Genesys Cloud - Main

 View Only

Sign Up

  • 1.  AMD Accuracy

    Posted 10-27-2025 13:20

    Hello,

    I understand that AMD (Answering Machine Detection) has seen significant improvements, reportedly leveraging AI and machine learning.

    I am specifically trying to determine the stated accuracy rate for human connect identification using the Genesys dialer's AMD feature. Some competitors are advertising figures as high as 94%, but I have not been able to find comparable data or advertised statistics from Genesys.

    Any information or official figures regarding the percentage of accuracy for human connect identification with the Genesys AMD would be greatly appreciated.

    Thank you.


    #Outbound

    ------------------------------
    Paul Cane
    Dialler Manager
    ------------------------------


  • 2.  RE: AMD Accuracy
    Best Answer

    Posted 10-28-2025 15:03

    Hello Paul,

    I don't believe we have published statistics on our Answering Machine Detection accuracy, but we do have this article from the Resource Center that goes into a little bit of detail on how it works.

    I will see what I can find and will share any information that I can here.



    ------------------------------
    Jason Kleitz
    Online Community Manager/Moderator
    ------------------------------

    Original Message


  • 3.  RE: AMD Accuracy

    Posted 2 days ago
    Edited by Christopher Samuels 2 days ago

    Hi Paul/Jason,

    Did you find any additional information about AMD accuracy? I would be interested to know what the reported accuracy figures are too.

    Regards,

    Chris 



    ------------------------------
    Christopher Samuels
    x
    ------------------------------

    Original Message


  • 4.  RE: AMD Accuracy

    Posted 2 days ago

    Hi Christopher

    Yeah we got an official response last week from Genesys.  
    If you fancy a chat related to this and the tech we are using with Genesys feel free to reach out to me Paul.Cane@ovo.com.

    Response from Genesys:
    Genesys does not publish accuracy (or error) percentages related to Answering Machine Detection (AMD), for an important reason: performance can vary significantly by scenario, including:

    • Country and carrier
    • Connection quality (latency, jitter, and noise)
    • Voicemail types in use
    • Answering scripts from phone providers
    • Number of silences paused before recording
    • "Human vs machine" detection rate configured on the dialer

    In other words, any "generic" index would not faithfully represent its real environment.

    What Genesys recommends and makes available is the concept of policy and operational tuning, adjusting behavior according to false positive vs false negative tolerance:

    • If prioritizing not losing a human, accepts more voicemails dropping to agents
    • If prioritizing protecting agents from VMs, accepts small loss of human contacts

    In addition, there's an important technical point:
    ➡ AMD by Genesys can work with adaptive machine learning, learning the patterns of real calls in your environment over time.
    This continuous learning is what leads to better hit rates, rather than a fixed metric that would be misleading.

    Practical suggestion:

    Run a pilot for 2-4 weeks in real campaigns

    Measure the following metrics in Genesys:
    – Percentage of AMD_Human who entered as voicemail
    – Percentage of AMD_Machine who entered as human

    Adjust the aggressiveness of the policy according to the objective of the campaign



    ------------------------------
    Paul Cane
    Dialler Manager
    ------------------------------

    Original Message


  • 5.  RE: AMD Accuracy

    Posted 10-28-2025 15:38

    Apenas para complementar a ótima resposta do Jason:

    A Genesys não publica percentuais de precisão (ou erro) relacionados ao AMD (Answering Machine Detection), por um motivo importante: o desempenho pode variar significativamente conforme o cenário, incluindo:

    • País e operadora
    • Qualidade da conexão (latência, jitter e ruído)
    • Tipos de correio de voz em uso
    • Scripts de atendimento dos provedores de telefonia
    • Quantidade de silencios pausados antes da gravação
    • Taxa de detecção "human vs machine" configurada no discador

    Ou seja, qualquer índice "genérico" não representaria com fidelidade o seu ambiente real.

    O que a Genesys recomenda e disponibiliza é o conceito de política e tuning operacional, ajustando o comportamento conforme a tolerância a falso positivo vs falso negativo:

    • Se priorizar não perder um humano, aceita mais correios de voz caindo para agentes
    • Se priorizar proteger agentes de VMs, aceita pequena perda de contatos humanos

    Além disso, existe um ponto técnico importante:
    ➡ O AMD da Genesys pode trabalhar com Machine Learning adaptável, aprendendo os padrões das chamadas reais no seu ambiente ao longo do tempo.
    Esse aprendizado contínuo é o que conduz a melhores taxas de acerto, ao invés de uma métrica fixa que seria enganosa.

    Sugestão prática:

    1. Rode um piloto por 2 a 4 semanas em campanhas reais

    2. Meça as seguintes métricas no Genesys:
      – Percentual de AMD_Human que entrou como voicemail
      – Percentual de AMD_Machine que entrou como humano

    3. Ajuste a agressividade da política conforme o objetivo da campanha

    #Outbound



    ------------------------------
    Fernando Sotto dos Santos
    ConsultorVIA VAREJO
    ------------------------------

    Original Message


Related Content