Years ago, amidst growing scrutiny of WeWork’s inflated valuation, I met with co-founder Adam Neumann. He requested two meetings before our scheduled stage interview, first at a WeWork location in San Francisco and then at his New York office.
While I was well aware that the figures, particularly those from SoftBank, attributed to WeWork made little sense, I ultimately understood the appeal that drew investors to him.
Similarly, it is easy to grasp why individuals are captivated by Jordan Belfort, widely known as the “Wolf of Wall Street.” Belfort famously admitted guilt in 1999 to charges of securities fraud and money laundering connected to Stratton Oakmont, the Long Island brokerage where salespeople engaged in pump-and-dump schemes involving penny stocks, while Belfort amassed a Southampton estate, a 167-foot yacht, and at least one helicopter. (He wrecked the latter two while under the influence of Quaaludes and cocaine, as dramatically portrayed in Martin Scorsese’s renowned film adaptation of Belfort’s autobiography, “The Wolf of Wall Street.”)
One might assume that someone who served 22 months in prison and was mandated to pay $110.4 million in restitution to 1,513 victims would not be given another opportunity. The victims ultimately received “far less” than they were owed, TMZ reported in July, noting that Belfort concluded his court-ordered restitution obligations in late April of this year.
Nevertheless, Belfort, possessing an electrifying charisma, consistently reappears, including in recent years as a Miami-based sales consultant and, at one point, a host of cryptocurrency workshops at his Miami Beach residence. (A New York Times journalist attended a weekend workshop several years ago where entry cost one bitcoin or its then-equivalent of $40,000.)
Now, he is venturing into the intersection of his expertise—sales—and the most prominent global trend, AI, by developing a voice-based AI platform. The distinguishing feature, he claims, is its ability to infuse voices with persuasive tonality that helps close deals, for instance, by framing declarative statements as questions (which “fosters engagement”) and even occasionally whispering (which “suggests information scarcity”).
Belfort states that the platform, named Tone, is already being utilized by companies experiencing more sales leads than their human teams can manage. Now, in a move that could also broaden its training data, Tone is launching a self-service platform for content creators and authors wishing to convert their books into “expressive AI-generated narration.” Users can choose from Tone’s existing AI voices or, as a voice-cloning feature rolls out, create a replica of their own voice.
It is plausible to argue that the market is already saturated with AI voice startups and technologies, and Tone carries additional risks: Belfort’s co-founder in this venture is long-time investor Shervin Pishevar, who has a controversial history of his own. Furthermore, we cannot independently verify the claimed functionality of the technology (neither co-founder is an engineer).
However, if my experience with Neumann taught me anything, it’s that skepticism regarding financial figures rarely hinders a compelling sales pitch. When Belfort seeks investors to support this new undertaking, finding enthusiastic backers will likely not be difficult.
Below is an excerpt from our discussion, edited for brevity and clarity.
What is the nature of your creation?
It represents a highly distinctive method for voice synthesis, rooted in a system of influence and persuasion that I’ve been refining since the late 1980s. It’s not merely about the words you speak, but how you deliver them. When not interacting in person, body language’s role is transferred to tonality. There are approximately 15 exceptionally impactful tonalities that individuals employ constantly. I coined the term “goal-oriented communication” for this concept. If my aim is to be perceived as an authority, I strategically apply specific tonalities.
Has this system been primarily trained using your own voice?
Initially, yes. As someone without an engineering background, I approached it from first principles: How would I instruct a human to do this? One tonality at a time. I possess 30 years of exceptionally high-quality data, and I began training the model. It was a pivotal moment because it not only replicated my voice precisely but also perfectly captured the tonality. Subsequently, I taught it a second tonality, and then how to connect them. The fundamental element in all of this is data. Even a minimal amount of flawed data can significantly undermine the progress achieved.
How did your collaboration with Shervin Pishevar originate?
Shervin and I have been close personal friends for an extended period, and we were having dinner one evening. I had recently been approached by an AI company claiming to have an AI capable of sales. I listened to it; it was terrible. But it planted the idea in my mind. I initially thought I would just teach existing models how to sell, but I couldn’t find a voice that was adequate. Necessity, as they say, is the mother of invention, so we developed our own.
Who constitutes your client base, and how do they utilize your service?
Our clients include law firms and health insurance providers. We are heavily involved in polling and the automotive sector. It can guide you through an entire health insurance purchase. It asks qualifying questions, such as your desired insurance type, budget, and ages. Then it selects a plan and elucidates its features and benefits. Our latency is approximately 600 milliseconds. If the payment is too high, it will suggest lowering it and explain how to increase your deductible. If you dislike the plan, it will switch you to an alternative. It manages the entire sales process, literally.
