Analyzing Excel Data and Preparing Chronologies with Copilot

These days it's widely appreciated not only among litigation support professionals, but also among attorneys and law firm employees in general, that generative AI can quickly and reliably perform many common tasks such as document review or document summarization which previously would have taken many hours to complete.

Here's a demonstration of just how easy it is to use Microsoft's Copilot AI to review data in an Excel file, and the content of a set of PDFs.

This is a .csv file containing baseball batting statistics from 1871 to 2023, which can be downloaded here.

If you simply upload the file to copilot and ask a general question, it can give you an accurate example. In this example, you can see that I have asked about the number of home runs hit by players for the New York Mets, but I have not specified the abbreviation used for this team ("NYN"), or the abbreviation used for a home run ("HR") used in the .csv file.

. . . and yet still copilot gives me accurate results. Note that it uses the abbreviations for player names used in the Batting.csv source file - so it's not getting the information from a general knowledge set.

It can also easily generate a visualization of the search results like this bar graph:

. . . and will upon request modify the chart to show the data according to a different metric.

Copilot may be very literal in interrupting the instructions you give it. Here when I asked for a list of players hitting higher than .400 in a season it included players with very few at bats.

. . .however it prompted me to give more specific instructions. Keep in mind that batting averages are not actually listed in the source data - copilot knows to perform the calculation by dividing the number of hits by the number of at bats.

Copilot also gives you the option to export the results from any operation it performs:

You will get your own Excel file to work with.

I also tested how copilot performed analyzing multiple document PDFs using trial exhibits from the Enron litigation which are available on the DOJ's web site: https://www.justice.gov/archive/index-enron.html

Copilot did not do a good job OCRing the PDFs - the supposedly text-searchable PDFs that it generated were actually files with blank pages.

The free version of copilot will only allow you to upload 3 files at once.

When I asked it to tell me the dates of three PDFs - each of a different document type - it not only gave me the correct dates, but also provided an accurate summary of each document without being asked to.

Note how for the PDF named, 'EXH063-00292.pdf' it's able to correctly identify the author and the pinpoint purpose of the letter.