Cloud-Based Machine Learning Tools Support Businesses’ Customer Service Process

How ML and AI Alert FedEx to Problems Early

FedEx has applied advanced machine learning across its business operations, implementing tools from Microsoft Azure such as Azure Databricks, Azure Machine Learning and Azure Data Factory, as well outside solutions that integrate easily with Azure, such as GitHub.

Anthony Norris, senior vice president of IT for global platforms and customer solutions at FedEx Services, explains that machine learning has taken its data intelligence to the next level, serving as a kind of canary in the coal mine to warn leaders about systemic challenges or problems with individual deliveries. “Machine learning can help us predict whether a package is going off track and we need to intercede,” Norris says. “It’s a huge value proposition to our customers.”

At a higher level, the system helps FedEx uncover broader challenges. For example, during the 2022 holiday period, issues with deliveries requiring customer signatures accounted for 25 percent of calls to FedEx’s call center, Gibson says, even though only 2 percent of its deliveries require such signatures. “That showed us that something wasn’t right with that experience, and we have teams who are able to take that insight and improve our operations,” he says.

In fact, FedEx has been inspired to create internal teams focused on a whole host of issues based on the data it has gleaned from machine learning. One focuses on solving challenges with time-sensitive pharmaceutical deliveries, for example. “AI has helped us develop what we call a package fingerprint,” says Gibson. “Based on that data, we created a monitoring and intervention team and dedicated dashboards to track high-priority packages.”

Before machine learning, FedEx’s method for gathering data intelligence was “rules-driven,” Norris says. “But when you have a complicated network, it’s hard to layer in rules that can address every single scenario.”

EXPLORE: How will AI affect cybersecurity in coming years?

Machine learning models can adapt to dynamic situations in real time and apply what it already knows to the matter at hand, allowing the company to put more intelligence into models. “For example, now the models can alert us and let us know what and when we need to communicate with customers. It’s become much more accurate, much earlier,” he says.

Norris and Gibson agree that when FedEx has better reliability and visibility, the customer benefits.

“The benefits of machine learning are multifaceted,” Norris says. “We can identify root causes of issues and opportunities for automation. AI and machine learning are really helping us do that.”

Natural Language Processing Drives Customer Engagement

Thanks to AI and machine learning, customers at Washington Federal Bank, a Seattle-based institution commonly known as WaFd, are glimpsing a future that isn’t yet available at the national level.

When former CTO Dustin Hubbard arrived at the company several years ago, his goal was to modernize how customers interact with the bank, particularly the call center. “I set out to rethink what should happen when people call in,” Hubbard says. “How do you flatten the experience, or quickly triage? I wanted the client to feel a very low amount of friction and get to the right person quickly.”

One way WaFd accomplishes this is through natural language tools such as Amazon Lex, which uses AI to transcribe spoken language, and Amazon Polly, a text-to-speech application that provides callers with a welcoming, natural-sounding automated phone voice. It’s all built on the Amazon Web Services cloud. The company also uses machine learning to analyze customer calls. The system gives agents real-time insight on why someone is calling, as well as the caller’s sentiment: Is the customer annoyed? Angry? Frustrated?

READ MORE: How the cloud has revolutionized how businesses run their contact centers.

“It’s good information for agents so that they don’t come into a conversation cold,” Hubbard says. “If a customer is angry with a chatbot before getting a person on the phone, the agent will know that and be prepared to handle that situation.”

One novel machine learning tool is WaFd’s voice authentication service for customers. Instead of having to answer security questions on the phone, bank clients can sign up for voiceprint recognition. It’s so reliable that it can distinguish a real, live voice from a fraudulent voice recording. Customers are adopting this service quickly.

“Last week, for the first time, more than half of our callers were using the voice recognition technology,” says Hubbard. “For the customers, it reduces the time it takes to complete a transaction by 90 percent. On the labor side, it lowers call volumes to the center itself, so that people who do need to talk to agents have shorter wait times.”

How Machine Learning Helps Business Find Data

Archive360’s machine learning journey has been a long one: It started more than 10 years ago to help customers more easily and efficiently search their own archival data.

“We’ve been using machine learning in this space ever since it was possible,” says Glenn Luft, vice president of engineering at the New York-based company. “Back then, it was about character recognition, creating transcripts from video and generating text from speech — anything we could do to extract meaning and content out of archival materials.”

The shift to the Azure cloud has created even more possibilities and cost savings for customers, who can search everything from legal files to videos in different languages. Tools such as Azure Video Indexer and Azure Cognitive Search work easily alongside solutions gathered from open-source AI marketplaces, and the machine learning component allows customers to provide feedback to continuously improve search results.

“Azure Video Indexer provides very good accuracy on the contents of a transcript. But my favorite feature is that it gives you a confidence score,” says Luft. “Human users can review that score and rate how relevant they think the results are. Machine learning watches those corrections and learns, and hopefully is more accurate next time.”