Customer Service Automation
Humana needed to reduce the cost and poor experience caused by an IVR that pushed too many provider calls to live agents for routine pre-service questions.
Reported impact
The voice agent handles inquiries at about a third of the previous cost, delivers nearly double the response rate of the old IVR and lets providers complete common inquiries in about two minutes.
Humana worked with IBM Watson and IBM Data and AI Expert Labs & Learning to replace an underperforming IVR with a Provider Services Conversational Voice Agent that spans IBM Cloud and on-premises voice infrastructure. After a three-month proof of concept, the joint team combined multiple Watson applications in a single assistant, while watsonx Assistant for Voice handled the voice interaction layer at Humana. The system uses AI to determine caller intent, verify access rights and return eligibility, benefits, authorization, referral and other pre-service information without a live agent. Humana also trained seven language models and two acoustic models around different input types so the assistant could cope with healthcare terminology and noisy call-center conditions. In effect, the architecture turns provider phone inquiries into a continuously tuned voice workflow that serves routine healthcare administration at scale rather than handing those calls off to outsourced centers.
Provider phone calls are processed by watsonx Assistant for Voice and custom speech models, IBM Watson interprets the intent and access rights for each call, and the system returns pre-service, eligibility, benefits, authorization or referral information without routing routine requests to a live agent.
"Humana receives over one million provider calls every month"
"More than 60% of these calls were related to routine, specific, pre-service questions with well-defined answers."
"The solution combines multiple Watson applications in a single conversational assistant, run on IBM Cloud, while the watsonx Assistant for Voice runs on premises at Humana."
"The solution receives more than 7,000 voice calls from 120 providers per business day"
"the solution achieves an average of 90%-95% sentence error rate accuracy level"
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
IBM Sample Dec 18, 2025 Oncoclínicas&Co Oncoclínicas&Co needed to unify fragmented patient communication channels, reduce wait times and automate routine interactions without losing the personalized and empathetic experience patients expected. 38% improvement in appointment attendance IBM Cloud Pak for Business Automation · IBM Cloud Pak for Integration · IBM watsonx.ai IBM Sample Jun 11, 2024 University Hospitals Coventry and Warwickshire NHS Trust UHCW NHS Trust needed a more efficient way to answer routine HR and workforce questions for more than ten thousand employees so specialist staff could spend less time on inbox and phone traffic and more time on complex… 10,500+ employees IBM watsonx Assistant · IBM watsonx.ai 
Google Cloud Sample Modified · Oct 9, 2026 Lillia Lillia needed to reduce the friction of manual meal, medication, and vital tracking for chronic disease patients while freeing human care coaches from routine tasks and still meeting enterprise reliability and data… Automates 95% of early patient care interactions using Gemini APIs Artifact Registry · Cloud Build · Cloud Load Balancing 