As 2024 comes to a close, I want to reflect on how Pindrop has innovated on behalf of our customers. In just one year, we delivered award-winning deepfake detection, scaled to analyzing +130M calls in real-time, and improved our industry-leading fraud detection accuracy by +13%. We also increased the number of calls authenticated by our product without any agent in the loop by +45%. Our teams launched 17 new capabilities, delivering an innovative product enhancement every 11 business days, on average.
Delivering deepfake detection to market—raising the bar for contact center security with cutting-edge technology to meet evolving fraud challenges.
In 2024, AI-generated synthetic content crossed the ‘uncanny valley’ – a concept that refers to the feeling of unease we experience when encountering AI that seems almost human-like—and fraudsters began misusing AI voice at scale to conduct impersonation scams, especially in the phone channel.
Deepfake detection for the call center
Customers, especially those who rely on voice analysis, must fortify their authentication approaches with protection against synthetic audio.
In response to this customer need, we launched our Pindrop® Pulse™ technology in February, our deepfake detection product focused on call centers. In real-time, Pulse detects audio deepfakes, voice clones, and replay attacks in contact centers, setting a new standard for contact center security. Just two months after launching Pulse, Pindrop was named the winner of the FTC voice cloning challenge in the large organization category, reinforcing the importance of this technology in protecting from the misuse of AI-enabled voice cloning for fraud. Since launch, we have continued to improve the accuracy of our deepfake detection engine, releasing 20 machine learning model enhancements in 2024, all addressing real-world challenges faced by our customers, such as removing bias associated with gender and speech accents, having multiple speakers on a call, background noise, distortions from use of speaker phones, and more. As of November, our customers are using Pulse to assess liveness of +130M calls on an annualized basis, the largest volume of calls analyzed in the industry today.
Today, Pindrop has some of the largest financial institutions using “liveness detection” as part of their multi-factor authentication and fraud approach to security.
Deepfake protection and tracing the source
In August, we announced Pindrop® Pulse™ Inspect, a deepfake product that helps organizations maintain content integrity by identifying AI-generated audio with up to 99% accuracy. Inside Pulse Inspect, we launched specialized ML models designed for high-fidelity digital audio, and lower-fidelity phone audio. We are proud to have TrueMedia.org as a collaborative partner, which uses Pulse Inspect to manage misinformation in elections worldwide, including U.S. Presidential elections. Our research on tracing the AI system that creates synthetic audio, a key component of Pulse Inspect, was accepted at Interspeech 2024. This capability to “trace the source” can help organizations like law enforcement or enterprise security teams to identify the attacker(s) behind an attack. This capability is currently available in preview to select customers.
Additionally, the solution offers the Pindrop® Pulse™ Inspect API, which enables customers to integrate our deepfake detection technology seamlessly into their workflows. This flexibility allows for both programmatic access and direct interaction through our user-friendly interface. Pulse Inspect API can analyze audio in real-time and is designed for high-volume detection use cases (e.g., detecting Grandparent scams on consumer phones, AI-generated content on social media platforms, and content streaming sites).
Continuously Improve Accuracy for Better Authentication and Fraud Detection
At Pindrop, accuracy is a job that is never fully done. This year, our team focused on delivering continuous improvements to the accuracy of our authentication and fraud products. This commitment to improvement helps keep our products at the forefront of the industry, providing our customers with the most reliable and effective solutions.
Enhanced search to simplify data analysis for fraud teams
This year, we added an enhanced search capability to assist fraud and authentication
teams with policy management and data analysis, simplifying their ability to get critical work done faster. For example, fraud teams can quickly search for information that is important to their organization and generate reporting directly from those results.
Specifically, enhanced search enables teams to do things like selecting multiple criteria and operators for a desired search request, customize search result columns, sort search results by column headings, export search results via a CSV file (up to 10k rows), and more.
Enhanced search tackles tedious tasks, streamlining workflows and keeping your team focused on their top priority: fighting fraud.
