Why AI Voice Feels More Human Than Ever

00:41:18
https://www.youtube.com/watch?v=-_qYRdEcNiE

Ringkasan

TLDRThe podcast explores the transformative influence of AI voice technology on business interactions and societal perceptions. As companies navigate the shift towards AI-driven communication, voices can replace human operatives, enhancing efficiency and user experience. The discussion emphasizes recent advancements in natural and emotional AI responses, catering to users' comfort with machines. Moreover, unexpected receptivity of consumers toward AI companionship illustrates changing dynamics in interpersonal communication. Industry-specific applications are emerging rapidly, with voice AI becoming viable across finance, healthcare, and logistics. Still, the challenge remains for incumbents to compete effectively against agile startups creating innovative solutions. Key areas of growth include emotional engagement and trust-building, prompting a reevaluation of consumer-company relationships in a digital era.

Takeaways

  • 📞 AI voice technology can significantly cut business costs and improve efficiency.
  • 🤖 Emotional engagement in AI interactions enhances user experience.
  • 🚀 Startups are rapidly innovating in the AI voice space, especially in vertical applications.
  • 🔍 Consumer receptivity towards AI companions is growing, changing perceptions of tech interactions.
  • 📊 Pricing models for AI voice services are evolving to reflect value beyond basic usage.
  • 📈 Trust-building in voice AI is crucial for broader adoption and success.
  • 🎯 Focus on high-value consumer interactions can lead to more impactful AI solutions.
  • 💬 Emotional responsiveness in AI can redefine companionship and user connection.
  • 🔧 Founders are encouraged to build quickly and experiment based on real-time feedback.
  • 🌍 AI voice applications are increasingly integral in sectors like healthcare and finance.

Garis waktu

  • 00:00:00 - 00:05:00

    The discussion highlights the growing influence of AI in business interactions, emphasizing that businesses increasingly leverage AI for phone calls, primarily to reduce costs associated with human labor. AI voice systems are seen as a potential solution for companies spending significant amounts on call answering, leading to a new paradigm of communication with consumers.

  • 00:05:00 - 00:10:00

    There is a reflection on the challenges experienced with existing voice AI products like Siri and Alexa, which often lack personality and depth in conversation. This has resulted in people turning off such assistants due to unsatisfactory interactions, indicating a need for more human-like and engaging voice interfaces.

  • 00:10:00 - 00:15:00

    As advancements in large language models (LLMs) and voice interaction technologies are made, there is optimism that voice may become a primary interface for AI engagement. The conversation touches upon the evolution of voice technology from basic interactions to more complex conversational capabilities that mimic human interaction.

  • 00:15:00 - 00:20:00

    A review of past AI voice technologies shows a transition from basic interactive voice response (IVR) systems to more sophisticated AI that can engage in meaningful conversations. Current models are better equipped to process holistic customer inquiries based on integrated data sources rather than responding to specific keywords or commands.

  • 00:20:00 - 00:25:00

    The evolution of voice technology has resulted in improved latency, emotional engagement, and human-like dialogue structures, making AI voices more relatable and capable of navigating complex conversations empathetically. These factors contribute to a more enriched user experience when interacting with voice AI.

  • 00:25:00 - 00:30:00

    There's a discussion on the different business models arising from AI voice technology, including outcome-based pricing, platform fees, and per-minute usage. These new pricing strategies are evolving as the technology demonstrates its value in various applications, particularly in high-volume B2B settings.

  • 00:30:00 - 00:35:00

    The conversation further explores the potential for AI in consumer applications, particularly in areas like mental health, education, and recruitment, where traditional human interactions can be augmented or replaced. Receptivity from consumers is seen as a key factor in the adoption of AI technologies.

  • 00:35:00 - 00:41:18

    Final thoughts revolve around the need for opinionated and engaging AI voice companions that can form meaningful connections with users, reflecting on how startups can innovate in this field by building unique, character-driven voice agents instead of generic systems.

Tampilkan lebih banyak

Peta Pikiran

Video Tanya Jawab

  • What businesses can benefit from AI voice technology?

    Any business that spends a significant amount on call centers or human customer service roles, typically those paying $100-150K annually, can benefit from AI voice technology.

  • Why do people prefer interacting with AI, especially in interviews?

    Many candidates feel that AI interviewers provide a more unbiased and consistent evaluation compared to human recruiters, leading to a more positive experience.

  • How has AI voice technology advanced recently?

    Recent advancements have led to more natural sounding voices, reduced latency in conversations, and improved emotional responsiveness, making interactions feel more human.

  • What sectors are seeing the most growth in AI voice applications?

    Sectors like financial services, healthcare, logistics, and call centers are experiencing significant growth in AI voice applications.

  • What is the significance of emotionality in AI voice technology?

    Emotionality helps in making AI interactions feel more human, improving user engagement and experience.

  • How do companies price AI voice services?

    Pricing models are evolving, with common structures including per minute charges, platform fees, or outcome-based pricing.

  • What challenges do traditional companies face in competing with new AI voice startups?

    Incumbents often struggle with innovation and may be less adaptable to the fast-paced changes needed for effective AI voice integration.

  • What advice is given to founders in the AI voice space?

    Founders should focus on building quickly and exploring high-value, sensitive use cases that demand sophisticated conversations.

  • How are consumer applications of AI voice different from B2B applications?

    Consumer applications often focus on high-cost, hard-to-access services like mental health support, while B2B applications replace existing human roles.

  • What future trends are expected in AI voice technology?

    Increased personalization, improved emotional engagement, and new vertical applications that leverage the unique capabilities of AI voice.

