Email Sentiment Analysis to improve customer service
What I love about my life as a #consultant is having the opportunity to hear customer problems and responding to them with something of value which improves their business in their own industry and market. What I love about being #Microsoft #Technology #consultant is working on a technology which not only cares about end users but also makes it easier for me (or any citizen developer) to come up with solutions easy to implement and with the #powerplatform and #msflow, many things don't even need to open my #visualstudio (which I love and open every day even if I am not coding - Sounds crazy, nah! :D). Let's get back to the track now!
My initial thought was that "How do I need to quantify if a customer is going to defect because they are not satisfied"? After pondering on few solutions, I could come up with the idea of "Email Sentiment" as KPI for customer defection. If a customer is not satisfied with a service, their first reaction is to send a bad email to the company (At least this what I do) before going to the social media. So I took the initial complaining email as a sign of losing customers. The next thing was how to implement the idea? And this is how I did:
The solution is extremely easy because #MicrosoftFlow provides an out of the box connector to the text analysis service and all you will need to do is to provide the service key and service endpoint. Below is how my #Microsoftflow looks like:
Azure returns the sentiment score along with its analysis as Positive, Negative and Neutral. The API returns a numeric score between 0 and 1. Scores close to 1 indicate positive sentiment, while scores close to 0 indicate negative sentiment. A score of 0.5 indicates the lack of sentiment (e.g. a factoid statement).
In my solution, I stored sentiment value as Whole Number, so I had to cast the decimal value between 0 and 1 to a number between 0 and 100. To do this, I used Operation step to multiply the sentiment score by 100 and cast it to an integer value. So I used the below formula:
Scenario
I had a request from my customer who was getting bombarded with case emails in its support department. The customer asked me to find a solution to prioritize emails based on urgency and probability of customers getting defected.My initial thought was that "How do I need to quantify if a customer is going to defect because they are not satisfied"? After pondering on few solutions, I could come up with the idea of "Email Sentiment" as KPI for customer defection. If a customer is not satisfied with a service, their first reaction is to send a bad email to the company (At least this what I do) before going to the social media. So I took the initial complaining email as a sign of losing customers. The next thing was how to implement the idea? And this is how I did:
Solutions
- The basis of the solution was to use Azure Text Analysis service to detect the email message sentiment. The underlying service being utilized was Azure Text Sentiment Analysis service.
- The next thing was to customize the email message entity to hold the sentiment value and potentially trigger a notification to manager or just sort emails based on their negative sentiment value.
- The last thing was to connect Power Platform to the Azure Text Sentiment Analysis service and get the sentiment value of email message from azure. I had two ways to implement this:
- The first solution was to write a customer action to call the service and pass the email text to the azure endpoint. On receiving the response of the analysis service, the action would return the sentiment as its return. Finally calling the action on a workflow which triggers on the Creation of Email Activity!
- The second solution was to use #MicrosftFlow and do everything without writing a single line of code. Obviously, I used this technique.
The solution is extremely easy because #MicrosoftFlow provides an out of the box connector to the text analysis service and all you will need to do is to provide the service key and service endpoint. Below is how my #Microsoftflow looks like:
Azure returns the sentiment score along with its analysis as Positive, Negative and Neutral. The API returns a numeric score between 0 and 1. Scores close to 1 indicate positive sentiment, while scores close to 0 indicate negative sentiment. A score of 0.5 indicates the lack of sentiment (e.g. a factoid statement).
In my solution, I stored sentiment value as Whole Number, so I had to cast the decimal value between 0 and 1 to a number between 0 and 100. To do this, I used Operation step to multiply the sentiment score by 100 and cast it to an integer value. So I used the below formula:
int(substring(string(mul(body('Detect_Sentiment')?['score'],100)),0,indexof(string(mul(body('Detect_Sentiment')?['score'],100)),'.')))
Note: #MicrosftFlow does not have round function so I had to convert the value to string and substring until the decimal point.
Key Points:
Key Points:All of the Text Analytics API endpoints accept raw text data. The current limit is 5,120 characters for each document; if you need to analyze larger documents, you can break them up into smaller chunks.- Your rate limit will vary with your pricing tier.
- The Text Analytics API uses Unicode encoding for text representation and character count calculations. Requests can be submitted in both UTF-8 and UTF-16 with no measurable differences in the character count.
Comments
Post a Comment