<img height="1" width="1" style="display:none" src="https://www.facebook.com/tr?id=319569654834053&amp;ev=PageView&amp;noscript=1">

What's New: Document Viewer, Higher Accuracy, Indemnity Extraction and More

by: diligenceengine | June 10, 2013
4 min read

We're constantly working to improve the DiligenceEngine system. Since our first test users tried the system nearly a year and a half ago, we've rebuilt our core interface to make it even easier to add documents to the system and access findings. We've significantly improved our accuracy and increased the number of provisions the system finds. We've made improvements on the business side, updating our website and bettering our sales and marketing. And our team is bigger and stronger than ever. One area where we've slipped: keeping blog readers informed about our progress. Sorry. Here are a few big things we've done in recent months.

Document Viewer

Over the winter, we added a document viewer to the system. It allows users to see highlighted versions of documents underlying summaries, and directly add to summaries by highlighting (without having to cut & paste). It's fully integrated into our system, so you can jump straight from summaries to the original text of a provision with one click. We've gotten very positive reviews on the document viewer. Here's a quick video of it in action:

[wpvideo A7KHqFag]

The viewer may look simple but is pretty cool compared with typical document viewers for two reasons:

  • Shows Text-Selectable Original Document. Our viewer shows the original document (whether it's a ten-year-old scan or an agreement in Word) and also allows access to the underlying language. It includes annotations like highlights, and allows users to select language and (i) copy to elsewhere or (ii) drop it directly into system-prepared summaries without cutting and pasting. Having a selectable original document is useful:
    • Showing the original document allows users to do tasks like check for OCR errors or see if a document's signed. Document viewers that don't show the original document force users needing to see the original to open another version of the agreement, which wastes time (and somewhat defeats the purpose of having a document viewer).
    • There are also real virtues to having a fully-selectable version of the document. In a fully selectable document viewer (like ours), users can:
      • see found provisions and results from word searches highlighted;
      • highlight new provisions and add them to summary charts automatically; and
      • copy text for use elsewhere.

    Showing either the original document or a fully selectable version is easy. Combining both is best for users. So that's what we built.

  • Browser-Based Without Java. Our viewer works in all modern browsers without having users launch a Java runtime environment. While it would be fairly easy to deliver a fully selectable version of the original document via a Java applet (at least one of the major virtual data rooms does this), many users are uncomfortable with using Java for security reasons. These features would also be fairly easy to deliver locally on a user's computer (i.e, you can copy from an OCRd original via Adobe Acrobat if the document is on your computer). While local installation works for some, there are real benefits to cloud-based software. And we think our system should be able to work without installation.

The document viewer and summary charts also show side-by-side (with a wide enough monitor and a modern browser).

Accuracy and Provisions

Our accuracy (as measured by recall) is now 90% or better on every substantive provision we cover (i.e., not necessarily on title/party/date, which are quite layout specific (and also quite easy for a human reviewer to fix (as opposed to, say, finding a change of control provision, which is hard for people to do)). As we've posted on previously, these accuracy numbers are based on thorough testing. And our testing system is itself much improved.

The system also now finds "indemnity" provisions. We believe this covers all the provisions that come up most often in M&A-related due diligence contract review projects. We are working to add more provision types. And may add ones where recall is high but not 90% if we feel the provision would be useful to our users at the found recall level. As it is, users can get more contracts reviewed in less time even if searching for provisions the system doesn't yet cover, thanks to the select & drop feature of the document viewer shown above. (We'll have more on this last point sooner or later.)


Our interface is looking good and simpler-than-ever to use. And our Word download templates are newly attractive.

Let us know what you think!

Share via:

Read more articles related to Machine Learning & Kira Systems :