Verify images with Zapier and Peltarion
Build an automated and effective system that verifies customer photo uploads.
Put yourself in the shoes of a car insurer. They receive a lot of insurance claims from their customers, and the insurer must ensure that every customer’s car is, in fact, damaged before paying any money. This is a tedious task but luckily a task that can be automated.
In this tutorial, we’ll show you how to set up a verification system that uses the customer’s information and image and then notifies systems administrators when something is wrong.
- Target audience: Intermediate users
- Estimated time: 15 minutes
You will learn
- Build a Zapier workflow
- Connect Peltarion to a Zapier workflow
Create a spreadsheet with insurance claims
Make your own copy of this spreadsheet with insurance claims that we’ve made. The spreadsheet includes information on each claim. Most importantly, it contains a link to a photo for each claim. You will use that photo in a Peltarion classification model.
Tips on tools for survey collection
You can also create your own spreadsheet that includes a photo for each claim. This photo must have a link so the Peltarion classification model can use it.
We used Typeform to set up a form that, from a chat window, collected dummy information about insurance claims from customers. You could also try AirTable, which also worked very well and, as an added benefit, also removes the need for a spreadsheet in the step below.
Create Zap: Peltarion ←→ Google Sheet
Now let’s set up a zap in Zapier.
Goal: Receive an email notification if a photo can’t be verified.
Log in to Zapier.
In the My apps window search for and add a Peltarion connection.
If the Peltarion connection is not in the list, search for "Peltarion".
We’ve pre-trained a demo model already that can classify different car damages.
Use the URL and Token below to access our deployed model. (You can also build and deploy a car classifier yourself in our car damage classifier tutorial.)
In the Zaps window create a Zap.
To connect the zap to your insurance claims spreadsheet.
App: Google Sheets
Trigger event: New Spreadsheet Row
Connect the zap to your Google account.
Select the spreadsheet with insurance claims you just copied from us.
To make predictions of photos added to the spreadsheet.
Action event: Prediction
Choose account: The car damage account you set up in the My apps window.
Set up the action: Select the data from the photos column.
Click the + icon to add another step (do not turn the zap on).
Add Action: Filter
To filter out the photos that are either predicted to belong to the class Unknown or where the Peltarion model isn’t confident in its prediction.
In Filter Setup & Testing:
Click Only continue if… and select Predictions in Peltarion and then Class.
Choose condition (Text) Contains.
Enter value Unkown in the last field.
This means that you will filter out all examples with the predicted label Unkown.
click + or
In Filter Setup & Testing:
Click Or continue if… and select Predictions in Peltarion and then Class Conviction.
Choose condition (Number) Less than…
Enter value 0.7 in the last field.
This means that if the two top predictions are very close, then the conviction will be low. That is, the model doesn’t give a clear prediction.
Math: Conviction = ln (1st candidate / 2nd candidate)
Click the + icon to add another step.
Add Action: Send outbound email
To send you an email if a photo is filtered out.
Search for Email by Zapier and add it (will then show as Action / Send Outbound Email).
Action event: Send outbound email. Enter this info:
Subject: Unverified insurance claim
Body: Click in the Body field
Select Insert data → New Spreadsheet Row in Google sheets → Photo
This will add a link to the unverified image to the mail.
Then you can add some text, for example, is unverified. Please take a look.
Done! Turn the zap on.
Time to verify that your zap works.
Add row to spreadsheet
Copy the row below and paste it into your spreadsheet:
Recieve a mail from the zap
Check you inbox for a mail from zapier. It might take some time since there are a couple of different tools that need to communicate with each other.
And that’s it! Your customized verification system is finished. In this tutorial, you learned to:
Build a workflow using apps you already use.
Deploy a Peltarion model in Zapier.
Integrate it into your processes.