AI-Powered Autofill with Multi-Source Retrieval

Due diligence questionnaires often contain dozens or hundreds of questions that teams answer repeatedly across projects. Standard autofill matches one similar question at a time, which can miss relevant context sitting in related responses or supporting documents. AI Autofill searches across multiple sources at once, understands both the questions and the actual response content, and combines what it finds into a complete, well sourced draft answer.

Teams typically use AI Autofill to:

• Draft responses from the Q/A Library, past project responses, and uploaded documents in a single run
• Control which sources the AI trusts first when more than one source could answer a question
• See exactly how each draft was produced, including where the wording came from and how much the AI changed it
• Review faster, because every response is labeled with a match classification and a confidence level

Running AI Autofill

Navigation Path

Steps

• Click Auto-Fill in the top toolbar and select Responses.

  

• Toggle to AI Mode at the top of the Auto-fill window to enable multi-source retrieval.

  

• Choose one or more sources: Library Q/As, Project Q/As, and Documents. When Documents is selected, you can use an existing document or upload a new one.

  

• Set the source priority and configure the filters, described in the next two sections.

  

• Click Auto-fill as Drafts to run. The AI generates responses across all applicable questions and saves them as drafts. Because they are drafts, your team reviews and finalizes each one before it counts as a final answer.

  

Choosing Your Sources

• Library Q/As draws from your firm's centralized Q/A Library, the repository of approved responses your team maintains over time.
• Project Q/As retrieves answers from previous and ongoing projects, capturing how your firm actually responded in specific contexts.
• Documents brings in supporting materials such as pitch decks, fund overviews, or compliance policies, giving the AI access to information that may not yet exist in structured Q/A form.

You can select any combination of these sources. The system pre-filters content to your firm automatically, so responses are drawn only from data you own or have been assigned.

Setting Source Priority

When more than one source could answer a question, source priority decides which one the AI trusts first. Previously, AI Autofill prioritized sources purely by date, so the most recent answer or document always won regardless of where it came from. You can now control this order for each run.

The Auto-fill window offers two priority modes:

• Most Recent (the default): keeps the existing behavior, where recency decides the winning answer.

• Customization: lets you rank your sources manually.

  

Ranking Source Types

In Customization mode, drag and drop the three source types, QA Library, Project Responses, and Documents, into your preferred order of authority. For example, you can place approved QA Library content above older project responses.

Ranking Individual Documents

Within the Documents type, expand the list to reorder specific files, such as placing a newly signed PPM above an older factsheet. This nested ranking applies inside the Documents type only and does not change the order of the other source types.

Smart Filters for Precision

Alongside priority, a set of filters controls which content the AI is allowed to consider:

• Content Date Range: restrict source content to a specific time window so the AI works from the most current information available.
• Include Responses from Specific Entities/Investor firms: limit source content to responses from specific entities for fund specific or strategy specific accuracy.

• Select Specific Projects: draw responses only from the projects you select, so the AI references the most relevant prior work.

  

How the AI Builds a Response

• Multi-Source Search: for each question, the AI searches all selected sources at the same time to find the most relevant matches.
• Content Understanding: the system analyzes not just question similarity but the actual content of past responses and documents, identifying passages that directly address the question.
• Source Selection: the AI prefers verbatim source language when a single source fully answers the question and is the most recent and relevant under your priority settings. It adapts the material only when it has to.
• Draft with Provenance: every response is saved as a draft carrying a match classification badge, a confidence chip, AI reasoning, and its source details.

Reviewing an AI Generated Response

• Match Classification Badge: Indicates where the wording came from and how much the AI changed it.
• Confidence Chip: Shows whether the content can be trusted to answer the question, using either High Confidence or Limited Confidence. The match classification badge and confidence chip move independently, so a response can be copied word for word from a weak source or carefully combined from strong ones. Limited Confidence keeps an attention-grabbing treatment so it stands out, while High Confidence is styled more quietly and no longer competes with the response for your eye.
• AI Reasoning and Sources: The AI reasoning is shown first. Source detail sits behind a collapsed Sources: chip and expands inline when you want a closer look, keeping the upfront view clean. The same order applies in Response History and the source detail panel: reasoning first, then sources.
• AI Usage Indicator: A purple star on the bar marks that AI was used to produce the response. It is a static marker and is separate from the purple sparkle reserved for DV Assist. A Q&A retrieved directly from the database carries no star and no confidence chip because AI was not used to produce it.
• Response Feedback: Use the refreshed thumbs up and thumbs down controls on the bar to share feedback on a generated response.

Match Classification: Verbatim and Adapted

Every AI generated response is classified automatically. No admin has to pre-designate approved phrases for the system to recognize a verbatim match. The badge reflects how the response was actually generated.

Verbatim Response

The entire response comes from one source, reproduced almost exactly. Differences in punctuation, formatting, or capitalization, and same meaning word swaps such as changing and to as well as, changing however to but, or dropping a redundant the, still count as Verbatim.

The classification breaks, even over a single word, when the response contains a changed fact, figure, name, or scope, a reordered list or clause, added or removed content, or material from a second source.

Adapted Response

Everything else: rephrasing a source, combining sources, or reshaping a response to a custom prompt. Even a response built entirely from word for word passages across multiple sources is Adapted, because the combination itself was never reviewed as a unit.

Reading the Badge and Confidence Together

Auto-Filled History

A complete Auto-Filled History log provides a full audit trail of every autofill action: when it was run, which sources were used, which priority configuration was in effect, and what was generated. You can also delete auto-filled responses in bulk if a re-run with different parameters is needed.

Standard vs. AI Mode