About
Assurance IQ provides personalised guidance and makes finding and using medicare, health, life, motor and home insurances for millions of North Americans each year. It enjoys patronage of over 17 million customers and uses data science and machine learning to narrow down options for coverage.
Situation
Assurance IQ receives millions of website visitors monthly, generating hundreds of thousands of leads. These leads are directed to the sales contact centre, where thousands of licensed insurance agents assist them over a phone call in making decisions tailored to their individual needs.
However, this process has several challenges.
- With over 6,000 insurance plans to be familiar with, agents face uphill tasks in staying updated on plan details and specific terms and conditions that change every year in the insurance renewal season from September to December.
- The problem is more acute when the agents are relatively new or recently hired. New agents typically take around six months to reach the same level of effectiveness as experienced agents.
- Next comes the shortage of time and the multitasking involved. Agents need to efficiently engage with callers in under 30 minutes, quickly understand their needs from a vast array of plans, explain why a particular plan is recommended, highlight its benefits that align with the customer's needs, and ultimately close the sale convincingly. Delaying a decision often results in lost sales opportunities.
CTO Nick Howard enlisted the help of Alltius to develop an AI tool for their agents - which would assist insurance agents in identifying plans with specific benefits as the customer conversation progresses, facilitate comparisons of these benefits across various plans, and ultimately provide valuable insights. Essentially, as agents engage with customers, the AI companion would offer talking points about the plans, highlighting not only how they cater to the customer's needs but also showcasing additional benefits that may not have been explicitly mentioned or recognized by the customer.
Solution
A cross-functional team, comprising software engineers, product managers, designers, and data scientists from both Alltius and Assurance, was formed. The setup of Alltius assistant on the web app platform was swift and straightforward, with hundreds of medicare plans from various sources ingested and coached within minutes. In parallel, Alltius cleared a stringent info-sec review of the platform by the auditing team at Assurance.
The team utilised Alltius' playground and APIs to assess the form and accuracy of the responses. The pilot project was divided into three phases that extended over a period of 10 weeks.
Phase 1 : Customising skills (2-4 weeks): The goal was to assess and fine tune three specific assistant skills that would provide the maximum value to Assurance’s sales agents.
- Narrow the search : Distil top 5-10 relevant plans from thousands given the buyer is looking for addressing 3-5 specific needs and health issues. For example, the agent could ask "My client is a 60-year-old woman in Atlanta who utilises physical therapy, has glaucoma, and prioritises dental benefits. Which plan would best suit her?"
- Compare plans : Next, the assistant could give a comparison across benefits, inclusions, and exclusions across a narrowed choice of plans specific to the customer needs. For example, the agent could effectively ask, “Provide a concise explanation of how Plan X stacks up against Plan Y and Plan Z on the stated customer needs N1, N2, and N3.”
- Pitch a plan : Finally, when the caller is now leaning towards a particular plan, help the agent give a 30-second comprehensive pitch highlighting how it addresses not only the stated needs but also the unstated needs of the caller. “Given that my customer has asked queries Q1, Q2, Q3 and has the needs N1, N2, and N3, provide me with a 30-second pitch that can help me close the sale.”
- Build the grid : Each season, a dedicated team of data entry operators and agents meticulously reviewed thousands of plans to extract and tally specific information in response to standard questions from each plan. For example, “What is the copay % or amount for out of hospital coverage?” This process was time-consuming and required careful attention to detail. The goal was for the system to quickly generate a grid containing M pieces of information from N plans, which could then be validated by humans for accuracy and cross-referenced to the specific sections where the information was located. What used to take several months could now be completed in just an hour for generation and a day or two for validation.
Alltius built these customised skills in a couple weeks and made it available for quality testing.
Phase 2 : Quality testing (2 weeks): The teams focused on evaluating the quality and accuracy of responses, a crucial step in addressing trust issues given the vast amount of information in 6,000 plans, each with multiple pages. The question at hand was whether the AI assistant could provide accurate responses at scale with the specified skills. Assurance's team utilised Alltius' APIs to quickly generate thousands of responses for a sample set. Through collaborative fine-tuning, they achieved a precision rate of 100% and a recall rate of 96.2% (responses provided when relevant information is found in sources). This was more than acceptable to take this to the next level i.e. agent testing.
Phase 3 : Prototype validate and POC (2 weeks): The task force took these assistants with a functional prototype to a varied mix of agent trainers, experienced agents and new agents. In all, over 10 individuals contributed to the validation process and the response was an overwhelming vote of trust on the utility of the assistants for the sales process and agent productivity.
Results
The assistants returned valuable time to the agents, allowing them to focus on their core strength of listening to customers with empathy and fully comprehending their needs. This enabled the agents to provide undivided attention to customers, rather than spending time reviewing plan details individually and memorising offerings.
Seasoned sales agents praised the assistants, stating, "This is excellent! It phrases things perfectly. It takes 6 months to get new agents up to mid-level of productivity. These tools would get them there in a third of the time."
The estimated benefits of the assistant implementation include:
- Lift in conversion: By providing more personalised, needs-focused, and precise pitches during 30-minute phone calls, a 5-15% increase in call-to-sale conversion rates was projected, enhancing the overall buyer experience.
- New agent ramp-up time: With onboarding time reduced by 67%, new agents could reach mid-level productivity three times faster, resulting in a 20% improvement in overall productivity over an average two-year tenure.
- Tech man-month savings: The build-the-grid tooling would save 2-4 engineer months annually that would have otherwise been spent on information sifting tasks across thousands of plans.
Way forward
WIth the promising results early on, the sponsors green-lit other projects to reconsider after the end-of-year insurance selling season. The first use case was auto-detection of buyer needs as the assistants make sense of the telephonic conversations. Next, would be to summarise the buyer needs and log across conversations so that different agents would instantly get context of past and present discussions. Additionally, the assistants could evolve into a knowledge navigator to get the best information, instantly and accurately across product, legal and operations manuals and policy wordings.