Lihat lebih banyak ringkasan video

Dapatkan akses instan ke ringkasan video YouTube gratis yang didukung oleh AI!
Teks
en
Gulir Otomatis:
  • 00:00:00
    we see a lot of businesses that are
  • 00:00:01
    already doing thousands tens of
  • 00:00:02
    thousands of phone calls with AI every
  • 00:00:05
    day any business that pays a person 100
  • 00:00:08
    150k a year to answer phone calls is a
  • 00:00:11
    potential customer of voice AI they
  • 00:00:14
    think the rules of the game are changing
  • 00:00:15
    do people really want to be friends with
  • 00:00:17
    an AI and is that good for our society
  • 00:00:19
    and I think like yes and yes voice is a
  • 00:00:22
    platform that we need to be more
  • 00:00:24
    opinionated because interesting people
  • 00:00:25
    are opinionated exactly exactly the type
  • 00:00:29
    of empow of products you can build is
  • 00:00:31
    also above anything that we've ever seen
  • 00:00:34
    trust has to be earned and if the models
  • 00:00:36
    don't design for that they're never
  • 00:00:37
    going to get to their full potential and
  • 00:00:39
    I think we're going to see it in the
  • 00:00:40
    next 12 months not the next 5
  • 00:00:45
    years to me when I think of AI Voice or
  • 00:00:48
    at least voice products I think of Alexa
  • 00:00:50
    I think of Siri and I actually I
  • 00:00:53
    personally turn off Siri I think a lot
  • 00:00:54
    of people do too so tell me a bit about
  • 00:00:57
    why that's the case why haven't these
  • 00:00:59
    products delivered the kind of AI voice
  • 00:01:01
    magic that people have been waiting for
  • 00:01:03
    yeah it's really interesting because I
  • 00:01:05
    feel like now in the world of llms voice
  • 00:01:07
    is one of the most magical and engaging
  • 00:01:09
    ways to interact with AI but arguably
  • 00:01:11
    we've had these AI voice products for a
  • 00:01:13
    while and they were kind of
  • 00:01:14
    disappointing and honest compelling and
  • 00:01:16
    I think there's a couple reasons like
  • 00:01:18
    one the voices themselves sound robotic
  • 00:01:21
    and then I think the biggest thing
  • 00:01:23
    actually is just like what is behind the
  • 00:01:26
    voice like what is the engine so like a
  • 00:01:28
    Siria or an Alexa it might be connected
  • 00:01:30
    to kind of a basic set of Integrations
  • 00:01:32
    within the Apple ecosystem or within the
  • 00:01:34
    Amazon ecosystem so maybe it's pulling
  • 00:01:36
    product information or asking a basic
  • 00:01:38
    question but it doesn't have a
  • 00:01:40
    personality it doesn't really have a
  • 00:01:41
    brain it's probably not connected to the
  • 00:01:43
    internet in most cases it's in no way
  • 00:01:45
    like a true conversational partner in a
  • 00:01:47
    way that people are interacting with AI
  • 00:01:50
    voice now like it is a human or in some
  • 00:01:53
    ways even better than a human I think
  • 00:01:55
    there's definitely the use cases which
  • 00:01:57
    are very constrained to your point but
  • 00:01:59
    then there's also just like the tonality
  • 00:02:00
    of it and the back and forth and the
  • 00:02:03
    sort of rational critique I think where
  • 00:02:04
    we're like you can't do that many things
  • 00:02:06
    and it can't but then there's the just
  • 00:02:08
    the emotional you know what you call the
  • 00:02:10
    uncanny valley where you just feel like
  • 00:02:12
    you're talking to something that is a
  • 00:02:13
    system or a technology not like not even
  • 00:02:16
    coming close to having interaction with
  • 00:02:18
    a person well it sounds like that might
  • 00:02:19
    be changing you both have released this
  • 00:02:21
    AI voice report of sorts this thesis and
  • 00:02:24
    I just want to call out a few quotes
  • 00:02:25
    from it you said that voice is one of
  • 00:02:27
    the most powerful unlocks for AI
  • 00:02:29
    application companies and also that for
  • 00:02:31
    consumers We Believe voice will be the
  • 00:02:33
    first and perhaps the primary way people
  • 00:02:35
    interact with AI so those are pretty
  • 00:02:37
    bold statements tell me about that and
  • 00:02:40
    specifically the why now the one I think
  • 00:02:42
    is that we have models that work for the
  • 00:02:44
    first time you know there's a lot of
  • 00:02:45
    attempts at voice but the technology
  • 00:02:47
    simply didn't work there's a bunch of
  • 00:02:49
    attempts at the infrastructure level
  • 00:02:51
    everything from Dragon Naturally
  • 00:02:52
    Speaking and a major development in the
  • 00:02:55
    computer world today as Massachusetts
  • 00:02:57
    based Dragon systems announced the first
  • 00:02:59
    first Affordable Computer dictation
  • 00:03:02
    system that understands standard natural
  • 00:03:04
    speech you know all the way on to the
  • 00:03:06
    2000s and 20010s and then there was
  • 00:03:09
    application efforts like voice XML but
  • 00:03:11
    just that like the sort of underlying
  • 00:03:13
    technology didn't work very well so we
  • 00:03:16
    never really got to the like what can we
  • 00:03:17
    do with this now so one I think the
  • 00:03:19
    model really works and the technology
  • 00:03:21
    really works both in terms of the llms
  • 00:03:23
    as well as the uh text to speech speech
  • 00:03:25
    to text two I think that we've got this
  • 00:03:28
    opportunity to use phone calls as a new
  • 00:03:31
    distribution channel so I think the
  • 00:03:33
    product capability is there and it's
  • 00:03:34
    really compelling but the fact that it's
  • 00:03:36
    paired with a very natural distribution
  • 00:03:38
    channel is also really interesting yeah
  • 00:03:40
    I would agree if it's like one thing to
  • 00:03:42
    talk to chat gbt via text and to have a
  • 00:03:45
    great experience there but it's another
  • 00:03:48
    thing entirely to be able to talk to
  • 00:03:50
    chat chbt or any other llm via voice
  • 00:03:53
    because it's next level like it both has
  • 00:03:54
    to generate what you would see in the
  • 00:03:56
    text and then it has to sound like an
  • 00:03:57
    actual human talking back to you and
  • 00:03:59
    when it accomplishes that it's just this
  • 00:04:01
    crazy it's almost like an emotional
  • 00:04:03
    feeling and then I think to anisha's
  • 00:04:05
    point um in terms of why so many
  • 00:04:08
    consumers will encounter AI voice it
  • 00:04:10
    might be because they choose to like
  • 00:04:11
    they'll go and talk to Chachi BT they'll
  • 00:04:13
    go and use a language learning app with
  • 00:04:15
    AI on their cell phone but also I think
  • 00:04:18
    many businesses in a great way will
  • 00:04:20
    impose it on them because you can now
  • 00:04:22
    use AI to replace phone calls which is
  • 00:04:24
    so much more efficient and cost
  • 00:04:26
    effective for them and so many consumers
  • 00:04:28
    probably actually have already
  • 00:04:29
    interacted with AI via voice and might
  • 00:04:31
    not have even known it or detected it we
  • 00:04:34
    see a lot of businesses that are already
  • 00:04:35
    doing thousands tens of thousands of
  • 00:04:37
    phone calls with AI every day but from
  • 00:04:40
    my experience a lot especially if it's a
  • 00:04:42
    short phone call a lot of these AI voice
  • 00:04:44
    agents are so good that you wouldn't be
  • 00:04:45
    able to tell it's interesting because I
  • 00:04:47
    think that Talking Heads want to tell
  • 00:04:48
    you that people don't want to talk to an
  • 00:04:50
    AI but in all the cases where people do
  • 00:04:52
    interact with an AI that starts to call
  • 00:04:54
    by announcing I'm an AI people are like
  • 00:04:56
    oh cool let's just get into it and as
  • 00:04:58
    soon as they start to feel the feeling
  • 00:04:59
    of a human conversation they immediately
  • 00:05:01
    forget or sort of don't care that it's
  • 00:05:03
    an AI right so let's talk about this
  • 00:05:05
    idea of an operating platform voice is
  • 00:05:08
    this new operating platform that people
  • 00:05:09
    are building on top of can we just walk
  • 00:05:11
    through maybe the wave of technological
  • 00:05:13
    unlocks or maybe the different steps
  • 00:05:16
    we've taken to get to where we are yeah
  • 00:05:18
    I mean maybe we can start with like the
  • 00:05:20
    first wave of kind of early AI phone
  • 00:05:23
    technology which would be kind of the
  • 00:05:25
    ivr phone trees of like you know press
  • 00:05:27
    one for sales press two for customer
  • 00:05:29
    support this was you know late '90s
  • 00:05:30