How does the AI manage a sales conversation that deviates from the predetermined script?
It avoids saying “That’s a great question” unless it genuinely is. I find that annoying; it’s overly ingratiating. In any sales interaction, there’s an ideal path, how we’d envision the perfect dialogue, with every word potentially mapped out. However, people have inquiries and concerns that lead the conversation off this straight line. So the challenge lies in a model that can sustain that perfect conversation, adeptly manage any curveballs, respond eloquently, and return to the main path, without damaging rapport or engaging in hallucination—or, as a salesperson would, lying. We have trained the model to accomplish this.
Has Tone generated any revenue thus far? Have you secured funding for it?
Revenue is probably around a million dollars—perhaps $700,000. We’ve secured friends and family and seed funding, totaling approximately $10 million.
I wish to address the obvious concern. You have a fraud conviction. Pishevar has faced allegations of misconduct that he denies. Why should a financial institution, a hospital, or a consumer brand place their trust in this product coming from the two of you?
Regarding myself, all I can say is that I did make errors. That was a very, very long time ago, and I have dedicated the past two decades teaching individuals worldwide. Companies globally have engaged me to impart sales and persuasion techniques. If certain companies choose not to collaborate with us, that is their prerogative. However, I believe more than enough will recognize the value of what we have created.
What is Tone’s conversion rate compared to a typical human sales force?
I still don’t believe it’s quite as effective as a top-tier human performer. A leading salesperson in the field would outperform it call for call. I anticipate that this will change within the next six months to a year. To enable a model to respond this rapidly, one does not employ frontier models. We utilize our own very compact, lightweight models, so their intelligence isn’t quite as extensive. I do think it will significantly outperform an average closer.
Where is it most effectively deployed? In opportunities involving lower-value sales? I would imagine that for higher monetary stakes, customers would demand more human interaction.
I don’t perceive it as a tool to replace human personnel. What I observe most often with clients is a simple lack of available human resources. They might generate far more leads than they have staff to call, or they require coverage at varying times of day or in diverse languages. I genuinely believe there will always be a demand for exceptional human salespeople.
You are in competition with companies like ElevenLabs and Cartesia, among many other voice AI firms. Is Tone your competitive advantage?
I don’t think sounding human is the core of this endeavor. They will all achieve human-like voices within six months to a year. Our distinction lies in the application of tonality: the intelligence behind when it selects a particular tonality. We hold a patent for it. ElevenLabs is an excellent company, and Cartesia as well, but their voices were not suitable for our needs.
The FCC ruled in 2024 that AI voices are subject to robocall regulations under the Telephone Consumer Protection Act. How does outbound sales calling function under these rules?
We exclusively make calls to opted-in individuals. You would not use AI for cold calling. We would prohibit that; you would face lawsuits from every TCPA attorney globally. It is either for opted-in leads or existing customers. The exception is polling, which has an exemption allowing calls to private residences without prior opt-in.
How do you ensure consent and prevent unauthorized impersonation in voice cloning?
We could launch the cloning feature tomorrow, but we haven’t. If you create a clone within our system, you speak into the computer, which identifies the creator, and that individual provides signed consent. If someone wishes to upload audio, which we advise against, a live voiceprint would need to be matched in real-time.
Who developed the technology? You are not an engineer, nor is Pishevar. I noticed a name on the patent, Jordan Anthony Metzner.
No. Our VP of engineering is Ben Sussman, a brilliant engineer. I serve as the ultimate arbiter and implementer of what achieves perfection in sound. I will prototype something in the most unconventional manner, and my engineers then rebuild it entirely from scratch.
Do you charge on a per-minute basis, or according to the results clients achieve?
It’s a combination. For streaming, it’s per minute. The content platform will operate on monthly subscriptions. And for some clients, the majority of our revenue is generated through success-based metrics.
Is this an autonomous company, or is it intended to be integrated within a larger AI enterprise?
If a suitable and substantial offer presented itself, and I deemed it sensible, who wouldn’t consider a discussion? However, my current strategy is to secure a significant amount of funding. I don’t believe we need to own data centers, but we require sufficient reserved computing power. I wouldn’t anticipate selling it in the near future, but the future is uncertain.
Paramount+ released a documentary about you this summer. Did you view it?
I did not. Believe it or not, I abstain from watching or reading anything about myself, whether positive or negative. I never read comments. When you are a public figure, you must live by your inner convictions. I operate based on my own internal moral compass.