AI Advances Answering Machine Detection
Artificial intelligence has infiltrated just about every component of the CRM technology stack, and outbound call centers are now starting to benefit from it as well. AI-driven answering machine detection (AMD) is now a reality, enabling contact center systems to detect with greater speed and accuracy whether outbound calls are picked up by voicemail and answering machines or people. New AI technologies are even making systems language-independent.
AI’s introduction in the AMD space is a recent occurrence, though AMD technology has been around almost as long as answering machines themselves. The earliest AMD iterations, which started to appear in the 1970s and 1980s, primarily focused on identifying patterns in recorded audio to differentiate between human voices and machine responses. Early AMD systems relied on simple audio analysis to detect silence or specific tones commonly found in answering machine greetings.
Over the decades since then, significant advancements have happened as technology evolved, starting with the integration of digital signal processing, which allowed for more sophisticated analysis of audio features like pitch, energy, and spectral characteristics, leading to improved accuracy.
But the systems still had problems. Back in 2008, Ofcom, the United Kingdom’s communications regulator with responsibility for the telecommunications, broadcasting, and postal industries, effectively banned the use of AMD technology due to the number of silent calls it caused.
Of course, since then, the technology has advanced substantially. In 2020, Aculab released AI-AMD, enhancing the traditional AMD system with a trained neural network. Many other contact center and communications platform vendors, including Twilio, Genesys, Sprinklr, Convoso, CallHub, LumenVox, Voxdesk, ByVoice, Five9, RingCentral, CallHippo, Regal.ai, CallTools, Vocalcom, and dozens of others, also offer AMD technology.
These newer, more modern AMD systems leverage AI, including machine learning, neural networks, and advanced analytics, to analyze audio patterns and classify responses as human or machine with much greater precision and speed. These same systems also incorporate advanced call progress analyzers, advanced speech recognition, and analytics and reporting tools that can help companies determine whether the phone numbers in their customer databases are useful, whether the phone numbers they are using have good reputations with callers (often, if consumers are unsure about incoming calls, they let them go to voicemail), and when their customers are available to receive calls.
AMD results can be delivered in near real time, with some vendors claiming that their systems can classify whether calls are answered by a human or their voicemail in less than two seconds with almost 100 percent accuracy.
Advanced algorithms that have emerged in just the past year or so can adapt to different accents, background noise, and variations in answering machine greetings. Modern systems can even further classify the recipient as residential or business depending on how the call is answered.
Standard answering machine detection algorithms use voicemail tone detection, which can be tested and refined using configurable time-windowing parameters.
“Now these systems analyze voice patterns, pauses, and other audio cues in real time, offering unprecedented accuracy and speed,” says Bobby Hakimi, chief product officer and cofounder of Convoso, a provider of outbound call center systems. “Deep learning models enhance detection by learning from vast datasets, while noise-filtering technologies improve performance in diverse environments. Some AMD tools now integrate with conversational AI, enabling the seamless delivery of prerecorded voicemail messages.”
Why Use AMD?
By knowing what’s happening on the other end of the call, outbound dialing teams can program their apps to hang up, play prerecorded tracks without connecting to an operator, leave customized messages, or redial the number later. This process allows agents to move on to the next call swiftly or ensures the intended message is delivered even in the absence of a live person. The goal is to optimize agent time and ensure effective communication.
Leveraging advanced algorithms to accurately detect whether a call is answered by a live person or picked up by voicemail, automated voicemail management enables agents to focus on productive conversations, reducing wasted time and effort, Hakimi says.
Advancements in voicemail detection now use machine learning algorithms trained on large datasets to distinguish between human voices and recordings, adds Iryna Melnyk, a content and SEO specialist at Jose Angelo Studios. “By analyzing pitch, tone, and cadence, these systems can accurately detect if a call reaches a live person or voicemail.”
Different types of answering machine detection technologies include rule-based systems, machine learning algorithms, and hybrid approaches combining both methods. In the past 18 months artificial intelligence has made this technology much more accurate in determining whether a human or machine/voicemail has answered a call.
Modern technology is also more accurate in providing feedback on false positives (mistaking a live person for an answering machine/voicemail) or false negatives (mistaking a machine/voicemail for a live person), according to Hakimi.
In fact, with AI, false negatives and false positives have decreased by at least 50 percent. Some vendors claim that with AI, the accuracy of their current systems is about 98 percent, compared to 40 percent for much older systems.
The more data that goes into training AI models and the more often that training is updated with new data, the better the accuracy, experts agree.
“At Convoso, we have a lot of data to tell us if we have detected a new type of answering machine or something strange, like calls picking up, then ringing, which shouldn’t happen, or something else, like a message that the number is no longer in service,” Hakimi says.
However, a CallHippo blog cautions: “The accuracy of AMD depends on several factors, including the quality of the detection algorithms and the complexity of the answering machine messages. Simple messages like ‘Hello, you have reached…’ are easier for AMD to detect accurately, while more complex messages or background noise can confuse the system.”
Likewise, automated voicemails, with their steady and robotic tone, are easier for AMDs to detect.
Another factor is speed. The quicker an AMD can detect an answering machine, the quicker an automated dialer can move on to the next call.