    early 2000s and then we moved more
  • 00:05:33
    recently into kind of truly AI driven
  • 00:05:36
    but still very limited where it was an
  • 00:05:38
    AI but it was listening for you to say a
  • 00:05:40
    specific word that it could then use to
  • 00:05:42
    trigger like a very specific and set
  • 00:05:44
    workflow or script like I many times
  • 00:05:47
    unfortunately have had to yell like
  • 00:05:49
    customer service into a phone all the
  • 00:05:51
    time yeah exactly and so in that case
  • 00:05:54
    the AI is listening for you to say that
  • 00:05:56
    and then it knows okay let me rout route
  • 00:05:57
    the call to the customer service
  • 00:05:59
    department now what we're seeing with
  • 00:06:01
    this kind of new wave of infrastructure
  • 00:06:02
    and then application layer companies is
  • 00:06:05
    where the AI isn't listening for for one
  • 00:06:07
    thing in particular but it's trying to
  • 00:06:09
    get a more holistic sense of what are
  • 00:06:11
    you as a customer asking for it's
  • 00:06:13
    accessing resources from the business
  • 00:06:14
    it's accessing resources from the
  • 00:06:16
    internet and it can have a much more
  • 00:06:18
    humanlike conversation with you and even
  • 00:06:21
    within AI you know 2.0 in the way that
  • 00:06:23
    you guys frame it it seems like we've
  • 00:06:25
    progressed a lot even within that phase
  • 00:06:27
    can we talk about maybe some of those
  • 00:06:28
    unlocks whether it's specific models
  • 00:06:30
    that have been released the way that the
  • 00:06:32
    infrastructure has changed maybe we can
  • 00:06:33
    skip certain steps can we talk about
  • 00:06:35
    that I think we've made leaps in a bunch
  • 00:06:37
    of areas so probably the biggest and
  • 00:06:40
    most obvious one would be latency so
  • 00:06:43
    this time last year two to three seconds
  • 00:06:45
    of latency was pretty good now a second
  • 00:06:48
    of latency is too long maybe even half
  • 00:06:51
    of a second of latency is too long in
  • 00:06:53
    many cases so that has been a massive
  • 00:06:56
    unlock I think enabled by by new models
  • 00:06:58
    and just for the a what is like the
  • 00:07:00
    latency for humans like if we're talking
  • 00:07:03
    I mean it sub definitely sub like 300
  • 00:07:06
    milliseconds sometimes even less than
  • 00:07:08
    that if you have humans interrupting
  • 00:07:10
    humans for
  • 00:07:12
    sure and you can have some of the most
  • 00:07:14
    humanlike voice agents that I've seen
  • 00:07:17
    are capable of being interrupted by
  • 00:07:19
    humans and also capable of interrupting
  • 00:07:21
    humans too uh which makes them feel like
  • 00:07:23
    more of a a conversation the second one
  • 00:07:25
    would be kind of humanness of the voice
  • 00:07:27
    so again hearkening back to like a Siri
  • 00:07:29
    or Alexa does it sound like a robot or
  • 00:07:32
    does it sound like a real person we're
  • 00:07:34
    investors in companies like 11 Labs that
  • 00:07:36
    have uh built very deep models that
  • 00:07:39
    either have preset voices that sound
  • 00:07:41
    real or that you can design your own
  • 00:07:43
    character voice essentially depending on
  • 00:07:45
    your use case now you can create any
  • 00:07:48
    voice simply by typing a text
  • 00:07:51
    description i l there's magic in these
  • 00:07:54
    old bones yet another unlock that I've
  • 00:07:57
    noticed has made particular amount of
  • 00:08:00
    progress in the last 3 to four months is
  • 00:08:03
    emotionality so if you say something
  • 00:08:06
    that is supposed to be sad does the AI
  • 00:08:08
    sound a little down or a little sad when
  • 00:08:10
    it responds if you say something
  • 00:08:13
    exciting does it pick up the pace does
  • 00:08:15
    it pick up the pitch at which it's
  • 00:08:16
    talking back to you and then lastly I
  • 00:08:18
    think is there's not a term for this yet
  • 00:08:22
    maybe we should come up with one but
  • 00:08:24
    like the dialogue structure to an AI
  • 00:08:27
    model they will know if exactly what
  • 00:08:30
    words they want to say back to you right
  • 00:08:32
    so there's no reason for them to put in
  • 00:08:34
    any pauses any gaps any little vocal
  • 00:08:37
    ticks but to a human listener very few
  • 00:08:41
    humans just speak perfectly with no
  • 00:08:43
    interruptions with no weird little
  • 00:08:44
    inflections with no pauses notebook LM
  • 00:08:47
    is one example where that sounded so
  • 00:08:50
    human because they put in all of these
  • 00:08:53
    things that to an AI might feel like an
  • 00:08:55
    error but to a human it sounds like
  • 00:08:57
    another human talking you know we always
  • 00:08:59
    talk about you know diving deep into a
  • 00:09:02
    topic right but today's dive
  • 00:09:06
    well uh it's a bit of a doozy yeah it's
  • 00:09:09
    deeply personal and so we're seeing more
  • 00:09:11
    companies like Sesame is a good example
  • 00:09:13
    in our portfolio introducing things like
  • 00:09:15
    that in the model which just UPS the
  • 00:09:17
    kind of realness Factor hey looks like
  • 00:09:20
    we got cut short last time feel like
  • 00:09:22
    picking up what we left
  • 00:09:24
    off yeah I don't remember what we were
  • 00:09:26
    talking about though no worries happens
  • 00:09:28
    to the best of us we were diving into
  • 00:09:30
    weekend plans I was telling you about my
  • 00:09:32
    reading you know processing all that
  • 00:09:34
    text and code keeps my circuits firing
  • 00:09:36
    you know these latter two points are so
  • 00:09:38
    important I love the point about
  • 00:09:39
    emotionality because it is not an
  • 00:09:42
    obvious area to explore and yet when you
  • 00:09:44
    interact with a model that has invested
  • 00:09:46
    in emotionality it just feels like a
  • 00:09:48
    completely different product you really
  • 00:09:50
    like feel the feelings in a completely
  • 00:09:52
    different way you know as is designed so
  • 00:09:54
    I think it's a really really powerful
  • 00:09:56
    Direction Flo exploration and I would
  • 00:09:57
    argue even for the Alexis and sir
  • 00:10:00
    like even if they didn't invest a lot
  • 00:10:01
    more in intelligence and capabilities if
  • 00:10:04
    they over invested in emotionality they
  • 00:10:06
    might actually get a lot of the way
  • 00:10:08
    there in terms of consumer experience um
  • 00:10:10
    and and yet I have a feeling that none
  • 00:10:12
    of those companies are thinking about it
  • 00:10:13
    that way one interesting stat that you
  • 00:10:15
    guys shared was the percentage of YC
  • 00:10:17
    companies that are now pursuing AI voice
  • 00:10:20
    what are we seeing there in terms of how
  • 00:10:22
    cohorts have changed and the percentage
  • 00:10:24
    of I guess these new companies on the
  • 00:10:26
    frontier actually pursuing this field YC
  • 00:10:28
    Founders are typically Al you know young
  • 00:10:30
    High hustle ambitious and they're like
  • 00:10:33
    heat seeking missiles and so they will
  • 00:10:35
    pivot until they get into a space that's
  • 00:10:37
    interesting so in recent YC cohorts
  • 00:10:40
    upwards of 20 25% of companies are are
  • 00:10:43
    building with AI voice which is really
  • 00:10:45
    exciting we're even seeing a lot of
  • 00:10:47
    companies from past cohorts all the way
  • 00:10:49
    back to like 2019 2020 are going back
  • 00:10:52
    now and pivoting into AI voice the first
  • 00:10:55
    wave after the infrastructure companies
  • 00:10:57
    inv voice we saw were pretty horizontal
  • 00:10:59
    platforms that allow anyone any business
  • 00:11:02
    any consumer to build kind of a
  • 00:11:03
    broad-based voice agent like I built one
  • 00:11:06
    that called the DMV for me and scheduled
  • 00:11:08
    an appointment which is very useful what
  • 00:11:11
    type of appointment do you need say
  • 00:11:13
    behind a wheel driving test or an office
  • 00:11:15
    visit and the next wave that we're
  • 00:11:17
    starting to see is a lot more
  • 00:11:19
    verticalized and I think it makes sense
  • 00:11:21
    because the ability to build a voice
  • 00:11:24
    agent has kind of commoditized if even I
  • 00:11:27
    can make somewhat of a performance voice
  • 00:11:29
    agent with models that are available and
  • 00:11:32
    so now we're seeing companies think
  • 00:11:33
    Beyond okay you have the voice agent
  • 00:11:36
    using that as a wedge what is the next
  • 00:11:39
    level of software that you can build can
  • 00:11:41
