Negotiate retention discounts on bills by phone
I’m Tikue, one of the co-founders of Kudos (joinkudos.com)We built this AI voice agent to call service providers, navigate IVR systems, wait on hold, and negotiate retention discounts on a user’s behalf.A few details that matter:By default, the agent negotiates within the user’s existing plan. Task: I’m Tikue, one of the co-founders of Kudos (joinkudos.com)We built this AI voice agent to call service providers, navigate IVR systems, wait on hold, and negotiate retention discounts on a user’s behalf.A few details that matter:By default, the agent negotiates within the user’s existing plan. It does not change plans unless the user explicitly provides flexibility or additional context.For example, in my own case with T-Mobile, I saved $840 per year because I told the agent I was open to plan changes. I had a single line sitting on an old multi-line family plan after friends gradually left over the years. Without that context, it would have only negotiated a discount on the existing structure.Savings vary a lot and are not guaranteed. This Xfinity example reduced the bill by $276 per year. Many outcomes are loyalty discounts, though legacy or misconfigured plans can produce larger reductions.Identity verification is a non-trivial part of the system. When a provider requires an OTP, the agent pauses and the user enters the code. We’ve spent significant time optimizing that step to ensure users are actually present and able to complete verification when needed, since timing and proximity matter.Some of the harder technical challenges:• Handling messy call state transitions across IVR trees, transfers, silence, interruptions, and unexpected flows • Enforcing guardrails so the agent does not make unintended commitments • Navigating automated front-line systems, which often means AI interacting with AI before reaching retention • Ensuring we get the OTP verification codes from users on time when they are requiredHappy to answer questions; tools: custom-built or unnamed agent. Human role: The original post does not say how a person stayed involved.; limitations: The system faces technical challenges with messy call state transitions, AI-to-AI interactions in automated front-lines, and enforcing guardrails against unintended commitments. Identity verification timing is non-trivial, requiring users to be present and responsive to OTP requests. The source is a self-reported founder post, so claims of ease or success should be treated as unverified..
Tools
custom-built or unnamed agent
Published on Agent Work Atlas:
Overview
I’m Tikue, one of the co-founders of Kudos (joinkudos.com)We built this AI voice agent to call service providers, navigate IVR systems, wait on hold, and negotiate retention discounts on a user’s behalf.A few details that matter:By default, the agent negotiates within the user’s existing plan.
Workflow
I’m Tikue, one of the co-founders of Kudos (joinkudos.com)We built this AI voice agent to call service providers, navigate IVR systems, wait on hold, and negotiate retention discounts on a user’s behalf.A few details that matter:By default, the agent negotiates within the user’s existing plan. It does not change plans unless the user explicitly provides flexibility or additional context.For example, in my own case with T-Mobile, I saved $840 per year because I told the agent I was open to plan changes. I had a single line sitting on an old multi-line family plan after friends gradually left over the years. Without that context, it would have only negotiated a discount on the existing structure.Savings vary a lot and are not guaranteed. This Xfinity example reduced the bill by $276 per year. Many outcomes are loyalty discounts, though legacy or misconfigured plans can produce larger reductions.Identity verification is a non-trivial part of the system. When a provider requires an OTP, the agent pauses and the user enters the code. We’ve spent significant time optimizing that step to ensure users are actually present and able to complete verification when needed, since timing and proximity matter.Some of the harder technical challenges:• Handling messy call state transitions across IVR trees, transfers, silence, interruptions, and unexpected flows • Enforcing guardrails so the agent does not make unintended commitments • Navigating automated front-line systems, which often means AI interacting with AI before reaching retention • Ensuring we get the OTP verification codes from users on time when they are requiredHappy to answer questions
Reported result
The source reports specific anecdotal savings: $840/year for a T-Mobile plan change and $276/year for an Xfinity loyalty discount. However, the author explicitly states that savings vary significantly and are not guaranteed. No aggregate performance metrics or failure rates are provided in the text.
Human role
The original post does not say how a person stayed involved.
Limitations
The system faces technical challenges with messy call state transitions, AI-to-AI interactions in automated front-lines, and enforcing guardrails against unintended commitments. Identity verification timing is non-trivial, requiring users to be present and responsive to OTP requests. The source is a self-reported founder post, so claims of ease or success should be treated as unverified.
Try it
Act as a voice agent developer. I need to build a prototype for a bill negotiation assistant. [replaceable inputs: Provider Name, Current Plan Details, User Flexibility Constraints]. Use Twilio for call control and ElevenLabs for voice synthesis. Implement a rule-based IVR navigator that pauses for human OTP input when required. Enforce a strict guardrail: do not agree to plan changes unless the user explicitly opts in. First, run a dry simulation with a mock IVR tree and a sample negotiation script. Show me the transcript and the proposed discount. Wait for my confirmation before attempting any real call or spending credits. If the simulation fails to handle a transfer or silence, report the error state immediately.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.