01 October 2024

We Put ChatGPT’s Advanced Voice Mode To The Test - Live Sales Call

2Frameworks

13Insights

Frameworks in this episode

Sales5 steps

Document-Grounded AI Sales Call Test

Turn product material into a timed AI sales simulation and evaluate the result

Read →

Productivity5 steps

Voice Walk-and-Talk Clarification Loop

Use spoken first-principles questioning to turn rough thoughts into a clear output

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster30:30

Voice Interfaces May Add New Behavior Instead of Killing Typing

The hosts question whether voice and augmented-reality interfaces will make typing obsolete. One counters that Google's earlier voice search expanded the total search market rather than cannibalizing text, suggesting voice and typing may remain complementary modes.

New interfaces do not always replace established ones
Voice search created additional behavior rather than eliminating text search
Different tasks may naturally favor voice or text
Personal habits could persist even as voice technology improves

“Google released voice search a long time ago and it didn't really cannibalize a lot of search it actually was additive”
31:00

“there's some things that are voice and there's some that are text”
31:00

#voice interfaces#typing#search#human computer interaction

Hot Take· 3

Hot Take19:30

Voice AI Adoption May Follow the Early Internet Curve

Kieran predicts that discomfort with AI interactions is temporary and compares it with early reluctance to transact online. The technology may become capable before mass consumer behavior catches up, after which AI interactions could feel ordinary or ubiquitous.

Technical capability and social acceptance advance at different rates
Customers may initially request humans even when AI performs better
Online transactions followed a similar trust-adoption lag
Voice agents may eventually become a normal company interface

“I just think we're at a period of time where these interactions for some folks or for some people will feel strange until they don't…”
Kieran · 20:00

“this is just a an adoption curve”
Kieran · 22:00

#adoption#consumer behavior#voice ai#technology

Hot Take25:00

Human Support Could Become Paid While AI Stays Free

The hosts consider a future in which free AI provides broad, continuous support while customers pay for scarce human attention. They also suggest that sufficiently capable AI might eventually become the premium tier rather than the cheaper fallback.

Free human support often lacks complete coverage
AI could provide broad coverage at lower marginal cost
Businesses might charge small fees for human interactions
Advanced AI could eventually outperform standard support and become premium

“you could imagine in the future that we will charge for human customer support and provide AI for free”
Kieran · 25:30

“there is a world where eventually AI support could even be premium support”
26:00

#customer support#pricing#automation#service design

Hot Take29:30

People Who Use AI Daily Will Pull Away From Those Who Do Not

Kieran predicts a stark capability gap between people who integrate AI into everyday work and those who avoid it. Persistent access to expert-level help across small tasks, major tasks, ideation, drafting, and variation generation creates an advantage that unaided workers will struggle to match.

AI applies to both small and substantial tasks
It can support thinking, drafting, and generating alternatives
Text and audio expand access to expert-like assistance
Routine usage may create a durable individual productivity advantage

“there is just going to be a stark difference between the people who are using these models in their day-to-day and people who are not”
Kieran · 30:00

“you just cannot compete with that”
Kieran · 30:00

#productivity#competitive advantage#future of work#ai adoption

Explainer· 4

Explainer08:30

How Voice AI Could Learn From the Best Sales Reps

Kieran argues that sales-call data can reveal behaviors separating top performers from weaker representatives. Because an AI can both analyze and conduct calls, it could rapidly incorporate successful cues and adjust its approach in real time.

Compare communication patterns across high- and low-performing reps
Identify call behaviors associated with better outcomes
Instrument successful behaviors into the AI agent
Adjust conversations in real time as evidence accumulates

“you can start to look at commonalities of best performing Reps versus like the you know worst performing rep”
Kieran · 09:00

“it will be able to adjust in real time to be able to make that call much much better”
Kieran · 09:00

#voice ai#sales data#bdr#machine learning

Explainer13:30

AI Can Master Product Knowledge but Still Feel Impersonal

The hosts distinguish informational effectiveness from the personal connection a human seller can create. AI can organize product knowledge and map it to customer needs, but rapport based on natural shared context may remain harder to reproduce convincingly.

AI can structure large amounts of unstructured product information
Accurate answers do not automatically create rapport
Human sellers can add contextual and personal nuance
Customer preference for humans may persist despite better AI answers

“the thing you want AI is better than a human the knowledge”
Kieran · 14:00

“humans have the ability to add nuance and personalization into those calls in a way that AI struggles with”
Kieran · 18:30

#sales#rapport#personalization#voice ai

Explainer16:30

The Remaining Voice AI Friction: Latency and Turn-Taking

The demo exposed a short delay between the human finishing and the AI responding, along with uncertainty about who should speak next. These issues did not erase the value of a brief appointment-setting call, but they made the artificial nature of the interaction obvious.