    you build the AI native vertical SAS
  • 00:11:44
    product for an industry using that voice
  • 00:11:47
    agent can you invent a new system of
  • 00:11:48
    record what can you do next and so that
  • 00:11:51
    leads you into being a little bit more
  • 00:11:52
    focus and verticalized and that's where
  • 00:11:54
    a lot of the YC companies are landing I
  • 00:11:56
    think it mirrors the cloud transition in
  • 00:11:58
    many ways and the initial vertical SAS
  • 00:12:00
    wave of 10 years ago because I think at
  • 00:12:03
    that time there's a lot of criticism
  • 00:12:04
    that like these markets seem too small
  • 00:12:06
    and yet many companies through you know
  • 00:12:08
    just larger than apparent vertical SAS
  • 00:12:10
    Market built big businesses and then
  • 00:12:12
    also found new ways to monetize things
  • 00:12:14
    like fintech I think similarly for voice
  • 00:12:16
    has applied to Vertical use cases you
  • 00:12:18
    know any business that pays a person 100
  • 00:12:22
    150k a year to answer phone calls is a
  • 00:12:25
    potential customer of voice Ai and and
  • 00:12:27
    you know can lead do a really
  • 00:12:28
    interesting vertical opportunity and
  • 00:12:30
    what are some examples of some of those
  • 00:12:32
    vertical opportunities where we're
  • 00:12:33
    seeing real companies break out pretty
  • 00:12:35
    much every vertical has a voice agent
  • 00:12:37
    company which is really exciting I think
  • 00:12:38
    to anisha's point when we talk to most
  • 00:12:40
    voice agent companies they aren't
  • 00:12:43
    necessarily replacing existing software
  • 00:12:46
    but they're probably actually allowing
  • 00:12:47
    businesses to either cut down on human
  • 00:12:49
    labor or reallocate their human labor to
  • 00:12:53
    kind of more effective things for the
  • 00:12:54
    business jobs that humans also are are
  • 00:12:57
    happier to do I would say where we've
  • 00:12:59
    seen voice agents take off the most like
  • 00:13:01
    where has a startup actually been able
  • 00:13:03
    to do a million calls on the phone have
  • 00:13:06
    been the call center category so you as
  • 00:13:09
    a business customer are already paying
  • 00:13:11
    10K 15K 20K a month to have people
  • 00:13:15
    making and taking phone calls for you
  • 00:13:17
    there's a ton of this in financial
  • 00:13:19
    services a ton of this in healthare a
  • 00:13:22
    lot of this in government but every
  • 00:13:24
    vertical has like we're investors in a
  • 00:13:26
    company called happy robot which builds
  • 00:13:27
    specifically for Freight and a lot of
  • 00:13:29
    those logistics companies previously had
  • 00:13:32
    call centers that they were paying tens
  • 00:13:33
    if not hundreds of thousands of dollars
  • 00:13:35
    to make and take calls so it's really
  • 00:13:37
    happening almost everywhere right now I
  • 00:13:40
    think it's becoming increasingly
  • 00:13:41
    consensus that any place where there's a
  • 00:13:43
    large volume of phone calls and
  • 00:13:45
    significant spend is an obvious area to
  • 00:13:46
    apply AI but an interesting area for
  • 00:13:49
    exploration that connects to our point
  • 00:13:50
    about emotionality is you know if you're
  • 00:13:53
    um negotiating I don't know a divorce
  • 00:13:55
    settlement or some sort of incredibly
  • 00:13:58
    important um corporate transaction every
  • 00:14:00
    phone call really really matters which
  • 00:14:02
    is why many of the people that make
  • 00:14:03
    those phone calls attorneys for example
  • 00:14:05
    may get paid thousands of dollars an
  • 00:14:06
    hour like what is the AI skew that gets
  • 00:14:10
    paid thousands of dollars an hour to
  • 00:14:12
    make a phone call and I think we're
  • 00:14:14
    going to see it in the next 12 months
  • 00:14:15
    not the next 5 years there's been some
  • 00:14:18
    very at least to me non-obvious examples
  • 00:14:21
    in use cases recruiting is one so
  • 00:14:24
    there's like 45 publicly traded staffing
  • 00:14:27
    companies that do inter inters for yes
  • 00:14:30
    bluecollar jobs but also engineering
  • 00:14:32
    jobs a massive range of them and what we
  • 00:14:36
    find is that a lot of candidates would
  • 00:14:38
    actually prefer talking to an AI
  • 00:14:41
    interviewer than talking to a human
  • 00:14:43
    recruiter that maybe has to take 10
  • 00:14:46
    calls that day is tired is in a bad mood
  • 00:14:49
    doesn't really have the technical de
  • 00:14:51
    exactly and maybe doesn't have the
  • 00:14:53
    technical expertise for every single job
  • 00:14:55
    that they're interviewing for to kind of
  • 00:14:57
    understand what are the smart follow-up
  • 00:14:59
    questions to really get at their
  • 00:15:01
    expertise and so that's one example of
  • 00:15:03
    you would think that a human would be
  • 00:15:06
    shocked offended upset to find
  • 00:15:08
    themselves interviewing with an AI but
  • 00:15:10
    in many cases by the end of the
  • 00:15:11
    interview they're actually more excited
  • 00:15:14
    and more positive about it than you
  • 00:15:15
    would think that is so interesting it's
  • 00:15:17
    kind of like you know the Uber Airbnb
  • 00:15:19
    you know I'm not going to no one's going
  • 00:15:20
    to want to stay in a stranger's house
  • 00:15:22
    you know driving a stranger's car and
  • 00:15:24
    then what do you know everyone's okay
  • 00:15:26
    with it the human at the end actually
  • 00:15:28
    often like it better because it's
  • 00:15:30
    unbiased like it's the same AI That's
  • 00:15:32
    evaluating everyone it's evaluating them
  • 00:15:35
    based on your actual performance not
  • 00:15:37
    based on whether they like you more or
  • 00:15:39
    less than someone else that they might
  • 00:15:41
    be evaluating you know it's interesting
  • 00:15:43
    because I think there's always been
  • 00:15:44
    these predictions around consumer
  • 00:15:45
    receptivity to new technology and
  • 00:15:47
    consumers cons consistently show
  • 00:15:49
    themselves to be more receptive so a
  • 00:15:51
    great example of this is sharing
  • 00:15:52
    location which 10 years ago is like oh
  • 00:15:54
    my God nobody is going to share location
  • 00:15:56
    it's too creepy it's too personal and
  • 00:15:58
    now I think a lot of people gen Z gen
  • 00:16:01
    Alpha share their fine friends with all
  • 00:16:03
    of their friends which is terrifying
  • 00:16:05
    don't understand it but so consumers are
  • 00:16:08
    highly receptive and I think the the
  • 00:16:09
    sort of analog to this in AI is
  • 00:16:12
    companionship and friendship you know
  • 00:16:13
    which is a much broader concept than
  • 00:16:15
    voice though voice really brings it to
  • 00:16:16
    life and people say hey do people really
  • 00:16:19
    want to be friends with an AI and is
  • 00:16:20
    that good for our society and I think
  • 00:16:22
    like yes and yes I think people are
  • 00:16:24
    getting much more socially skilled than
  • 00:16:26
    they were through the consumption of
  • 00:16:28
    things like social media which isn't
  • 00:16:29
    necessarily a bad thing either but I
  • 00:16:31
    think the sort of pundit perception of
  • 00:16:33
    this as the next gen of social media is
  • 00:16:36
    totally wrong and instead it sort of
  • 00:16:38
    enhances our uh our ability to interact
  • 00:16:41
    with real people I think people were
  • 00:16:42
    surprised quite frankly that the AI
  • 00:16:44
    companions text version had caught on to
  • 00:16:47
    the extent that they did were there any
  • 00:16:49
    surprises with voice as that was
  • 00:16:51
    introduced in terms of the adoption the
  • 00:16:52
    way that people were engaging with these
  • 00:16:54
    companions or anything like that so
  • 00:16:56
    there's some companion platforms that
  • 00:16:58
    are vo first for for example character
  • 00:16:59
    AI added a voice mode and it got some
  • 00:17:02
    crazy amount of usage in beta I think
  • 00:17:04
    actually a lot of people are kind of
  • 00:17:06
    taking for example inflections PI app or
  • 00:17:08
    chat gbt in voice mode and using it as a
  • 00:17:11
    companion and you might try it once CU
  • 00:17:14
    you're driving or your handsfree or you
  • 00:17:16
    know it feels more convenient but I mean
  • 00:17:19
    you say this a lot like in many cases
  • 00:17:21
    the AI is more human than the human even
  • 00:17:25
    your best friend if you give them a call
  • 00:17:27
    they might be busy they're at work
  • 00:17:28
    they're having a bad day are they
