Most conversations about AI in diligence focus on the model: which one, how capable, how it reasons. In practice, the more common bottleneck sits a step earlier: can the tool actually see your data in the first place, across every system it lives in, without someone manually assembling it first.
The familiar version of this problem: an associate pulling documents out of a dataroom platform, cross-referencing notes from a call-recording tool, checking the CRM for deal history, and pasting all of it into one working document before any actual analysis starts. None of that is diligence. It’s data collection that happens to precede diligence.
Before any analysis starts
Standard connections, not bespoke ones
This is the same problem most of software already solved for itself: instead of building a custom integration for every pair of tools, you standardize the connection once and reuse it everywhere. Applied to a deal team’s stack, it means a dataroom, a document repository, and a CRM can all be reached the same way, without an engineering project every time the firm adopts a new tool.
That standardization doesn’t make the underlying reasoning any better on its own. It just removes the tax on reaching the data at all.
Access is not the same problem as trust
Being able to read four systems at once mostly buys you a very fast copy-paste. The harder and more valuable layer sits after the read: when the system makes a claim, can it point to the exact sentence and document it came from? Does it say nothing when the evidence is thin, instead of filling the gap with something plausible-sounding?
An unsourced claim is treated as a failure, not an edge case. Ungrounded figures come back blank rather than invented.
That’s a separate problem from connectivity, and it’s the one we’ve built around directly. That principle extends to what counts as readable in the first place: a scanned or image-only PDF that would have been invisible to earlier tooling now gets pulled into the same sourced workflow, though it’s recent enough that we’d still tell you to give those extractions a second look for now.
Reach, not just read
On the connectivity side itself, we’ve been extending what the workspace can pull in directly. Most recently, that means importing a live dataroom straight from DocSend into the knowledge base, rather than someone downloading and re-uploading files by hand.
What’s harder, and further out
Standard connections between a tool and your data are a solved-enough problem now. The less solved one is handing a clean result from one workflow to the next — a diligence output becoming the starting input for the model, becoming the starting input for the memo — without a person manually re-uploading or re-explaining context in between.
That’s a real direction, and it’s earlier and less proven, industry-wide, than the access layer is. Worth building toward. Not worth overselling yet.