One of the first and most enduring things I learned during my onboarding into the world of insurance is that underwriting is half art, half science, and all nuance. If tools within the underwriter experience don’t align with this precarious balance, they won’t work.
User research
Design
Testing
QA
1 product manager
1 product designer 👋🏻
3 engineers
2023
Pricing a cyber policy is not so simple as choosing a number. There are no less than 10 numerical modifiers applied to a base price to determine the final amount a client will pay. These data-driven risk modifiers are offset by a human judgment factor: a sliding discount called a credit which underwriters can apply to accounts they deem worthy.
Pricing complexity only deepens when a policy is going to renew. If account details haven’t changed, clients expect the price of their renewals to be flat (unchanged from last year). In most cases where price increases more than 10%, underwriters lose that business!
However, the way we price and assess risk changes in tandem with the cybersecurity landscape, which is constantly in flux. As such, prices very rarely stay naturally flat year over year and would need manual adjustment to remain viable.
Despite a long history of tension between the need to keep pricing consistent and a continually fluctuating pricing model, underwriters had no way to easily calculate the amount they must manually credit to reach a target price.
In order to refine our understanding of stakeholder and user concerns with the current state, we conducted 7 qualitative interviews with new business and renewal underwriters, as well as the Senior Vice President of cyber underwriting.
We asked a standard set of questions on their experience with pricing: Were there ever cases when they were trying to reach a very specific price and must adjust credits in order to do so? How often? What was their current approach to completing this task? Which aspects were painful, if any?
We learned the following:
In order to reach a flat price, underwriters were forced to repeatedly nudge the one pricing lever which they had control over - their sliding discount - to guess and check how changes in value would affect the final price of a policy, re-rating the account with each nudge! These nudges would not always make the price move in the way one would expect because of all the other complicated modifiers at play to produce the final price.
The result was an average of 10 minutes per account spent nudging the sliding discount, often 6-8 decimal places deep, and continued re-rating to finally reach a desired premium amount.
Surely there must be a better way to structure the experience of this task!
One of the most important times for underwriters to be able to price quickly is when they are on the phone with a broker trying to do business. Not being able to produce an answer on whether a desired price is possible or not creates unnecessary friction in the negotiation process which over time causes relational damage to a broker-underwriter partnership. In the long term, this results in notably less business, less won accounts, and a growing damage to gross written premium.
“Brokers care about the dollars versus the decimals… If I’m on the phone with a broker, they aren’t going to have time for me to play with these calculations”
Enter our Data Science (DS) team, the ones who keep our pricing engine tuned so the underwriters can use it to write business effectively.
Specifically, one member of our DS team did something extraordinary: he created a local tool to simulate rating calculations. Based on whichever inputs were filled in, the simulation could calculate what the other inputs must be in order for all the pieces of the pricing puzzle to fit together properly. As you might expect, once underwriters heard about this tool, they started going to him all the time to ask “how much do I need to credit in order to hit x price?”.
Problem solved! Just kidding, this workaround actually presented a number of challenges:
The overwhelming conclusion? Things were not great, but maybe we could improve them.
If you can't construct a corkboard with thumbtacks, photos, and yarn, an affinity map will do
Using Data Science’s simulation tool as a foundation, how might we improve the way underwriters experience the task of precision pricing?
What if we turned the action and result on their heads? Currently, underwriters could only input a credit value and then be returned a premium value. What if we created a new pathway for action, reversing that flow and updating the interface so that underwriters can input a desired premium amount and be returned the credit value necessary to reach it?
Based on our research, our benchmarks for success were that this feature should:
We also defined a leading indicator of success: if we did a good job, we could expect to see a dramatic reduction in average time-to-underwrite per submission.
Early concept sketches
Early designs considered both in-flow and modal solutions.
Modal iteration
In-flow iteration
Adding the tool to our current pricing panel would sustain stronger proximity of features in the interface area associated with the pricing task. The downside was that adding a new feature may also add undesirable complexity to an interface area that supports a high-stakes, high concentration task. In simpler terms, we wanted our tool to be available without throwing a user’s cognitive load out of balance.
In user and stakeholder check-ins, we received consistent feedback that pulling users out of the pricing flow would reduce adoption of our tool and potentially even break an UW’s train of thought, turning a helpful feature harmful. As such, we proceeded with our in-flow option, paying special attention to footprint size.
We also experimented in what the ideal output action would look like. At first we displayed the credit value necessary to hit a target premium amount but did not automatically update the amount.
A less assertive return value in the form of an alert
I was hesitant to change values without an additional layer of user confirmation, but also worried about input issues and other forms of human error when that responsibility was placed on the user.
Given the option between copying and pasting the value or automatically adjusting the credit value, users overwhelmingly preferred automatic adjustment.
Prior to build, in order to make sure the intended design would serve user needs, we conducted a round of moderated usability testing with the same 7 underwriters we had initially gathered qualitative feedback from.
We presented users with the prompt “Please credit this account so that the premium value is exactly $10,000”, after which they shared their screens and navigated through a prototype of our intended updates to the pricing flow.
All 7 respondents successfully navigated through the prototype and used the new tool to set the proper premium value without guidance. After completing the task at hand, our team received feedback that the new flow felt intuitive, would definitely help with phone conversations on pricing, and radically reduced underwriting time required for precise pricing.
With build complete and the feature safely in a staging environment, we conducted a final round of confirmation by letting users try the tool prior to launch into production. Being able to use the feature fully in the software as opposed to a somewhat limited prototype only magnified the positive feedback for this feature. One respondent was so excited she actually got a little misty, remarking that this tool bolstered confidence in her own ability to underwrite. This had us a little misty, too :-’)
After release, in order to determine this feature’s impact on the time-to-underwrite metric, we examined the conversion rate between our calculate action and the send quote action.
As a reminder, time to underwrite previously hovered around the 10 minute mark - our feature reduced this metric dramatically.
Why was pricing still taking longer in some cases? The fact that only 90% of submissions convert within an hour suggests increased complexity within some accounts. These “sticky accounts” may require more care, or even approval from underwriting management which would affect conversion time dramatically.
Having been in production for a year now, this feature remains one of the most consistently complimented when our team collects quarterly feedback to calculate our underwriter Net Promoter Score.
A few words from our NPS respondents:
There have been a few pricing updates that required us to patch the tool so it would continue to function properly, but outside of those moments, the main point of critique we see is that underwriters would like to see this feature expanded to other product lines. I sincerely hope we can implement this in the future!