In an effort to protect revenue, even the largest global systemically important financial institutions (G-SIFIs) are investing in their systems and teams to offer exceptional and seamless customer service across all channels. In my role at Snorkel AI, I recently had the pleasure of working on a project with one of the largest U.S. financial institutions to revolutionize how their call centers, perhaps the most information-rich channel, are informing customer experience management.
I recently spoke about this project with Matt Casey, Snorkel AI’s data science content lead. You can watch our conversation (embedded below), but I’ve also summarized the main points here.
Call centers form the heart of customer service for many large organizations. Understanding this, leaders in many industries have a history of applying AI to customer call center engagements to improve the experience through predictive call routing, post-call sentiment analysis, and more.
While these efforts have had a tangible impact, corporations have struggled to apply these tools to the high-value application of real-time customer experience analytics. Where these systems exist, they rely on manual human labeling and focus on a narrow set of experience types, leaving them expensive to maintain and sluggish to adapt during times of dynamic change—precisely when organizational agility is most valuable.
Expanding the breadth of experiences their AI automation system could identify would enable companies to spot emerging trends as early as possible. This could enable leadership teams to launch initiatives to drive down expenses by understanding the root cause of customer behavior. They could also prioritize product and service efforts to enhance customer experience.
However, building an AI-powered system to uncover customer intent requires high-quality labeled data—a lot of it. To obtain this labeled data, employees must manually review each conversation, imposing a linear economic cost curve on the effort. Typically, shrewd assessments of the economic realities dictate that coverage can only be maintained for a portion of customer intents, leaving blind spots for emerging trends to go unnoticed. Furthermore, time-to-value of the AI-powered intent system is lengthy, giving it a profile more akin to a capital investment than a nimble experiment. Worst of all, an AI-powered intent system built on data that took nine months to refine can deteriorate rapidly as the competitive environment changes.
Despite these challenges, large enterprises regularly invest in AI-powered customer intent systems because they are powerful customer experience management tools.
What if those economic constraints were elevated? Could you deliver a leap in customer experience if you weren’t forced to live with blind spots? What would the ROI profile of your AI-powered intent system look like if you could benefit from economies of scale when refining your data? How much easier would it be to secure funding if it took weeks instead of months to refine your data?
For this G-SIFI, the prospect was too good to pass up. Here is how Snorkel Flow’s core capabilities created a new economic reality for customer experience management.
With Snorkel Flow, we deployed programmatic labeling functions across hundreds of intents. These labeling functions ranged from simple keyword-based functions to sophisticated sentiment analysis and other model-based approaches. Snorkel Flow intelligently combines and denoises these signals into a more powerful signal.
Snorkel Flow’s labeling function approach also allows users flexibility with their labeling schema. Subject matter experts often discover a mismatch between a project’s schema and its data over the course of an initiative. With traditional manual labeling, this kind of discovery means restarting the labeling process from zero or accepting the mismatch.
In Snorkel Flow, users can simply add, remove, or modify a handful of labeling functions and update their entire data set in as little as a few hours.
This flexibility is also helpful when the AI-powered intent system needs updating. As business needs, competitive atmosphere, or broader external factors change, Snorkel Flow users can return to the platform and modify a subset of labeling functions as needed.
Successful machine learning projects hinge on effective collaboration between data scientists and domain experts.
In this case, the institution’s operational teams had years of experience and deep insights into what their customers hoped to accomplish when they called into a call center. Through Snorkel Flow, we empowered them to not only label transcripts but to explain the reasoning behind their labels. Data scientists then turn these explanations into labeling functions that Snorkel Flow applies to hundreds or thousands of records, breaking out from the linear economic cost curve of prior approaches.
In many instances, Snorkel Flow’s user interface allows analytically savvy SMEs to directly contribute their knowledge without needing deep technical expertise, further accelerating time-to-value.
Let’s break down the implementation process and the significant outcomes that followed.
Using Snorkel Flow, we created a system to automatically classify customer engagements into intents. This system encompassed a variety of labeling functions:
Importantly, these signals are not used directly by the final AI-powered intent system. Instead, these signals are indirectly represented by the labels, which are used to train the model that the bank deployed.
Non-programmatic labeling approaches struggle to keep up with changing contexts. For instance, customer concerns during the pandemic were starkly different from pre-pandemic times. While that’s an extreme example, enterprises constantly cope with emerging trends—from extreme weather to regulatory changes to new identity-theft attack vectors.
With Snorkel Flow, our client’s AI-powered intent system could adapt much more quickly to new trends. Not only does this allow for efficient maintenance, it also enables teams to easily peer back in time to assess how far back a trend emerged and conduct scenario planning for customer experience initiatives.
We recently helped deploy the first version of the enhanced AI-powered customer intent system. The system is already improving the operational efficiency of the intent system and routing freshly minted intent data to product and services teams who will leverage it to reduce call center expenses while improving customer experience.
With Snorkel Flow’s continuous updates and new features, the institution is prepared to iterate on this initial version, refining and expanding its capabilities. Furthermore, the insights gained from call data will not only improve call center operations but also inform broader business strategies.
The success of the call center initiative represents step one in a multi-year plan. Other teams within the bank are exploring similar methodologies for various customer interaction points. Having a consistent and scalable solution across different customer touchpoints ensures cohesive and comprehensive insights. This holistic view of customer intent promises to provide the financial institution with a powerful new capability to provide a frictionless and more personalized customer experience.
This case study showcases machine learning’s potential to simultaneously transform operational efficiency while also delivering a big leap in customer experience. By leveraging Snorkel Flow, this global systemically important financial institution successfully addressed the scalability, time-to-market, and obsolescence challenges that customer intent identification initiatives face.
As enterprises continue to embrace digital transformation, the ability to quickly adapt to changing environments while maintaining high service standards will be a key differentiator. At Snorkel AI, we are proud to contribute to our clients’ successes by providing innovative solutions to help them realize the full benefit of AI.
Deploy production AI and ML applications 10-100x faster with Snorkel’s experts, using our proprietary technology.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。