How might we help agents help customers?
What’s the real problem with the workflow?
Preview of the solution
Historically, Capital One has placed the determination of whether or not a transaction is fraud or a dispute on a customer agent with minimal guidance in the internal filing system. Agents rely on training materials and their own judgement to determine the whether a transaction is fraud or a dispute leaving room for errors. Well-meaning agents (and customers) make mistakes that result in high rates of false fraud, operational costs, fraud losses, and negative customer experiences.
Timeline
Nine months
Skip to:
OVERVIEW
Lead Product Designer
User Research
Sketching & Ideation
Interaction Design
Collaboration
Product, Tech Lead, Data Scientist, Content Designer
Deliverable
Internal platform, Desktop
Our internal customer service tool, Empath, lacks ample guidance and tools to assist agents in their determination or fraud vs disputes.
The challenge here was to build a workflow that guides the agent to treat each transaction uniquely. We wanted to create an experience that was a diagnosis and less of a guessing game.
My role
DISCOVERY
What are our biggest user pain points?
How might we approach this work with the vision state in mind?
The process of filing a fraud or dispute claim was two unique work flows that were accessed through two separate entry points on the platform. This means the agent is determining whether or not a claim is fraud or a dispute based on their own knowledge and guesstimate. In order to prevent human error, we needed to find a way for our system to guide the agent rather than having them determine a claim on their own.
Current experience
Although the original request was to find a way to build one workflow, I wanted to lean into the vision state goal. I looked at the approach as a continuous iterative process that would span over many pilots.
Empath, is a partner in servicing, not a leader. To create a harmonious long term solution, I wanted to approach this from the perspective of creating Empath as the guide, not the assistance in getting agents to the right solution.
The Career Foundry’s Design Thinking Process inspired by Stanford’s d.school
Observing for inspiration
To get to the core of the issue, we needed to marry past agent observations with current agent conversations. This meant talking to agents directly and also watching them conduct fraud and disputes calls to fill in the gaps. We had the opportunity to be eye-witnesses with this internal tool.
Building trust with stakeholders
To set ourselves up for design success, I engaged our product partners, tech lead and associated dev team, and operations teams to be a part of the vision state conversation at the beginning. Their excitement and participation in picking apart research synthesis and more created a natural investment for them in the design process.
Deploy in pilots to constantly iterate
To give ourselves a safety net given the gargantuan amount of customer service agents at Capital One, we proposed 3 unique deployments rolling out the new experience to batches of agents at a time. This ultimately built confidence in the design result, but was successful in itself so just delayed the delivery of a better experience in the end to agents.
Before jumping into the design, we wanted to dive deeper into qualitative research to get a 360 view of the agent experience. Speaking with 11 agents , we focused on surfacing the end to end experience including typical and atypical customer scenarios.
We discovered 3 key insights from these conversations:
The system lacks sufficient guidance.
Data is not accessible to agents when they need it.
Data is not accessible to customers when they need it.
Agents shared that they’re lacking the context they need to be able to execute their tasks quickly and efficiently.
“I have to use all of my hacks to find what the right info. A new tab here, a new window there. It is tiresome and takes up time on the call… which makes me look bad as an agent.”
Agent interviews give us the clear picture
DEFINE
Building the vision state together
One Hallway
The concept for One Hallway was a visualization for our user flow. The hallway has one single entry point for all claims. All claims go down the same path until they need to exit out of a door from the hallway to be handled uniquely. We also toyed with the idea of an airplane but we don’t want any claims flying off the wings!
Working alongside our product team, operations lead, and tech lead, we came up with the below list of business outcomes to measure our success once every phase of One Hallway was out in the Empath world.
Define “success” for this effort
HOW MIGHT WE
guide agents to the right outcome for all customer claim circumstances?
DESIGN
Solving a true logic challenge
Just keep iterating
We created one logical flow including future enhancements in the vision state. AU Check is a system automated check that flagged a transaction made by an authorized user, not the account manager themselves. Additionally, a system check for recurring transactions flags potentially legitimate transactions like free trials or subscriptions that may clear up customer confusion.
Matrix Questions are the determining factor before we take them down the fraud intake path. If a customer doesn’t recognize a transaction and did not give the merchant their credit card information, then it is definitely fraud. The combination of these questions determines the next step like everything in One Hallway.
The check for potential false fraud was the gatekeeper to fraud intake. We found that if we were to build a manual checking experience for the MVP, we needed to ensure that it was very clear what should and should not be vocalized over the phone to make calls as efficient as possible.
Through usability testing, we learned that we needed to be more direct with instructions. We needed to create a clear difference in the UI between what should be vocalized and what can come up in conversation naturally. Agents are looking to the system to guide them, and every step of the flow needs to reflect that.
“I need the system to reflect how conversations actually happen so there are no awkward pauses and I can get things done fast.”
Additionally we consolidated the two entry points into one. We changed the language to Resolve a Transaction Problem to include dispute and fraud scenarios.
The MVP is out the door!
EVALUATE
The goal of this workflow was to execute a dynamic guided experience. Through our system checks for authorized user and recurring transaction, as well as our matrix questions that determined pre fraud or pre dispute intake, we built an experience that reacted to each input by the agent. While reacting it also acted as the leader in diagnosis, taking that pressure off of the agent.
So how did it perform?
Our MVP rolled out to 10% of US and Philippines agents for its first delivery phase. We received positive feedback and a few notes for the MVP+1 and iterations we could make before rolling it out to 100% of agents.
97% believed that the workflow was helpful.
95% of agents prefer the new workflow to the previous way of filing.
91% said customers seem happy with the new workflow.
What’s next?
The transaction probing checklist is a manual experience that agents have to comb through. The ideal future state is to reduce the amount of these checklist items through false fraud detection via a machine learning model. The more we can lead and automate, the less heavy lifting and work for our user, the agent.