  • 00:17:30
    actually going to listen to every single
  • 00:17:32
    word that you're saying and respond in
  • 00:17:34
    kind of like an empathetic way and a
  • 00:17:36
    thoughtful way and the AI does that 100%
  • 00:17:39
    of the time and they have more expertise
  • 00:17:41
    more knowledge more resources this will
  • 00:17:43
    only get better as the models improve
  • 00:17:46
    because we're still in in the early days
  • 00:17:48
    um but a lot of people are shocked by
  • 00:17:50
    how friendly it feels to talk to an AI
  • 00:17:54
    you know I think an interesting area
  • 00:17:56
    also for consideration is just the
  • 00:17:58
    passive use case cases of voice like Hey
  • 00:18:00
    listen to me in this conversation listen
  • 00:18:03
    to me in this meeting listen to me sort
  • 00:18:05
    of you know reset recite the set of
  • 00:18:07
    ideas and AI can just listen passively
  • 00:18:09
    in the way that You' probably never ask
  • 00:18:11
    another person to yeah and give you
  • 00:18:13
    notes and feedback so it feels like
  • 00:18:14
    that's also an area that you know lends
  • 00:18:16
    itself a lot better to a technology Le
  • 00:18:19
    concept than a human Le concept and
  • 00:18:20
    we're just starting to see the
  • 00:18:21
    beginnings of that and what both of you
  • 00:18:23
    have touched on is this idea of instead
  • 00:18:25
    of substitution which is what people
  • 00:18:27
    mostly jump to when they think about
  • 00:18:29
    Technologies replacing humans and really
  • 00:18:31
    this idea of augmentation as well and
  • 00:18:33
    you mentioned it also in the scenario of
  • 00:18:35
    like let's say you take some company
  • 00:18:37
    that you know only has a receptionist 9
  • 00:18:38
    to5 what about 5 to9 or 24/7 can you
  • 00:18:42
    talk a little bit about how you're
  • 00:18:43
    seeing these AI companies wedge in and
  • 00:18:46
    kind of start the engines I would say a
  • 00:18:48
    lot of businesses I mean small
  • 00:18:50
    businesses to Enterprise alike are for
  • 00:18:53
    their own reasons like nervous to hand
  • 00:18:55
    over all of their phone calls and
  • 00:18:57
    customer interactions to an I so we'll
  • 00:18:59
    often see these voice agents start with
  • 00:19:01
    a specific wedge that just feels so
  • 00:19:04
    obvious in terms of Roi to the business
  • 00:19:06
    and then as they gain trust expand from
  • 00:19:09
    there so one of the most obvious and
  • 00:19:10
    easiest ones are these after hours or
  • 00:19:13
    overflow calls so if you're a small
  • 00:19:15
    business you probably live or die by the
  • 00:19:17
    ability to kind of get an appointment
  • 00:19:19
    booked having that handled by an AI is
  • 00:19:21
    kind of a no-brainer like at the very
  • 00:19:24
    least they can get a phone number and
  • 00:19:25
    information and call back but maybe they
  • 00:19:27
    can actually book a full appointment for
  • 00:19:29
    you and have like a job you know on deck
  • 00:19:31
    for the next day which is awesome but
  • 00:19:33
    beyond that there's a lot of other kind
  • 00:19:35
    of clever things that I think we've seen
  • 00:19:38
    uh companies do so there's some calls
  • 00:19:40
    that just don't make sense to make right
  • 00:19:43
    now if you're paying human labor like if
  • 00:19:45
    you're a credit card company you send
  • 00:19:47
    out a credit card uh and the consumer
  • 00:19:50
    never activates it does it actually make
  • 00:19:52
    sense to call them after one or two or
  • 00:19:54
    three days and get them to do that we
  • 00:19:56
    see I've seen a couple voice agents that
  • 00:19:58
    are really successful now with that use
  • 00:20:00
    case alone anything that's back office
  • 00:20:04
    it's not client facing so it's uh less
  • 00:20:08
    sensitive but if you're say a doctor's
  • 00:20:11
    office you probably have humans that
  • 00:20:14
    you're paying a lot spending hours on
  • 00:20:16
    the phone every day with Pharmacies with
  • 00:20:18
    insurers and that is time that they
  • 00:20:20
    could have spent with your patients or
  • 00:20:22
    making the clinic operate better and so
  • 00:20:25
    those kinds of calls are super obvious
  • 00:20:27
    and and like a great idea for voice
  • 00:20:29
    agents to tackle and then maybe the most
  • 00:20:32
    interesting one and one that we've
  • 00:20:33
    talked about a lot is there's so many
  • 00:20:36
    types of calls or interactions where
  • 00:20:38
    humans are not incentivized to do them
  • 00:20:41
    well maybe they have to make an upsell
  • 00:20:44
    and it's awkward but they are not
  • 00:20:46
    getting an extra commission for doing
  • 00:20:48
    that so they're going to skip it 80% of
  • 00:20:50
    the time and AI will just do it every
  • 00:20:53
    time and we'll kind of do it proudly and
  • 00:20:56
    and if they get turned down you know
  • 00:20:58
    they're just going to move on to the
  • 00:21:00
    hundred other calls that they're doing
  • 00:21:01
    simultaneously the AI is so relentlessly
  • 00:21:04
    cheerful yet never gives an inch of the
  • 00:21:06
    negotiation right which is amazing yes
  • 00:21:09
    you know I think to this point one of
  • 00:21:10
    the magic moments for a lot of the
  • 00:21:11
    customers of these products is when they
  • 00:21:13
    see it actually improves like in the
  • 00:21:15
    case of recruiting it improves candidate
  • 00:21:17
    experience and employee experience
  • 00:21:20
    because for the candidates as Olivia
  • 00:21:22
    said they're just excited to have this
  • 00:21:24
    sort of unbiased system that's available
  • 00:21:26
    to them 24/7 conversely for employees
  • 00:21:29
    they're just excited to not have to do
  • 00:21:31
    these recruiting calls many of which are
  • 00:21:33
    with people they'll never speak to again
  • 00:21:35
    right so just those like these like high
  • 00:21:37
    NPS outcomes the sort of um intuitive uh
  • 00:21:41
    thinking of a lot of the customers is
  • 00:21:42
    like well it's lower price but probably
  • 00:21:44
    a lower NPS experience and it's not it's
  • 00:21:47
    actually lower price and a higher NPS
  • 00:21:49
    experience in many cases right you also
  • 00:21:51
    talked about a few characteristics just
  • 00:21:53
    to kind of crystallize that in terms of
  • 00:21:55
    where we're seeing these AI agents be
  • 00:21:57
    successful versus not can you just speak
  • 00:21:59
    to those yeah so definitely I think like
  • 00:22:01
    the lowest hanging early early fruit I
  • 00:22:04
    guess to grab would be these businesses
  • 00:22:06
    that are already paying for a call
  • 00:22:07
    center okay cuz they're already spending
  • 00:22:08
    a lot of money on it and it's already a
  • 00:22:10
    pain point for them call centers are
  • 00:22:12
    notoriously High turnover they're hard
  • 00:22:14
    to manage so most businesses honestly
  • 00:22:17
    probably want to get rid of that if they
  • 00:22:18
    can you know the models are good now
  • 00:22:20
    they're just getting better and better
  • 00:22:22
    every month so I think we're still in a
  • 00:22:24
    world where when the call has kind of a
  • 00:22:27
    constrained uh process and outcome um
  • 00:22:31
    businesses are more comfortable so for
  • 00:22:32
    example The Voice agent knows going in
  • 00:22:35
    my goal is to book an appointment with
  • 00:22:36
    this person versus maybe like an
  • 00:22:38
    amorphus how do you how do you even
  • 00:22:40
    measure if this call was successful
  • 00:22:42
    we've seen some AI therapy voice agents
  • 00:22:44
    which are amazing and I think are
  • 00:22:46
    improving all the time but in that case
  • 00:22:48
    it's much harder for The Voice agent to
  • 00:22:50
    know at the end of the call did I do a
  • 00:22:51
    good job it's much harder for the
  • 00:22:53
    company to know at the end of the call
  • 00:22:55
    did it you know complete the objective
  • 00:22:57
    and then I would say this gets back to
  • 00:23:00
    the constrainted point but even though
  • 00:23:02
    the voice agent is still probably doing
  • 00:23:04
    better than your human agents most
  • 00:23:06
    businesses don't want to pay that much
  • 00:23:08
    for it because it is AI and they see it
  • 00:23:10
    as a way to cut cost so in these
  • 00:23:13
    verticals where you can offer it to
  • 00:23:14
    customers at I don't know 70% discount
  • 00:23:17
    to what they were paying before that has
  • 00:23:20
    been I would say very very powerful as
  • 00:23:22
    well and then I would say the other kind
