pairs aligned to specific evaluation criteria. AI Autofill From Mapped Responses builds directly on this foundation. Instead of drafting IC memo and Analyst Evaluation (AE) responses manually, users can now use AI to generate first-draft responses from those already-mapped Q&As, directly within the autofill workflow.
This feature is designed to reduce manual drafting effort, improve consistency across evaluations, and give teams a stronger starting point for review without introducing content from outside the mapped context. Each generated response is labeled to show whether its wording was taken as-is from a mapped source or adapted by the AI, and how confident the AI is that the answer is accurate.
Who This Is For
This feature is available to Requestor users and analysts working within Analyst Evaluation projects who have already mapped Responder responses to a template.
Prerequisites
To use AI Autofill From Mapped Responses, the following must be in place:
- An Analyst Evaluation project must exist and be in progress
- Questions must have been mapped to the relevant templates
- DV Assist must be enabled on your DiligenceVault account
If mapped responses are not yet in place, the AI mode for this feature will not have source content to generate from. Refer to the articles Creating an Analyst Evaluation Project and Mapping Responses to a Report Template before proceeding.
How to Use AI Autofill From Mapped Responses
- Navigate to the relevant Analyst Evaluation project from your DiligenceVault dashboard.
-
In the Analyst Evaluation tab, click the Autofill button and select Responses.
- Inside the autofill modal, select the AI Mode option. From the available modes, choose Mapped Responses as the content source. This instructs the system to generate the response using only the Q&As that have been mapped within the project.
Note: Currently only one content source can be selected, i.e. either Mapped Responses or Documents.
Setting Source Priority When Using Documents
If you select Documents as the content source and more than one document is used, you can choose which documents AI Autofill trusts first. Source priority is not available for Mapped Responses or when only one document is used.
- Most Recent (default): The most recent document takes priority when sources overlap. This is the existing behavior.
-
Customization: Drag and drop the documents into your preferred order of authority, for example placing a newly signed PPM above an older factsheet.
-
Initiate the autofill. The AI processes the mapped Q&As and generates a first-draft response for the selected field. Progress is displayed on the progress bar, and a notification appears on the right of your screen once the autofill is complete.
-
Review the generated response in the Autofill Source bar below it.
Once generated, each response displays:
- A purple star, which marks that AI was used to produce the response
- A match label (Verbatim Response or Adapted Response), which shows where the wording came from
- A confidence level (High or Limited), which shows how well the mapped content supports the response
- The AI reasoning, shown first, explaining how the answer was constructed
- The mapped Q&As used as source material, grouped under a collapsed Sources: chip
-
Click the Sources: chip to expand the list of mapped Q&As. To review source detail, click into any mapped Q&A. A side panel opens showing the reasoning first, followed by the full context for each contributing source.
- Review the draft response alongside the source information. Edit the content directly (by clicking on the text box) if needed before saving it to the project.
Note: Editing this content will remove the source attribution, as it is no longer strictly AI-generated.
Understanding Match Labels, Confidence, and Source Attribution
Every AI-generated response is accompanied by a transparent breakdown of how it was produced. This is designed to support analyst review, not replace it. Match and confidence are shown as two separate signals because they answer different questions: the match label tells you where the wording came from, and the confidence level tells you whether to trust it. The two move independently, so a Verbatim response can still have Limited confidence.
Match Label: Verbatim or Adapted
Every response is classified automatically. No admin needs to set up approved phrases in advance for the system to recognize a verbatim match. When a single mapped source fully answers the question, the AI prefers to use that source's language as-is and only adapts the wording when it has to. The label reflects how the response was actually generated.
-
Verbatim Response: The entire response comes from one mapped source, reproduced almost exactly. Changes to punctuation, formatting, or capitalization, and same-meaning word swaps (such as "and" to "as well as" or "however" to "but") still count as Verbatim. A changed fact, figure, name, or scope, a reordered list or clause, added or removed content, or a second source breaks it, even if only one word differs.
-
Adapted Response: Everything else, such as rephrasing a source or combining several sources. A response built entirely from word-for-word passages across more than one source is still Adapted, because the combination itself was never reviewed as a single answer.
Match labels apply to text paragraph, single-line text, and Yes/No with explanation questions. They do not apply to standalone Yes/No, dropdown, multi-select, grid, date, numeric, or file upload questions. These show the confidence level only, even when the AI read and interpreted text to select a value, because a selected value has no wording to classify.
Confidence Level: High or Limited
Confidence judges accuracy, not origin. It is a relative indicator of how well the mapped content supports the answer to the question. A Limited result signals that closer review is warranted before accepting the draft. Limited confidence is highlighted so it stands out, while High confidence is shown more quietly so it does not compete with the response for your attention.
- Verbatim, High: One source, and it fully answers the question. You can skim and approve.
- Verbatim, Limited: One source, but on its own it is a weak or incomplete answer.
- Adapted, High: Combined or rephrased, and fully and accurately supported.
- Adapted, Limited: Combined or rephrased, and the sources conflict with each other or leave information incomplete to fully answer the question.
AI Reasoning
A plain-language explanation of how the AI interpreted and synthesized the sources to produce the output. The reasoning now appears first in the Source bar and opens with a line naming the classification and what changed.
Mapped Q&As Used
The specific question-and-answer pairs drawn from the mapped content that informed the response. These now sit behind the collapsed Sources: chip so the reasoning stays in focus. Click the chip to expand the full list.
The purple star, match label, confidence level, and reasoning are always visible alongside the generated response. The purple star is a fixed marker meaning "AI was used here" and is different from the purple sparkle used for DV Assist. A Q&A retrieved directly from the database, without AI, shows no star and no confidence pill.
Autofill History and Audit Trail
To see the autofill history, click the Autofill button and select View Auto-Filled History.
All responses generated through this feature are saved with full metadata. The autofill history for a project records:
- The source type used (Mapped Responses)
- The contributing projects from which the mapped Q&As were drawn
- The source priority setting used for the run, when Documents is the content source
This audit trail supports internal review processes and provides a clear record of how each draft was produced, which is particularly relevant for IC-level workflows where traceability matters.
After completing an autofill session, the AI Autofill Summary provides a consolidated view of:
- The content source used (Mapped Responses)
- A list of all projects that contributed to the mapped Q&As for the session