The Report:
The Challenge:
In sales you often get big, lumpy Requests for Proposal's (RFP's) that need a lot of thoughtful consideration before writing the response....but all that means a lot of reading!.... and that sounds like hard work.
Wouldn't it be great if there was a way to extract the key information from the documentation and represent it in a way that provides an overview understanding of what is (and is not) important to them?
Below is my attempt to do just that. More of a fun attempt to use 'M' to see if it can do as much as other languages accomplish with regards to text analytics. It cant but it was still a good learning experience all the same and we have found that there are a couple of use cases where this 'tool' has been useful.
The following is the result with the methodology associated with how this was accomplished provided further down.
​
The Approach:
Extract text from proposal file: This step is completed manually by converting the file to .txt.
The .txt files were saved to a cloud location. In this case I used a SharePoint Online Document Library.
Power Query was used to extract the raw text from the .txt files and prep it before running the standard Text Analytics functions across the text. Shaping the data at this stage was rather complicated with the following being the key issues with links to solutions found.
removal or additional stop words and bulk find and replace of others: this required the use of a separate excel file which the stop words could be entered into (or the replacement value).
The above gave some modicum of control over shaping of the text content but is by no means as effective as using Python.
4 . Building the data Model: This was a pretty simple based on the schema below.
5 . Building Sentiment Bands: The use of What If Parameters were used to control what was categorised as positive and Negative sentiment levels. The banding options can be found under the more information circle on the report.
The Conclusion:​
Did it work? I think on the whole it did. When running through a number of proposals it accomplished what it was designed to do. Highlight the key words and themes that were coming through the different documents in the RFP packages. It would often highlight some key area of high importance to the client - yes something that a human reader would also pick up but it is interesting to augment a reader's perception with what is being shown. In the end the sentiment is probably not needed as the context of the word in use is important but buried. Plus I dont think the sentiment really applies that well to the type of technical language used in these types of documents.
So overall some value but more of a 'learning' exercise to see what Power BI could natively do in this space.