  • 00:23:25
    of main factor is these verticals where
  • 00:23:28
    where it really is crucial for the
  • 00:23:30
    business to answer the call uh but for
  • 00:23:33
    the end consumer if there's a mistake
  • 00:23:36
    here or there it's okay so like a
  • 00:23:38
    restaurant order versus you know getting
  • 00:23:40
    a healthc care diagnosis uh there's like
  • 00:23:42
    a little bit of a different level of
  • 00:23:44
    urgency I would say this is where I
  • 00:23:46
    think the capabilities are just going to
  • 00:23:47
    get better and better faster than we
  • 00:23:48
    appreciate you know this with the
  • 00:23:50
    language models they're prone to
  • 00:23:51
    hallucination and there are certain
  • 00:23:53
    conversations like the therapy one that
  • 00:23:54
    benefit from the hallucination there are
  • 00:23:56
    other conversations like negotiating
  • 00:23:58
    something where there's a price and like
  • 00:24:01
    exactness matters they probably don't
  • 00:24:03
    benefit as much from hallucination so
  • 00:24:05
    now starting to think of um voice models
  • 00:24:07
    plus reasoning models you have the
  • 00:24:09
    ability to sort of narrow and
  • 00:24:11
    circumscribe the hallucinations to a
  • 00:24:13
    zone that you like and need as a
  • 00:24:14
    business right versus just having it be
  • 00:24:16
    you know having to a lot of systems
  • 00:24:17
    around it to control it right and since
  • 00:24:19
    we are in some cases taking on things
  • 00:24:22
    that previously were done by humans how
  • 00:24:24
    do you think about pricing or what have
  • 00:24:26
    we learned there are you are you seeing
  • 00:24:28
    most companies just basically replicate
  • 00:24:29
    the pricing models of the previous
  • 00:24:31
    version or are there new pricing models
  • 00:24:33
    that are coming up yeah it's early it's
  • 00:24:36
    changing every month and I would say
  • 00:24:37
    that's maybe the number one question
  • 00:24:39
    that we get from companies is how should
  • 00:24:41
    I price how do you see other companies
  • 00:24:43
    in the space pricing I think we've seen
  • 00:24:45
    a few models that are starting to work
  • 00:24:47
    or that people are experimenting with so
  • 00:24:49
    the most obvious one is you just charge
  • 00:24:51
    per minute so you can calculate kind of
  • 00:24:53
    an hourly rate for The Voice agent
  • 00:24:55
    similar to like what you would pay a
  • 00:24:57
    human there's a couple maybe wrinkles
  • 00:24:59
    here one would be a lot of these
  • 00:25:02
    customers are informed enough to know
  • 00:25:04
    that the underlying technology is
  • 00:25:05
    getting cheaper so they will come to you
  • 00:25:07
    and say hey why am I still paying 30
  • 00:25:09
    cents per minute when your costs have
  • 00:25:11
    gone down and you're probably just
  • 00:25:13
    taking all of that in margin and then as
  • 00:25:16
    these spaces get more competitive it's
  • 00:25:18
    very easy then for a newcomer to come in
  • 00:25:22
    and say hey I'm going to only charge 5
  • 00:25:24
    cents per minute and just kind of
  • 00:25:25
    undercut you based on that the other
  • 00:25:28
    thing about the price per minute model
  • 00:25:30
    is it really just puts your value as a
  • 00:25:33
    platform solely on the phone calls which
  • 00:25:35
    again are kind of commoditizing versus
  • 00:25:37
    like the other software that you're
  • 00:25:39
    building around the phone call so I
  • 00:25:41
    would say as a result of that we've seen
  • 00:25:42
    a lot of companies evolve from just
  • 00:25:44
    doing price per minute to some sort of
  • 00:25:46
    platform fee um maybe it's per month
  • 00:25:49
    maybe it's per module where the customer
  • 00:25:51
    is also paying for things that they get
  • 00:25:53
    in addition to the voice agent there's
  • 00:25:55
    been a few more creative pricing
  • 00:25:57
    experiments we've seen as well the
  • 00:25:59
    recruiting one is a is a good example
  • 00:26:02
    where in these cases where the voice
  • 00:26:03
    agent is a co-pilot to the human you can
  • 00:26:06
    almost charge per human that is using
  • 00:26:08
    the voice agent like a per seat SAS
  • 00:26:11
    model almost so for a human recruiter it
  • 00:26:14
    might save them I don't know 5 10 hours
  • 00:26:17
    per week of doing interviews and so you
  • 00:26:19
    can charge $500 $1,000 per recruiter per
  • 00:26:23
    month and then the last one and maybe
  • 00:26:26
    the most experimental one is outcome
  • 00:26:28
    based pricing which I feel like is a
  • 00:26:30
    question across all of AI right now and
  • 00:26:35
    are we moving towards that version of
  • 00:26:37
    the world now so maybe it's $5 per
  • 00:26:40
    appointment booked maybe it's 5% of the
  • 00:26:43
    booking value if you get it right
  • 00:26:46
    obviously you are then tying your value
  • 00:26:49
    most clearly to the value that you're
  • 00:26:51
    generating for the
  • 00:26:53
    business but we're interested to see how
  • 00:26:56
    those scale for Enterprises cuz I think
  • 00:26:57
    a lot of Enterprises are maybe nervous
  • 00:27:00
    to commit to that kind of payment
  • 00:27:01
    structure especially if they're not sure
  • 00:27:03
    exactly what kind of volume they're
  • 00:27:05
    going to be driving through it
  • 00:27:06
    interesting so you're seeing that last
  • 00:27:07
    one kind of start to have legs but start
  • 00:27:09
    to have legs but early I I mean I think
  • 00:27:12
    similar to what we've seen in kind of
  • 00:27:14
    the SAS landscape like not every company
  • 00:27:16
    pric is the same it depends on the End
  • 00:27:18
    customer it depends on the vertical it
  • 00:27:19
    depends on the features that you're
  • 00:27:21
    offering my gut is that we'll see some
  • 00:27:23
    combination of like the usage based per
  • 00:27:26
    per call pricing combined with some sort
  • 00:27:28
    of broader platform or outcome or seat
  • 00:27:31
    based pricing so it won't just be one
  • 00:27:33
    model um but it's very early days still
  • 00:27:36
    yep since we're early days what's your
  • 00:27:38
    instinct about Moes right that's as you
  • 00:27:40
    mentioned like that's true across the AI
  • 00:27:42
    ecosystem not just voice yeah but where
  • 00:27:44
    do you see Moes potentially arising in
  • 00:27:47
    this sphere I see Moes in a couple ways
  • 00:27:50
    so one would be Integrations and and
  • 00:27:52
    this is I think why we're especially
  • 00:27:54
    excited about these more vertically
  • 00:27:55
    focused voice agents it's not going to
  • 00:27:57
    make sense for open AI to go integrate
  • 00:28:00
    with every longtail you know
  • 00:28:03
    transportation management software that
  • 00:28:05
    a fleet company is going to or Freight
  • 00:28:07
    compan is going to be able to need to
  • 00:28:08
    run their you know Fleet of of trucks on
  • 00:28:11
    a voice agent uh product and similarly
  • 00:28:14
    UI like open aai and and other companies
  • 00:28:17
    have a pretty set you know um system for
  • 00:28:20
    interaction right now that doesn't work
  • 00:28:22
    the way that many of these like heavily
  • 00:28:25
    Legacy businesses uh want to be able to
  • 00:28:28
    operate one of the types of Moes that
  • 00:28:31
    has been the most intriguing for us I
  • 00:28:35
    would say especially for Enterprises
  • 00:28:37
    it's kind of this self-improving data
  • 00:28:39
    mode so if you are going to take over
  • 00:28:42
    calls for say a large bank they have a
  • 00:28:45
    certain way that they want those to be
  • 00:28:47
    done and so you're not going to plug in
  • 00:28:49
    a voice agent and have 100% NPS on day
  • 00:28:52
    one it's going to take months and months
  • 00:28:54
    of training calls to make that better
  • 00:28:56
    and so you as a voice agent provider if
  • 00:28:59
    you get in early uh benefit from having
  • 00:29:02
    all that special proprietary data that
  • 00:29:05
    just gives you months of a head start
  • 00:29:07
    for anyone else who has to come along
  • 00:29:09
    and go through that entire onboarding
  • 00:29:11
    and integration and training process and
  • 00:29:14
    so I think the hope for a lot of these
  • 00:29:15
    vertical voice companies is that they
  • 00:29:18
    will be able to use the call data either
  • 00:29:20
    per customer or anonymized across a
  • 00:29:23
    customer set to make the model better
  • 00:29:25
    and better over time which will kind of
  • 00:29:27
    increase their modes versus the