Responses still had roughly one to two seconds of latency
Turn-taking was less fluid than natural human dialogue
Short calls may tolerate the delay better than long conversations
Improved reasoning and lower latency should make exchanges more natural

“there's still a little latency there's still like one to two seconds that if you were in a natural conversation you wouldn't have”
16:30

“there's also some still like when does the AI talk when do I talk”
17:00

#latency#turn-taking#voice ai#user experience

Explainer23:30

Why the Voice Demo Was Not Yet Ready for Broad Deployment

At recording time, the voice API had not been released, preventing businesses from simply connecting the demonstrated capability to websites or phone systems. The hosts speculate that computational cost and unresolved pricing were major reasons for the delay.

The consumer demo did not provide a production deployment path
API pricing and availability remained unknown
Serving every call could be cost-prohibitive
Early deployments may need to target selected audience segments

“the voice API hasn't been released”
23:30

“I think it's probably pretty expensive for them to do this and they're trying to figure out how to price it”
24:00

#voice api#pricing#deployment#openai

Story· 1

Story15:30

Why Adding Automation Increased Website Chat Engagement

One host reports that introducing basic automation into website chat increased the rate at which visitors initiated conversations. The lower perceived commitment and pressure of not immediately engaging a human made more people willing to begin.

Automation handled the most common questions and use cases
Visitors experienced less commitment and social pressure
The website's chat rate increased after automation was introduced
AI may be especially suitable for low-pressure initial engagement

“what we saw is our chat rate actually went up”
16:00

“more people who visit that page opted to chat because there was less commitment and less like pressure because you weren't dealing with a human”
16:00

#chatbots#conversion#automation#customer experience

Tool· 1

Tool28:00

Use Voice to Interrogate Documents and Compare Major Purchases

The hosts describe speaking with AI about consequential choices and uploaded documents. Examples include comparing cars against personal priorities and asking questions about a company's newly released financial earnings.

Explore competing products through conversational comparison
State the criteria that matter most to receive a tailored analysis
Upload a PDF and question its contents by voice
Use dialogue to summarize dense financial or business documents

“tell me all the competitors tell me if these are the things I care about most tell me what's the best fit for me”
28:30

“you can just upload a PDF and just talk to that PDF with your voice”
29:00

#document analysis#decision making#voice ai#research

Takeaway· 3

Takeaway20:30

Tone and Sentiment Make AI Calls Feel Warmer

The new voice system appeared able to detect vocal tone and sentiment while responding with better pacing and rhythm. The hosts believe these qualities can make short AI calls feel materially warmer, even if they do not yet feel fully human.

Tone and sentiment detection improved the interaction
Voice pacing and rhythm affect perceived warmth
Lower latency would further reduce artificiality
Brief appointment-setting calls may not require perfect human imitation

“I felt more warmth than I expected in that call”
21:00

“when you remove the latency and when you add in tonality and you add in the the kind of human elements of that conversation I…”
Kieran · 22:30

#sentiment#speech#voice ai#customer experience

Takeaway24:30

Disclose the AI and Preserve a Route to a Human

The hosts argue that businesses should identify AI agents and let customers move to a human representative. For inbound calls, they expect real-time notice and consent to be necessary, with some callers immediately declining the AI interaction.

Tell customers when they are interacting with AI
Offer an accessible human alternative
Obtain appropriate consent for inbound voice interactions
Expect acceptance to vary across customers and jurisdictions

“you're still going to have to have consent in real time when somebody calls in”
24:30

“you'll have to give them that option”
25:00

#disclosure#consent#compliance#human handoff

Takeaway27:00

Start Voice AI With Warm Demand and Layered Support

For business deployment, Kieran favors bounded interactions such as inbound calls, conversations with warm rather than cold demand, and support that can shift from text to audio. He expects support to improve further when models can see the product and guide users visually.

Prioritize inbound interactions over unsolicited cold calls
Test lightweight BDR calls with warm demand
Let support conversations move from text to audio
Combine audio and visual understanding for product guidance

“a lightweight bdr like Business Development calls to warm demand not cold warm demand I think is a really good thing to test with support”
Kieran · 27:00

“support is likely going to be much better where you have a model that can do audio and visual and Gad you around the product”
Kieran · 27:30

#inbound sales#customer support#multimodal ai#bdr