.png)
Shift Technology is a French start-up that developed a platform to detect fraudulent insurance claims.
At Shift, I had the opportunity to design a system of features that would enhance the utility and UX of Shift’s detected Network visualization.
.png)
Attempts to benefit from deceitful claims is a huge problem around the world. It means billions in stolen money, and ultimately creates mistrust between insurers and beneficiaries.
Shift uses Artificial Intelligence (AI) to help fraud investigators and claim handlers know when a claim can be trusted. The AI pulls out information to support complex investigations and helps to automate claim reviews, which helps insurers feel confident in giving deserving customers the help they need.
The more data that was analyzed by Shift, the more important it became to insurers to have a global picture of different relationships or patterns in the broader fraud detection scope. Understanding the data through discrete entities wasn't enough.
When looking at one entity, it isn't clear if they are part of a greater fraud network, or an isolated case.
In early 2020, the product team began receiving feedback about the value and potential of Networks in solving investigations. Here was the main feedback about the current situation of the Network at the time:
Project duration: 3 months
.png)
Problems with: unclear action icon meanings, lack of visual hierarchy & structure, small node icons, generally outdated UI design.
In order to serve our users, and to understand the initial feedback better, the product team and I conducted interviews and observed the investigation practices of AXA France's investigation team. We spoke to:
Dedicated to uncovering fraud and fraud links, and rely on networks to discover connections. They needed to easily identify patterns and communicate those with team members.
AXA France fraud investigator looking through documentation to understand fraud cases.
Here are the main three issues we discovered that would inform our work going forward:
The Network's UI was small, unbalanced, and hard to navigate. Buttons were unclear, there was no zoom, and reviewing was an overall uncomfortable experience.
Elsewhere on Shift's product, users would get alerts if entities were found to be fraudulent or suspicious. On the network, that information was not represented.
While investigating the network, there was no way to share findings. Users would find interesting patterns, but could not document or share them with team members.
To do this, we sliced the feature 3 ways:
The network had to be easy to review and interact with. I proposed:
We created two new network nodes that would visually distinguish fraudulent and suspicious entities from the rest.
We added new interaction functionalities that would support our users' review and documenting needs:
I'll be focused on the markup feature, as the others are confidential.
Before exploring any designs concepts, there was information we needed to look into.
To understand markup and pattern identification in different contexts, I analyzed the interactions and designs of other networks like IBM, Linkurious, and Tableau.
I continued collecting UX feedback from users and ensured that the new features took their concerns into account. Visibility, navigation, and simplicity were key.
I worked closely with the tech team to ensure my designs could be adjusted to the technical infrastructure they were already using. We experienced many hardpoints around the limitations of this infrastructure and had to work around it.
While looking for design solutions, I made sure that I was taking Shift's existing design system into account. I implemented existing patterns where I could, and otherwise I meaningfully created new ones.
The network needed to visually support large amount of visual information. Most of the network had to be re-designed, so I explored different design treatments to ensure discovering and reviewing the information was simple.
After strategizing how the Network UI would evolve as more features were added, we defined how we would ensure scalability and reduced cognitive overload.
Weekly design critiques from the PM and the Engineering team also helped to define designs that were not limited by the technical infrastructure of networks. I also worked with Data Scientists at Shift to ensure the designs covered complex usecases and edgcases.
We continued to collaborate with our users and scheduled monthly feedback sessions. Then, we tested with internal and external users to improve our UX. We learned it was important to:
After defining the concept and validating experience with all stakeholders, I moved on to creating the specs that would be delivered to development.
The network was reimagined to be clearer and easier to navigate.
Screen territory and navigation
The Network UI was made much bigger. An expand option was added, as well as a zoom for easier navigation.
Icons and buttons
The icons in the legend and buttons in the toolbar were clarified. Relationship icons were also refreshed.
.png)
When investigating, users needed to know if entities were confirmed fraud cases, or just under suspicion.
Fraudulent node
Fraud nodes need to be clearly identifiable. The checkmark icon confirms the known fraud, and the red indicates the concern.
Suspicious nodes
Suspicious nodes need to act as alerts. The alert icon indicates the need to review, and the orange indicates the warning.
Users could use markers to identify patterns and communicate findings to other team members using pre-configured marker types.
Actions dropdown
Users could add markers to one or more nodes at a time by selecting them and applying the marker action from the network toolbar.
Adding markers
Users could then select specific markers to apply to nodes. An icon representing the marker would appear on each of the nodes.
.png)
Users would typically only need to investigate one fraud theory at a time. That's why filtering on markers would be useful!
Using the filter
Users could filter on an in-use marker from the filters component. Non-applicable markers would then disappear.
Discovering patterns
The remaining markers would only apply to a specific suspicion, and users could more easily review the investigation.
.png)
To avoid cognitive overload, the user would need to be able to remove information that wasn't pertinent to their investigation.
Configuring view settings
Added information, like markers, notes, etc, could be removed from the view using the view dropdown.
Scaling for the future
These configurations also meant that in the future, Shift could add more information to the network.
.png)
The success of this feature was measured in three significant ways:
Multiple investigations teams informing us that they are very happy about the new experience and features on the network feature.
The network features enabled Shift to beat out dozens of other companies (including IBM) to secure coveted tenders worth millions of euros. Shift also raised over 200 million in funding after shortly after the release of these features.
This feature saw immediate usage numbers on Amplitude:
Target: 30% to 60% of users trying the network at least once
Result: 70%
Target: 35% of users using the network investigations features regularly
Result: 40%
Engineers: Louis De Courcel, Alexis Hadad, Houssam Otarid, Kassem Abboud
Product Manager: Emily Harrison
Design Team: Allison Kapps