Continuous Scoring to stay ahead of evolving fraud threats with more accurate, efficient, and continuous protection
For customers who seek enhanced fraud protection, Pindrop® Protect Continuous Scoring offers a superior “fraud look-back” capability that can detect up to 22% more fraud while helping prevent up to an estimated $5.1M in fraud losses2 annually, based on customer-specific implementations.
Pindrop® Protect’s AI-driven, real-time Continuous Scoring module leverages industry-leading Pindrop® Protect technology to automatically rescore historical calls every 5 seconds, creating new fraud cases with 90% accuracy, low false positives, and a low case review ratio. Continuous Scoring is integrated into a customer’s existing user interface and reporting mechanisms and fits seamlessly within their existing fraud case policies.
This cycle of learning and re-examining makes the Continuous Scoring process self-optimizing, improving your ability to continuously detect fraud from a ‘look back’ process.
Audio-Based Spoof Detection from a single keypress
Audio-based spoof detection enables a core solution, Pindrop® Passport, to determine whether a call is spoofed with a single keypress. This new spoof detection method removes reliance on call metadata or the data collected about a phone call, such as the caller’s phone number, call start time, duration, etc., which many customers struggle to acquire consistently (or at all) from their telephony carrier.
This new spoof detection method is one in a series of improvements to caller ID validation for Passport released in 2024 that work together alongside existing spoof detection techniques. With multiple factors to determine caller ID validation, Passport increases accuracy for existing customers and broadens the pool of calls eligible for scoring, including those on less-prominent carriers and customers outside of the U.S. Combined, these factors have proven effective in identifying emerging spoofing methods observed in specific call center scenarios.
Streamlined Integrations to Accelerate Customer Value
At Pindrop, we understand our solutions work as part of our customers’ tech stack to improve the customer experience, increase customer identity capabilities, and streamline fraud operations. Our top priority in 2024 was to make it easier for customers to deploy Pindrop solutions in their environments to get more value from their investment.
This year, we launched integrations to capture call traffic in real-time from the leading Call Center as a Service (CCaaS) Platforms. For example, customers using Genesys® AudioHook can now easily configure their Genesys environments to send call traffic in real-time to the Pindrop solution without depending on their telephony network teams. Customers using NICE and Five 9® can now integrate with Pindrop solutions using the SIPREC protocol from within the admin consoles of these systems. We realize that not all of our customers are using CCaaS platforms. For such customers, we launched new integrations with telecom providers such as Lumen, New Era, and Masergy to provide pre-built integration options for real-time call data.
Additionally, we launched an embeddable user interface widget (Agent Screen Pop) for serving real-time notifications to agents and collecting their call feedback, which can be integrated within the UI of their most commonly used CCaaS tools. This year, we released multiple iterations of our Agent Screen Pop to expand coverage for customers using Genesys, NICE/CXOne, and Amazon Connect. In November, we launched a Case Management API for our customers to help them pull fraud case data from the Pindrop solution to any enterprise case manager of their choice, especially for customers who have omnichannel fraud management teams using enterprise-wide case managers.
2024 marked a year where patience met precision in our product strategy. From launching Pulse at the start of the year – now processing +130M calls monthly – to enhancing our core ML models for +13% better fraud detection, authenticating 840M calls 45% higher than 2023, to streamlining integrations across major CCaaS platforms, we’ve proven that thoughtful innovation beats speed-to-market. Our approach at Pindrop isn’t just about building new features but solving real customer problems through deep technical expertise and customer collaboration. Thank you to our customers, partners, and to our internal product, research and engineering, and many other teams who made these advancements possible.
Delivering customer delight is a journey, and we are excited about the next stage of this journey, making further progress in areas where our customers expect and deserve a superior experience. As we look to 2025, we’re not just keeping pace with the evolving security landscape – we’re defining it.
Interested in learning more about how we innovate for our customers? Schedule a demo today.
1Derived from beta of large national bank – alert rate of rescored calls that were fraudulent compared to alert rate of original fraud calls
2Fraud uplift rate extrapolated from beta findings that were then applied to other Pindrop Protect customer call volume with industry standard assumptions of fraud loss detection and average loss to arrive at the potential savings for a customer with a large call volume
