User feedback analysis scenario help

I want to analyse user feedback that gets sent to my company email account, we use google gmail business account. Here’s how the email feedback is managed by myself:

  1. User submits feedback on the site
  2. I receive the email
  3. It is then labeled automatically using gmail filter
  4. I have set up multiple labels to tag each email based on page section type

In Make I’d like to run an on demand scenario that allows me to use chatgpt to analyse the feedback, and I can set a date range as well, then send the summary to a google spreadsheet, and a google doc.

The google spreadsheet would be more tabular style to show data points etc - and the google doc would be a long-form summary.

I’d like to anaylise: Sentiment Analysis, Pain Point Identification, Feature Requests & Improvement Suggestions, Bug & Technical Issue Detection, Actionable Insights & Prioritization, User Experience (UX) Analysis, Keyword & Theme Extraction, Emotional Analysis & Tone Detection, Engagement & Motivation Analysis and Customer Journey Analysis to name a few.

Any help would be much appreciated.

What you are describing seems pretty straightforward. The best way to do that is to create the prompts in ChatGPT until they work, then put them in a module in Make in your scenario automation. I would make specialized modules for each so you don’t have to create one prompt that does everything (although if you can, it will be less costly in API usage). Then you take the information from each module and put it in a Google Doc.

I do that with a Google Forms submission. When the form is submitted, it calls my scenario, which analyzes the content with ChatGPT, creates a Google Doc, and sends me the link to the Google Doc to my email.

Your biggest challenge will probably be to do the prompt engineering to your liking.

L

Thanks, I haven’t attempted anything like this before so trying to figure it out as I go.

Could I ask ChatGPT to generate the analysis then parse it as JSON for each specific feedback type e.g Sentiment?

Then map those to specific columns in a spreadsheet, and then one for the google doc summary?

Yes you can. If you are able to do it in ChatGPT directly, you should be able to use that same prompt in the ChatGPT module and get a simlar result. Things to consider:

  • Every call to the module uses API credits so you need to be lean in your prompting
  • While gpt-4o is the better model, it’s more expensive than 4o-mini. See if you can get it to work correctly with 4o-mini in ChatGPT, especially if you expect a large volume of calls
  • Since Make works with JSON, make sure that ChatGPT only output the JSON without the commentary it like to add every time

Have fun!

L

I managed to get it working, and outputting as JSON so I could map it to the spreadsheet. I plan to add a router since there are multiple categories we have so will have separate spreadsheets modules for each.

Here’s the current scenario build:

I’m trying to extract some information from the email body, which is the next challenge - do you have any advice on this?

There’s basically 2 URL’s I’m trying to extract and populate a column in the spreadsheet.

Also is there anyway to change the scenario to target a different gmail folder?

I can just easy to that manually in gmail module, and change the destination google sheet in the last module. But I wonder if there’s a way to this more quickly?

As I have around 20 feedback folder categories I would be running the analysis through.

As I don’t want to create 20 google sheet modules as it gets a bit complicated and less manageable, e.g if there’s a change to the last module, then all other sheet modules need to be updated.

HI,

There’s a lot of questions. Think about asking each of them separately in different threads so others can see what you’re doing.

Extraction of email content: Where are you having the problem? Is it your prompting? Is it something else? You can use a text parser with a regular expression to extract the URLs. Then you can transfer them to a column in your spreadsheet. But without seeing the output, it’s hard to tell.

What do you mean by this? I’m not quite sure what you mean. How do you do it when you do the work manually? THat’s what you need to plan for in your automation. If you are able to stremline the steps you take to do the work manually, you should be able to do the sam with Make.

Can you aggfregate all your results so you put everything in a single Google Sheet? I don’t have a clear vision of what you are doing, so that might not be a pertinent comment.

If you can, share your scenario snapshot in a different conversation, explain what you’re trying to doand where you are stuck and let’s see if we can help.

L

That’s the scenario where it is at the moment, so far I’ve got it doing pretty much everything I need. All I need to do to change the source is open the gmail module and change the target label folder.

I did run into a few issues in the ChatGPT module where I had to force stop it since it was taking way longer than usual, it did vary as well e.g it could hang for 4 or 8 runs. So I need to figure our what could be causing this.