Experts are quick to point out, though, that AMD shouldn’t be too quick to hang up once it determines that a voicemail or answering machine picked up the call. One reason is that there are people who will pick up from voicemail mid-call (iPhones offer this capability).
According to Hakimi, most answering machines will usually pick up after four rings. At six seconds per ring, that’s a total of 24 seconds. Then the AMD needs a few seconds after that to determine that a human hasn’t answered.
“The amount of time before giving up on the call depends on what is going on with the answering machine [i.e., the type of response]. We typically wait up to 18 seconds before detecting if it’s voicemail and whether to hang up on the call.” The actual detection takes only a fraction of that time, Hakimi explains.
Sales Pressure Drives AMD Advances
“The evolution of AMD has been fueled by the growing pressure modern sales and marketing teams feel to connect with more customers with greater speed and efficiency while avoiding compliance risks,” Hakimi adds. “AMD helps modern call centers meet these challenges and is essential to their technology stacks.”
Melynk agrees: “The driving force behind these advances is the need to optimize productivity and increase conversion rates. Businesses are constantly seeking ways to improve their outbound dialing strategies, and the ability to accurately detect voicemails allows them to better allocate their resources and target their efforts toward live conversations. This ultimately leads to higher sales and revenue generation.”
Convoso offers the following example of the benefits of using AMD technology: If 10 agents make 100 calls per day at a 20 percent contact rate, they each make 20 contacts per day. If you have a 10 percent conversion rate at $100 per conversion, that’s $500,000 annually. If high-quality AMD increases your contact rate by just 10 percent, you’d bring in an extra $250,000 per year.
This 10 percent increase in contact rates is a conservative example, according to Convoso. Medigap Life, a Medicare insurance brokerage in Boca Raton, Fla., switched to Convoso’s outbound dialing solution with AMD and quickly saw contact rates increase up to 80 percent. Its more than 200 agents are being connected to three times fewer voicemails.
Senior Healthcare Advisors also saw a tremendous transformation. Before using Convoso’s AMD, agents had to disposition as many as 100,000 voicemails every day. With Convoso, the Medicare sales company doubled contact rates and boosted agent efficiency by 50 percent. Its agents are reaching 100 times fewer answering machines now compared to its previous system.
AMD delivers measurable impact for CRM, sales, and marketing, Melnyk adds. It increases contact rates by routing live calls to agents immediately while automating voicemail drops, which saves time and resources. Sales teams can focus on high-value leads, and marketing gains a more efficient pipeline. CRM data becomes cleaner and more actionable, improving lead management and nurturing processes.
AMD is not just a tool—it’s a catalyst for better productivity and compliance, which is why it is essential in today’s competitive outbound dialing landscape, according to Hakimi.
But those are not the only financial considerations with modern AI-powered AMD systems. “The AI has a cost associated to it, but [the cost] is significantly lower than having a dedicated agent listening in. The cost of the AI is pennies compared to what it would cost for a human body to listen in in real time,” Hakimi says.
Best Practices for Configuring AMD
If you want to leverage your answering machine detection system to its full potential, it’s essential to configure it correctly. Telynx recommends the following best practices:
Tune the parameters. As with any system, the default settings might need some tweaking to meet your call center’s specific needs. Adjust parameters like machine detection timeout and machine detection speech threshold based on the nature of your calls and your audience’s demographics.
Regularly test and refine. Call patterns and customer behaviors change over time. Regularly testing and refining your AMD system ensures optimal performance, minimizing errors and enhancing customer interactions.
Train your agents. Ensure your agents understand how AMD works and what to expect. This training will help them transition smoothly between calls and handle potential false positives or negatives.
Monitor false positives. Keep an eye on the number of calls incorrectly classified as answered by humans. If this number is high, you might need to adjust your settings or consider a different AMD solution.
Stay updated. AMD technology is rapidly advancing, especially systems based on AI and machine learning. Stay updated with the latest developments and be ready to upgrade or adjust your system as needed.
Respect regulations. Some jurisdictions have regulations concerning automated calls and voicemail drops. Ensure your use of AMD complies with all relevant laws and regulations.
AMD Compliance Concerns
The compliance aspect is extremely important in certain industries, including debt collection, medicine, financial services, and law, because regulations require that certain messages not be left on answering machines. In some jurisdictions, for example, leaving messages requires consent, and then companies leaving messages are required to state their company name and the toll-free number for people to call back, and to notify call recipients that they can be placed on the company’s do-not-call lists.
But challenges aside, it’s still worth it for companies to deploy AMD technologies in their call centers. Answering machines are not going away anytime soon. In fact, Tech Vortex Solutions recently valued the global market for answering machines at $1.5 billion in 2022 and expects it to reach $2 billion by 2030, growing at a compound annual rate of 4.5 percent.
Experts also expect that as more centers recognize the merits of AI-powered AMD systems, they are poised to become the gold standard. For contact centers that have already made the shift, the operational efficiencies and customer satisfaction gains are undeniable. For those that haven’t yet embraced the technology, the message is clear: AMD is not just an advantage; it’s going to be a business imperative.
Phillip Britt is a freelance writer based in the Chicago area. He can be reached at spenterprises1@comcast.net.