Investigating Google Sheets AI

Google Sheets AI

Google recently announced enhanced functionality in Sheets with the addition of Artificial Intelligence (AI) to the “Explore” button. To satisfy my curiosity, I decided to take it for a test drive and blog the results.


I needed some data so I went to Google Trends and executed an analysis of the recently released movies “Wonder Woman” and “Guardians of the Galaxy Vol. 2”. I filtered the results to the region of the “United States”, the time span to the last seven days, and the category to “Arts and Entertainment”.

I downloaded the CVS results associated with the chart titled “Interest Over Time”.

I opened Google Sheets and created a new spreadsheet titled “Movie Comparison”. To add the data, I selected “Import” from the File menu and uploaded the CVS file with the “Import Action > Replace Current Sheet” selected. The end result was a data grid of three columns and one hundred sixty three rows. The first column was text and the next two numerical.


Without highlighting any of the cells I pressed the Explore button. As I expected the panel came up but the analysis was not meaningful. Highlighting the cells which contained data quickly produced a different result.

Explore mimics Google Now as it communicates with a series of cards and the interaction is live. That is to say, if you select a different range of cells while the Explore panel is active, if will recalculate and adjust its analysis to the change without the need to refresh the screen.

Status Card

The first card displays various stats about the data.

  • Sum of Highlighted Numeric Cells
  • Average of Highlighted Numeric Cells
  • Min of Highlighted Numeric Cells
  • Max of Highlighted Numeric Cells
  • Count(A) of all Highlighted Cells

Answer Card

The second card allows you to ask questions about the data. Well ok then, the following are my questions.

  • Question: Most popular time
  • Answer: 05/27/2017

This question is ambiguous because I did not include the movie title. We can infer the answer is the most popular time for the most popular movie. Looking at the data for 05/27/2017 confirms the inference. Obviously Google was able to determine by “most popular time” I meant the greatest number.

  • Question: Most unpopular time
  • Answer: Didn’t understand the question

The reverse question was not as successful. Google was not able to determine by “most unpopular time” I meant the lowest number.

  • Question: Lowest Wonder Woman time
  • Answer: Bottom time by Wonder Woman = 05/28/2017

Rephrasing the question and being more specific was successful.

  • Question: Highest Guardians time
  • Answer: Highest time by Guardians of the Galaxy Vol. 2 = 05/27/217

The test here was to use of the word “Guardians” as oppose to the full title. Google was able to get the inference.

  • Question: Average Guardians count
  • Answer: Didn’t understand the question

This question used mixed action words and Google did not try to guess. The quagmire is do I want the “average” or do I want a “count”.

  • Question: Average Guardians
  • Answer: Average of Guardians of the Galaxy Vol. 2 = 1.720238095

Removing the word “count” lifted the confusion and Google was able to process the request.

  • Question: Wonder Woman exceeds Guardians
  • Answer: Time with Wonder Woman: (United States) > Guardians of the Galaxy Vol. 2: (United States) = Every data point

As you can see from the result Google was able to translate the word “exceeds” to “>”. “Every data point” is my summation of a long list of results.

Formatting Card

This is just a quick and dirty way of adding color to various cells to enhance readability.

Analysis Card

This section is very interesting as it is Google’s take on what the data means.

Group Chart

The first item displayed is a line chart of all selected cells. In the circumstance where the cell data is text, such as a mailing list, Google may display a pie chart of the count by zip code. Options include embedding the chart into your spreadsheet and view the chart full size. If you decide to embed the chart into your spreadsheet all formatting options become available such as changing the type. Under the chart was the following comment.

The biggest difference between “Wonder Woman: (United States)” and “Guardians of the Galaxy Vol. 2: (United States)” is 99 at “157”.

This analysis is a little tricky. What Google is saying is the greatest difference of the two movies occurred at line 157 (row 161) of the selected data and the difference is 99 on a scale of 0 to 100. Wonder Woman peaked that day with 100 and Guardians of the Galaxy had 1. It would be interesting to see the comment from a more complicated sheet of accounting data.

Argh! Sell the business before it’s too late. You are losing money hand over fist. :0

I’m sure Google’s response would be less emotional in this scenario.


Many may be asking what in world is a histogram?

A histogram is a graphical representation of how many times different, mutually exclusive events are observed.

Each event has a rectangular bar which shows the count or frequency. In this dataset we are looking at searches over time (7 days). Under the first histogram the following is Google’s explanation of the “Wonder Woman” data.

Ranges from 7 (“2017-05-28T05”, “2017-05-28T06”, “2017-05-28T07”, and 3 others) to 100 (“2017-06-03T01”), but 80% of values are less than or equal to 35.

Under the second histogram the following is Google’s explanation of the “Guardians” data.

Ranges from 0 (“2017-05-30T05”, “2017-05-31T04”, “2017-05-31T05”, and 10 others) to 5 (“2017-05-27T17”, “2017-05-27T18”, “2017-05-27T19”, and 3 others).

Individual Charts For those who want to see the data separately, Explore creates separate charts.

Wonder Woman:

Outlying values for “Wonder Woman: (United States)”: peaks at 73 at “2017-06-02T01” and 100 at “2017-06-03T01”.


Outlying values for “Guardians of the Galaxy Vol. 2: (United States)”: peaks at 5 at “2017-05-28T16”, 5 at “2017-05-28T18”, and 5 at “2017-05-28T19” and dips to 1 at “2017-05-28T02”.


My Google Site

Wrap up

Wow! In this circumstance the amount of information is overload but it does identify the more popular of the two movies based on the number of Google searches.

I will admit in this test the human brain would have come to the same conclusion by scanning the columns; but what is important, Google came to the same conclusion with documentation as to why in about a second. And, Google would have come to a new conclusion if the dataset changed on the fly (ie. 2nd, 3rd, and fifth days only) in about the same amount of time. This a great example of the power of the cloud. I don’t need a “Tensor Processing Unit” in the device I am using, I just need to use one in the cloud for a brief moment.

There are limits to the technology. I was definitely not satisfied with the results when I asked “Gal Gadot’s telephone number”; but, this is a step in the right direction as it provides folks with a very compelling reason to use Sheets instead of some other product.