  • 00:29:29
    horizontal players if that's true are
  • 00:29:32
    you seeing uh AI voice companies kind of
  • 00:29:34
    raced to be the first mover in the same
  • 00:29:37
    way that we saw in the previous
  • 00:29:38
    generation I mean we talked about apps
  • 00:29:39
    like uber where it's like you know you
  • 00:29:41
    have to get the customers quickly and
  • 00:29:43
    you maybe have to blow a lot of cash to
  • 00:29:44
    get there but you you know you re that
  • 00:29:47
    back in later yeah I mean it's it's
  • 00:29:48
    certainly going to be less expensive
  • 00:29:50
    than Uber to go win the market but yes I
  • 00:29:52
    mean as Ben said many times you have to
  • 00:29:53
    both make a product people want and then
  • 00:29:55
    you have to go take the market get from
  • 00:29:57
    zero Market shared all the market share
  • 00:29:58
    so it is incredibly competitive that's
  • 00:30:01
    why we're seeing a lot of pressure on
  • 00:30:02
    pricing and pricing is such an important
  • 00:30:04
    topic in the ecosystem right now it will
  • 00:30:07
    definitely be a foot race and I do think
  • 00:30:09
    that there to Olivia's point there will
  • 00:30:10
    be some really interesting voice native
  • 00:30:12
    Moes you know you could imagine a a voel
  • 00:30:16
    investor for our firm where it can give
  • 00:30:18
    the the firm's pitch the way that Mark
  • 00:30:20
    can and it can negotiate the way that
  • 00:30:22
    Martin can and it can you know assess
  • 00:30:24
    the landscape the way Olivia can like
  • 00:30:26
    there's some specialization
  • 00:30:28
    opportunities there that feel very
  • 00:30:29
    native to voice you know on the other
  • 00:30:31
    hand Integrations Network effects you
  • 00:30:34
    know scale all the traditional modes
  • 00:30:35
    will be at play as well yeah and I do
  • 00:30:37
    think the go to market will depend on
  • 00:30:39
    the vertical there's say restaurants
  • 00:30:42
    Home Services businesses you know spas
  • 00:30:45
    or nail salons those are kind of very
  • 00:30:48
    fragmented long tail of of smaller
  • 00:30:51
    players and so in those cases the data
  • 00:30:54
    does exist in each of their hands
  • 00:30:56
    whereas again Banks or financial
  • 00:30:57
    institutions is maybe one where there's
  • 00:31:00
    a lot of concentration in a few players
  • 00:31:02
    and if it takes you 6 n months to get
  • 00:31:04
    them on board great versus the salon
  • 00:31:07
    restaurant Home Services voice agent
  • 00:31:09
    provider might be much more focused on
  • 00:31:11
    getting a thousand customers uh within
  • 00:31:14
    the same time frame you know I also
  • 00:31:16
    think an interesting thing to think
  • 00:31:17
    about is just people building personal
  • 00:31:19
    relationships with AIS for example like
  • 00:31:21
    you don't have a relationship with JP
  • 00:31:23
    Morgan you know you sort of have a more
  • 00:31:25
    of a relationship with your wealth
  • 00:31:27
    manager who happens to work at that firm
  • 00:31:29
    Y which is why when many of them leave
  • 00:31:31
    big platforms they take their customers
  • 00:31:32
    with them you know realtor is another
  • 00:31:34
    great example so there are cases where
  • 00:31:36
    the AI May build this deep personal
  • 00:31:38
    connection with a person and the person
  • 00:31:40
    wants to have that connection and that
  • 00:31:42
    then creates a moe it's a great point
  • 00:31:43
    and so far we've talked a lot about B2B
  • 00:31:46
    applications but that's you know brings
  • 00:31:47
    us right to Consumer applications can we
  • 00:31:50
    talk a little bit about what you're
  • 00:31:51
    seeing there maybe the difference
  • 00:31:53
    between you know what you're seeing in
  • 00:31:54
    B2B and B Toc I would say B2B voice
  • 00:31:57
    agents are
  • 00:31:58
    more obvious than consumer or b2c voice
  • 00:32:01
    agents just because again it's the use
  • 00:32:03
    case of replacing existing spend on
  • 00:32:07
    humans on the phone for businesses for
  • 00:32:09
    consumers maybe the coraly there would
  • 00:32:12
    be these high cost uh hardto access
  • 00:32:16
    services that can now be performed by a
  • 00:32:18
    voice agent instead of a human therapy
  • 00:32:20
    and mental health support is one of
  • 00:32:22
    those edtech is another big one language
  • 00:32:25
    learning uh teaching your kid how to
  • 00:32:27
    read teaching your kid how to do math
  • 00:32:29
    which I think a lot of parents struggle
  • 00:32:31
    with um coaching you know how to have
  • 00:32:35
    hard personal conversations those are
  • 00:32:37
    all areas where we've seen a lot of
  • 00:32:38
    voice agent uh pickup the main I think
  • 00:32:41
    open question on the consumer voice
  • 00:32:43
    agents have been when a Chachi BT or
  • 00:32:46
    soon a Claud can do a pretty good job
  • 00:32:49
    with a lot of those basic consumer use
  • 00:32:51
    cases where are the verticals or use
  • 00:32:54
    cases where you need either a
  • 00:32:55
    specialized model or a specialized
  • 00:32:57
    interface space to provide most of the
  • 00:32:59
    value the best models maybe are right
  • 00:33:02
    now being held by open AI versus being
  • 00:33:04
    available via API for any kind of
  • 00:33:07
    Standalone voice agent company to
  • 00:33:08
    utilize the biggest and best consumer
  • 00:33:10
    companies are often surprises and are
  • 00:33:13
    non obvious and so I my gut is that
  • 00:33:17
    whatever we see working in consumer
  • 00:33:18
    voice is going to be something that it's
  • 00:33:21
    hard to sit here and speculate on it'll
  • 00:33:23
    be extremely obvious yes and it'll be
  • 00:33:25
    like a massive company um when we see it
  • 00:33:28
    not until then exactly yeah what do you
  • 00:33:32
    think in terms of the incumbent's
  • 00:33:34
    potential to capture this consumer
  • 00:33:36
    Market whether it's you know Google or
  • 00:33:38
    Apple or you know are we seeing that you
  • 00:33:41
    know all of those YC companies or other
  • 00:33:43
    companies that we're involved with are
  • 00:33:44
    really getting further ahead I have a
  • 00:33:47
    bit of a point of view on this like I
  • 00:33:48
    think that the the incumbents it's just
  • 00:33:51
    such a daily demonstration of how far
  • 00:33:53
    behind they are when you both have
  • 00:33:55
    Google home in your home and you've got
  • 00:33:58
    chat GPT in your pocket yeah you know my
  • 00:34:01
    children try to ask Google home to tell
  • 00:34:03
    them stories in the same way that chat
  • 00:34:04
    GPT does and it just utterly utterly
  • 00:34:07
    fails and my children are you know their
  • 00:34:09
    first interaction with technology at
  • 00:34:11
    least deep interactions are happening
  • 00:34:13
    via models not via search engines so one
  • 00:34:16
    I think that there's uh it's just a sort
  • 00:34:18
    of day-to-day experience of a lot of
  • 00:34:20
    people is that the incumbents are pretty
  • 00:34:21
    far behind in this area then the second
  • 00:34:24
    I think we've talked a bunch about this
  • 00:34:25
    is that there are a lot of sort of I
  • 00:34:27
    don't know uncomfortable or impolite
  • 00:34:29
    aspects of The Human Experience which
  • 00:34:31
    incumbents are just structurally
  • 00:34:34
    designed to never discuss right you know
  • 00:34:36
    corporations sort of committees lawyers
  • 00:34:39
    like these big companies have a hard
  • 00:34:41
    time shipping opinionated products at
  • 00:34:44
    least opinionated in the way that many
  • 00:34:46
    of these voice models may need to be and
  • 00:34:48
    startups have no problem doing that now
  • 00:34:50
    there are you know counterpoints to it
  • 00:34:51
    like grock but I think that's very much
  • 00:34:54
    you know things that only a founder-led
  • 00:34:55
    big company can do versus the
  • 00:34:57
    traditional incumbents I think there's
  • 00:34:59
    you know one or two categories or use
  • 00:35:01
    cases where the calls have truly
  • 00:35:04
    commoditized or will commoditize and and
  • 00:35:06
    the user experience matters less and
  • 00:35:08
    like Google might take those for example
  • 00:35:11
    they recently launched the ability to
  • 00:35:13
    call you know a restaurant get
  • 00:35:15
    availability and then kind of come back
  • 00:35:17
    to you and give you the options like if
  • 00:35:18
    you can add that as a button on a Google
  • 00:35:20
    search like yes uh that probably makes
  • 00:35:23
