Data scientists and agencies must work together to get the best results
Analyzing the performance of brokers and agencies requires more targeted efforts to identify individualized areas for improvement rather than using more generalized data. For the best results, data scientists need to collaborate closely with other areas of a company to ensure that the correct business insights are present in the final product.
This is according to Justin Milam, associate director at Willis Towers Watson (WTW), who said that “analytics of broker performance need to be more specific and you need to ask the right questions to get actionable insights from them.”
In a conversation with Insurance Business, Milam detailed what modeling techniques can be adopted to further understand the business, the types of questions to ask to get useful insights, and what challenges can arise when working with a data scientist.
Surpassing previous modeling systems
Traditionally, the heuristic approach to analyzing a company and its employees has proven to be an accessible means of obtaining more immediate and digestible information on loss rates, number of deals signing up, conversion rates and other information.
“This data would then be used to determine bonuses, if an agent needs an audit, or if additional training is required to optimize and increase productivity, among other things,” Milam said.
“When starting out, some simple one- or two-way interactions may be the most appropriate model to build until you feel comfortable with the methodology,” Milam said, emphasizing the usefulness of the heuristic approach as a transition to more sophisticated measures.
These more advanced modeling techniques include a generalized linear model (GLM), in which the target variable is the loss rate, or a non-parameterized gradient-boosting machine (GBM).
Milam recommends overlapping the GLM and GBM techniques to recover data that may be lost or unaccounted for with each separate process.
Using these methods can provide a more nuanced view of a company’s current book and what can be changed for future growth opportunities.
“In the models, you can see if your agents are writing multiple lines of business, the credit ratings of that business, past claims, as well as how an agent’s profile can determine whether or not they will be successful.” milam said.
“You can also look at changes in the business over time. For independent agencies, there could be challenges where if a particular company enters or exits the market, they could really see their line of business change.”
“Taking a common starting point to understand what a company wants to achieve is key, especially if new systems or techniques are being incorporated,” Milam said.
For example, if a company is trying to find out if a recent hire can meet its standards, a barometer for success must be clearly defined. Whether it’s a low churn rate, a high conversion rate, the number of deals written, or the likelihood of longevity within a given company, each of these will affect a statistical analysis and produce variable results.
Being able to work with a data scientist to outline a narrower analytics framework will help generate targeted insights and won’t risk impacting employees or lines of business that may not be relevant.
Recognize the challenges of updated analytics
A 2021 WTW report found that only 10% of companies were using advanced analytics in the management of their agencies or brokers, pointing to widespread skepticism toward data science and a confirmation of the “if it ain’t broke, it’s not broke” philosophy. don’t fix it.” ”
“The maturation of analytical culture is not something that many want to embrace with open arms, so it is crucial to facilitate it in a way that doesn’t seem sinister,” said Milam.
“You really need to break down data silos to present relevant information. You want to make sure that the data you are using is what the agents are looking for. If you come up with some calculation for a loss ratio that the agency uses for other diagnoses, there will be skepticism around using that.”
Presenting the findings is just as important, as many runners may not be as receptive to an Excel spreadsheet as a data scientist and may find an infographic or pie chart more accessible.
However, certain business considerations may override the use of a model as the data scientist may have initially envisioned. For example, “if a model is built to reduce binding authority for agents with high projected loss rates, it may be difficult to gain buy-in from field executives and general agents if the producer has historically had a low loss rate.” “Milam said. .
While this can be frustrating for some data scientists, it is worth adding value in any capacity to a typically neglected area of analytics under any circumstances.
Keep up with the latest news and events
Join our mailing list, it’s free!