Digital insurance assistant: How the cloud is becoming a driver of innovation

Develop fast and innovative solutions based on the SAP Cloud Platform – that was the challenge at the SAP InnoJam 2018 in Basel. Our five-person team from Q-PERIOR won first place with the development of a digital insurance assistant. The task: To get an innovative and operable project rolling in just three days. The prototypes had to be developed on the basis of new applications with cloud-based SAP technology.

The idea: a digital insurance assistant

The focus was on assessment of hail damage in a manner that is as automated and uncomplicated as possible, i.e.specifically the creation of damage reports in the event of hail damage to a vehicle.

When implementing the idea, we additionally imposed the following limitations on ourselves:

The solution shouldn’t be another app
The way I communicate in my personal life is the way I want to communicate with my insurer
The damage report should be producible on-site and within max. 60 seconds
The damage report should be producible on-site and within max. 60 seconds

The infrastructure

But how can we develop this kind of digital insurance assistant ourselves without having our own infrastructure to meet such an ambitious objective? This is where SAP steps in to assist with their SAP Cloud Platform. There is an option to setup a trial account for testing purposes. It can be done in just a few short minutes on the official SAP Cloud Platform site. We then subscribed to the requisite services using the trial account’s cockpit. The APIs available online are also practical can be tested directly in-browser: https://api.sap.com/package/SAPLeonardoMLFunctionalServices

Of course, we weren’t able to leverage the full potential of the SAP Cloud Platform for our project. You’ll find further information on the official SAP site: https://cloudplatform.sap.com/

Technology used

The solution we developed consists of three core components to meet the various tasks in the overall solution.

We used the bot platform Recast to attain interaction between the damage report service and the effective messenger. This enables the linking of various channels. In our case, a connection was established with Facebook Messenger. However, linking via Telegram, WhatsApp, and other social channels is also easy to do. The chatbot interprets and structures the texts sent in Messenger. For example, the chatbot recognizes whether the text includes information about location and time.

It then forwards the structured information to an engine we developed using node.js. In order to render our own API accessible on the Internet, the node.js application was installed using Cloud Foundry on the SAP Cloud Platform.

The following should be seen as an excerpt from the source code, in which we present the user with a summary of the damage incident at the end of the process:

app.post('/summary', (req, res) => {
  console.log("MEMORY: " + JSON.stringify(req.body.conversation.memory) || "");
  console.log("Location: " + JSON.stringify(req.body.conversation.memory.location) || "");
  console.log("Datetime: " + JSON.stringify(req.body.conversation.memory.datetime) || "");

  res.send({
    replies: [{
      type: 'card',
      content: {
        title: 'News report: Hail damage',
        subtitle: 'Damage report: ' + req.body.conversation.memory.claimId,
        imageUrl: req.body.conversation.memory.imageUrl,
        buttons: [{
            title: 'Confirm report',
            type: 'postback',
            value: '/confirm'
          },
          {
            title: 'Cancle',
            type: 'postback',
            value: '/cancel'
          }
        ]
      }
    }],
  });
});

If the user confirms the damage incident, all of the data will be stored in the blockchain and the workflow process:

app.post('/confirm', (req, res) => {
  request.get({
    url: 'https://xxxxxxxxxxxxxxxxxx.hanatrial.ondemand.com/workflow-service/rest/v1/xsrf-token',
    headers: {
      'X-CSRF-Token': 'Fetch',
      'Cache-Control': 'no-cache, no-store, must-revalidate'
    },
    jar: true
  }, function(err, httpResponse, body) {
	  ...
      var token = httpResponse.headers["x-csrf-token"];
      var cookie = httpResponse.headers["set-cookie"];
      var requestCookie = "";
      var requestCookieArray = [];

      if (storage.getItemSync(token)) {
        cookie = JSON.parse(storage.getItemSync(token));
      } else {
        storage.setItemSync(token, JSON.stringify(cookie));
      }

      if (cookie) {
        cookie.forEach(element => {
          requestCookieArray.push(element.split(';')[0]);
        });
        requestCookie = requestCookieArray.join('; ');

        request.post({
          url: 'https://xxxxxxxxxxxxxxxxxx.hanatrial.ondemand.com/workflow-service/rest/v1/workflow-instances',
          headers: {
            "X-CSRF-Token": token,
            "Content-Type": "application/json",
            "Cookie": requestCookie
          },
          body: {
            "definitionId": "claiminboundprocessing",
            "context": {
              "documentId": req.body.conversation.memory.documentId,
              "userName": req.body.conversation.memory.userName,
              "claimId": req.body.conversation.memory.claimId,
              "blockchainId": req.body.conversation.memory.blockchainId,
              "conversationId": req.body.conversation.id,
              "imageUrl": req.body.conversation.memory.imageUrl,
              "mlTags": req.body.conversation.memory.mlTags,
              "damage": req.body.conversation.memory.damage,
              "imageHash": req.body.conversation.memory.imageHash || "",
              "location": req.body.conversation.memory.Location,
              "datetime": req.body.conversation.memory.Datetime
            }
          },

A solid guide on integrating Recast and node.js can be found here: https://recast.ai/blog/nodejs-chatbot-movie-bot

The final product

The end user starts the damage report process by sending an image to their insurer via Messenger – in this case, a photo of a vehicle damaged by hail. The image is then analyzed in the background: The first step classifies the image. If the object shown in the image is an automobile, then the probability of hail damage is additionally computed. Accordingly, the chatbot asks the end user about the damage incident’s location and time. After responding to all questions, the end user receives a summary of the information. If they confirm this information, then the damage will be logged with the insurer. After confirmation, all information is securely stored in the blockchain and the internal insurance process is automatically triggered via workflow.