How Stuller turns lost sales into product opportunities and brings QA coverage to 100% with Level AI

For more than 50 years, Stuller has been at the heart of the jewelry industry, supporting jewelry professionals around the world as a leading manufacturer, supplier, and technology developer. Led by a desire to serve, Stuller strives to offer an unmatched selection of quality products, expert and friendly service, and experiences that are easy, fast, and reliable. 

For Stuller's Customer Care leadership, reviewing individual customer interactions and relying on limited reporting was becoming increasingly time-consuming and restrictive. The company's previous quality management platform lacked the conversation analysis capabilities needed to investigate recurring issues across thousands of customer interactions. Supervisors had limited evidence for coaching decisions, while valuable customer feedback remained buried in individual customer communications.

Stuller turned to Level AI to connect quality assurance, customer insights, and agent activity, giving its teams a way to investigate recurring opportunities for improvement and bring customer feedback into product and operational decisions.

Disconnected systems and manual sampling hide systemic friction

Quality assurance carries significant meaning for Stuller and its frontline employees. Their mission is to S.E.R.V.E., and they did not have a way to objectively manage that expectation in the previous platform for all their 800,000+ interactions. Agent evaluations influence the company brand and performance, yet supervisors had only a small sample of conversations on which to measure performance.

Before Level AI rollout in March 2026, manual QA covered just 0.3% of eligible interactions. Each review took an average of 26 minutes and 16 seconds, making it difficult to increase coverage through manual effort alone.

That limited evidence made individual performance difficult to assess. An agent could be cited for poor empathy without a clear record of which behavior was missing. Another might repeat a customer's order correctly while entering the wrong product number into the system, an error that listening to the conversation alone could miss.

The consequences were particularly serious in Stuller's ordering environment, where a single incorrect digit could lead to the wrong product being fulfilled. Shatoya, a member of Stuller's quality assurance team, explains the difficulty of investigating these mistakes: "We have had interactions where we needed to do further research, and it was not as easy to uncover the root cause of an issue."

The underlying information was spread across Five9 for telephony, Salesforce for customer records, and Verint and UKG for personnel documentation. Coaching records also required manual documentation, adding administrative work for supervisors.

The lack of conversation analysis also affected business decisions. Customer complaints about product availability, delayed orders, and billing issues were scattered across individual interactions and proactive survey responses. Leadership had no consistent way to determine which concerns reflected recurring issues or how widely customers were affected. There was clear intuition among the subject matter experts, but that needed to be grounded with data so they would know where to start first.

Stuller connects customer insights, quality scoring, and agent activity through Level AI

Stuller began implementing Level AI in February 2026, integrating with Five9 and its existing Salesforce workflows. Over the following months, the company expanded its use of automated Quality Assurance, Voice of Customer, AI Workers, and Screen Recording.

AI Workers uncover lost revenue opportunities and inform product decisions

Angela, a Merchandising Product Manager at Stuller, began using Level AI Workers to investigate why customers who requested specific jewelry products were ultimately declining to purchase.

Angela noticed that customers kept asking for a particular series in a setting Stuller had previously considered impractical to produce. Individually, these requests were recorded as lost sales. When she examined them together using AI Workers, they revealed consistent demand for a product the business had overlooked.

The team confirmed that the design could be produced and began including the series in its holiday lineup. The findings also informed inventory planning, with customer demand providing evidence for stocking the new product. "We already know customers are going to buy it, so we can invest more stock. It makes that whole process easier," Angela explains.

Another successful use case occurred when Stuller made a business decision to modernize the checkout experience on Stuller.com. Customer feedback began reaching leadership. Meghan, Vice President of Customer Experience and Logistics, wanted to understand whether those concerns reflected a widespread customer problem or the experiences of a smaller group.

She used Level AI Workers to identify and examine over 300 conversations about the new checkout experience. The findings told a different story: approximately 70% of customers had adjusted to the change and simply needed guidance navigating the process. A smaller group had encountered specific problems that required attention.