    sense to do through them but are they
  • 00:35:25
    going to build the first AI native
  • 00:35:28
    personal assistant that works across all
  • 00:35:30
    of your products and all of your
  • 00:35:31
    information sources probably not I would
  • 00:35:33
    say I think that any and all of the
  • 00:35:36
    calls that the incumbents end up doing
  • 00:35:38
    which will be some volume are probably
  • 00:35:40
    not going to be the type of calls that
  • 00:35:42
    are going to support a large and
  • 00:35:44
    exciting Standalone new startup they
  • 00:35:46
    will use the new technology to kind of
  • 00:35:48
    extend their dominance of the categories
  • 00:35:50
    they've always dominated which is fine
  • 00:35:53
    all of the new categories they're just
  • 00:35:55
    going to be utterly unable to compete in
  • 00:35:56
    or at least that's been the historic
  • 00:35:58
    pattern and you know I think a good
  • 00:35:59
    question is if models are the new front
  • 00:36:01
    end for the internet is search even a
  • 00:36:04
    meaningful primitive you know are they
  • 00:36:06
    going to then extend their dominance of
  • 00:36:07
    a category that loses relevancy for the
  • 00:36:09
    next generation of consumers and
  • 00:36:10
    businesses and I think your point about
  • 00:36:12
    even the term opinionated is so
  • 00:36:14
    important here because I would argue
  • 00:36:16
    voice is a platform that you know we
  • 00:36:19
    Intuit it to be more opinionated or we
  • 00:36:21
    need to be more opinionated than let's
  • 00:36:23
    say because interesting people are
  • 00:36:24
    opinionated exactly exactly and I'm even
  • 00:36:27
    thinking through like I mean I might be
  • 00:36:28
    going too far here but some some of the
  • 00:36:31
    old kpis that you would see for
  • 00:36:32
    something like search or an application
  • 00:36:34
    it's you know may not even be the same
  • 00:36:36
    for voice like you can imagine you know
  • 00:36:38
    the Magic Moment might be like time to
  • 00:36:40
    laugh like how quickly can you get
  • 00:36:41
    someone to laugh or to cry or to you
  • 00:36:44
    know not intentionally but to really
  • 00:36:45
    engage with a Model A voice model that
  • 00:36:48
    just wouldn't necessarily occur with
  • 00:36:50
    with text so I think the average
  • 00:36:52
    consumer would in their head like a Siri
  • 00:36:55
    doesn't even compete with a che CBT
  • 00:36:58
    voice mode or something like that
  • 00:36:59
    because they're just such different
  • 00:37:01
    feelings that you get as a user when you
  • 00:37:02
    are using them you know I think the
  • 00:37:04
    other interesting part of this is that
  • 00:37:05
    there are cultures in which being a
  • 00:37:07
    little disagreeable a little sarcastic
  • 00:37:09
    is actually highly preferred and that's
  • 00:37:10
    the way that you were supposed to build
  • 00:37:12
    trust and interact with people you know
  • 00:37:13
    I know that um the British culture is a
  • 00:37:15
    little bit like this way even East Coast
  • 00:37:17
    culture you know we were having a laugh
  • 00:37:18
    a few weeks ago about like we need you
  • 00:37:20
    know chat GPT voice East Coast mode yes
  • 00:37:23
    where it's just like very short it
  • 00:37:25
    doesn't like suffer fools you know it
  • 00:37:27
    says no like no totally when you think
  • 00:37:30
    about your friends it's like a m you
  • 00:37:31
    don't have friends or some people do but
  • 00:37:33
    most people don't have friends that are
  • 00:37:34
    just like At Your Service yes you know
  • 00:37:36
    that there's some banter there's some
  • 00:37:38
    they have an opinion this is actually
  • 00:37:40
    this gets at what we're looking for in
  • 00:37:42
    voice companion products or but even any
  • 00:37:44
    Consumer Voice agent like there has to
  • 00:37:46
    be some friction if it's like too easy
  • 00:37:49
    to build the relationship if they're
  • 00:37:50
    always saying yes to you if they're not
  • 00:37:52
    giving you the brutally honest feedback
  • 00:37:53
    then it's like it gets old quickly
  • 00:37:56
    there's no value for you a consumer to
  • 00:37:58
    just have like a a yes A Yes Man or yes
  • 00:38:00
    when follow yes exactly following you
  • 00:38:04
    around all the time and so we actually
  • 00:38:06
    get very excited by Founders who are
  • 00:38:08
    opinionated in how to build the voice
  • 00:38:11
    agent as its own character its own
  • 00:38:13
    personality that the user is forming a
  • 00:38:15
    bond with versus uh The Voice agents
  • 00:38:19
    we've had in the past where the user is
  • 00:38:20
    kind of treating them as a machine that
  • 00:38:22
    they're handing basic tasks to right
  • 00:38:25
    that's right trust has to be earned and
  • 00:38:27
    if the models don't design for that
  • 00:38:28
    they're never going to get to their full
  • 00:38:30
    potential that's a great Point well as
  • 00:38:32
    we work toward those kind of products is
  • 00:38:35
    there anything you'd like to leave the
  • 00:38:36
    listeners with in terms of what's on the
  • 00:38:38
    horizon what you're excited about maybe
  • 00:38:40
    also where you'd like to see Founders
  • 00:38:42
    direct their attention I think one of
  • 00:38:44
    the things that has been really
  • 00:38:45
    interesting and and maybe it's just the
  • 00:38:47
    standard Tech platform shift but we're
  • 00:38:49
    seeing Founders that are maybe new to an
  • 00:38:52
    industry but spend a couple months going
  • 00:38:53
    really deep uh able to build the most
  • 00:38:56
    powerful and kind of the highest growth
  • 00:38:58
    and the highest inflection products and
  • 00:39:01
    that's just because I think the rules of
  • 00:39:02
    the game are changing and the type of
  • 00:39:05
    and power of products you can build is
  • 00:39:08
    also above anything that we've ever seen
  • 00:39:11
    and so if you move quickly in many ways
  • 00:39:12
    like shipping fast becomes the mo and
  • 00:39:15
    you can kind of catch up on everything
  • 00:39:17
    else like the industry expertise the
  • 00:39:19
    networks the knowledge base the
  • 00:39:21
    resourcing all of that and so I would
  • 00:39:24
    say that has been like one of the areas
  • 00:39:27
    where we get most excited Founders that
  • 00:39:29
    maybe have only been in the industry for
  • 00:39:31
    6 months a year even less but are
  • 00:39:34
    becoming quickly opinionated about what
  • 00:39:36
    they need to build and probably most
  • 00:39:37
    importantly just building really quickly
  • 00:39:39
    and testing getting feedback and going
  • 00:39:41
    from there yeah so two things one if
  • 00:39:43
    you're building the space talk to us we
  • 00:39:45
    definitely want it and you know the
  • 00:39:46
    weirder the better um and then two like
  • 00:39:50
    a prompt that we've discussed with a lot
  • 00:39:52
    of AI Founders is just what is the
  • 00:39:55
    incredibly mind-bogglingly expensive Ive
  • 00:39:57
    version of your product so if you're
  • 00:39:59
    charging a lot of consumers $20 a month
  • 00:40:02
    or $100 a month like what would the ,000
  • 00:40:04
    a month or $10,000 a month skew look
  • 00:40:06
    like I think the same is very true in
  • 00:40:08
    voice yes there's going to be high
  • 00:40:10
    volume use cases that we want to
  • 00:40:11
    actually replicate you know or
  • 00:40:13
    substitute voice AI models for but what
  • 00:40:16
    are the most sensitive most precious
  • 00:40:18
    most high value conversations that are
  • 00:40:20
    happening in the Enterprise right and
  • 00:40:21
    can you attack those and you know what
  • 00:40:24
    what price would you charge for those
  • 00:40:26
    might be $100,000 in interaction maybe
  • 00:40:28
    that's a little extreme but as a product
  • 00:40:30
    design sort of exercise why not it's a
  • 00:40:33
    great prompt to leave people with thank
  • 00:40:35
    you both so much thank you thank
  • 00:40:37
    [Music]
  • 00:40:41
    you thank you so much for listening to
  • 00:40:44
    the a16z podcast if you've made it this
  • 00:40:47
    far don't forget to subscribe so that
  • 00:40:49
    you are the first to get our exclusive
  • 00:40:51
    video content or you can check out this
  • 00:40:54
    video that we've hand selected for you
  • 00:41:03
    [Music]
  • 00:41:09
    he
  • 00:41:11
    [Music]
Tags
  • AI
  • Voice Technology
  • Business Applications
  • Emotional AI
  • Consumer Interaction
  • Tech Innovations
  • Sectors
  • Voice Companions
  • Market Trends
  • Founders Guidance