The analysis gave Meghan two things: a more complete picture of customer reactions, helping the team focus on genuine issues rather than treating every report as evidence of a widespread problem; and a new approach to communications planning when a larger change is happening, allowing the leadership team to lead change both internally and externally with customers. 

Beyond individual investigations using the AI Workers, Stuller has used the Product Gaps Worker to identify just under 200 unique product enhancement requests from customer conversations. These findings gave product teams a consolidated view of customer needs that would otherwise remain scattered across individual interaction records.

Voice of Customer identifies the friction behind contact volume

Allison, the lead of Stuller's Voice of the Customer Program, uses Level AI to investigate recurring issues affecting customers. She examines conversation trends and prepares weekly, monthly, and quarterly findings for leadership.

The analysis revealed lots of smaller issues that amounted to the same root cause in addition to areas we need to focus on to create a more exceptional experience for our customers. But it also allowed us to group our conversations into recurring themes, showing us that there are consistent conversations that can be handled in a self-service model on Stuller.com. It just wasn’t clear to the customer. An example of this is when a customer needs to “Add to an order”. Online, this simply requires the customer to place another order, and they will automatically group. However, in today’s experiential world, there is much clarity around “Adding” to an existing order versus placing a “New” order. As a result, there were approximately 19,000 conversations that led to this topic. Because of this, the Stuller UI/UX Development team went to work and created a mock-up of an additional function that provides more clarity on Stuller.com when you need to “Add to an Order”. It was simple, but it created the perception that communication was required to complete this action. With this change, the Customer Care team will be able to focus on more complex service needs.

These findings gave Stuller a clearer understanding of where customers encountered difficulties, including problems that required coordination across multiple departments. The team used the findings to investigate opportunities for clearer communication, earlier notifications about delays, and more consistent handling of complex orders.

InstaScore identifies the behaviors that matter in agent coaching

Stuller's QA team also wanted to replace broad performance judgments with evidence about what agents did during conversations. With Level AI’s Auto-QA, the team could evaluate 100% of agents' conversations on specific rubrics, covering acknowledgment, validation, reassurance, and other behaviors associated with empathy and call control. 

The team could now identify the exact behaviors an agent needed to improve. Instead of broadly instructing employees to demonstrate empathy, supervisors could focus training on whether agents acknowledged a customer's frustration, offered reassurance, or communicated understanding. Following targeted training, Britany, an Operations team member who specializes in QA, reported that these specific soft-skill behaviors, previously observed in approximately 1 in 10 calls, were appearing in 1 in 3–4 calls.

The team also continued their calibration journey and developed separate evaluation criteria for conversations involving both English and Spanish. The broader plan was to use automated scoring for most routine evaluations while preserving manual reviews for critical tasks requiring human verification. This has been a game changer for the QA team at Stuller to pinpoint areas of focus and support the Learning and Development efforts instead of having to be stuck with manual QA all day, every day.

Stuller improves QA performance and brings customer feedback into business product decisions

Within months of implementation, Stuller recorded measurable progress across quality assurance and customer analysis. The key areas of improvement included: 

• 27% shorter manual evaluations: Average evaluation time declined from 26 minutes and 16 seconds in March to about 19 minutes across April and May.

• New holiday product opportunity: Analysis of lost sales revealed recurring demand for multiple series that Stuller subsequently confirmed it could manufacture and began adding to its holiday lineup.

• 85.93% average InstaScore: Quality scores rose from 79.26% in February to 85.93% in June, an increase of 6.67 percentage points because of being able to guide agents on how to give a better experience.

• 99 high-performing agents: By June, 99 agents had average scores of at least 85%.

• More Seamless Business Processes: Conversation Analytics supported the addition of three new processes Stuller.com, allowing the customer to self-serve. These will be put into 2027 technology requests during their business planning season.

• 186 product enhancement requests: AI Workers identified recurring needs for product teams to investigate.