Neural Notes: Phonely's US$500,000 boost sets the tone for AI receptionists

Welcome back to Neural Notes, a column where I take a look at some of the most interesting (and sometimes weird) AI news of the week. In this edition: Melbourne startup Phonely lands US$500,000 from Silicon Valley accelerator, Y Combinator, for its AI receptionist.

Phonely was founded just eight months ago by former PhD researchers Will Bodewes and Nisal Ranasinghe. The AI-powered answering service aims to transform phone support for businesses. According to the company, its AI receptionist is able to handle up to a million calls simultaneously.

From there, Phonely can be hooked up to scheduling software and other existing internal platforms and route calls as needed. After a call, Phonely claims to be able to provide analytics as well as surface the most important information from a call to send straight to a company’s CRM, phone or inbox.

Phonely says it’s able to do this for 70% cheaper than a human receptionist.

Phonely wants to target personalisation and multi-language support

Personalisation is a key element of the Phonely platform, which claims to be able to use a company’s URL to build a custom AI receptionist in minutes.

It’s able to do this by extracting information from every subpage of the site to train the assistant. Soon this will include the specific languages used by the business.

At the time of writing Phonely is restricted to English, but it will be rolling out multiple languages over the next two weeks. The aim is for the company to support a variety of different voices and 80 languages by the end of June.

In the meantime, Phonely will be able to deploy the fresh injection from Y Combinator, which was announced just this week.

According to Phonely CEO Will Bodewes, the money will be poured primarily into expanding the team.

However, some will also be set aside for HIPAA compliance since Phonely deals with the likes of medical records, as well as something that you don’t often hear founders talk about when it comes to funds- marketing.

“A little bit of marketing sells a lot on development,” Bodewes said.

When it comes to future features, Bodewes said that Phonely is really focusing on improving the performance of the AI receptionist — not only processing a lot of information but then knowing how to act on that.

“One of the more challenging pieces is actually interfacing with the different softwares that businesses use, so that they can actually schedule appointments or have the correct workflows, the right integration so that it makes it easy for businesses to use,” Bodewes said.

“I think a second piece of the pie is just tracking and improving how these voice models work, and ensuring that they get accurate responses every time. It’s not a simple problem to augment the whole job of a receptionist over the phone, there’s a lot of different working pieces, a lot of different parts.”

Challenges and future plans

Phonely also wants to expand into other areas in the future, such as SMS.

“We’re already developing a lot of this AI technology to be able to do voice well — we can do a similar thing over SMS, right? So we can roll out that feature to be able to do voice calling, and then SMS responses to really give those businesses unique responses and interact with their customers better,” Bodewes said.

Unsurprisingly, getting to this point in just eight months was not without its technical challenges. For Phonely, latency was a significant hurdle to clear early on, wanting the AI to respond to customer interactions in under a second.

Model accuracy has also been critical, with the AI needing to be trained to prevent speech errors and understand the context of different interactions. This required extensive data training and continuous refinement.

“Getting the language models to interact like humans with scheduling software is tough. We’re essentially teaching a computer to use software with no user interface, and it has to learn instantly and do it correctly every time,” Bodewes said.

Solving these problems has been no easy task. Bodewes confessed to working 80-plus hour weeks over the past eight months. Bodewes says it’s because he loves what he does and because the team is committed to being the best.

He also advises other budding entrepreneurs to just get out there and do it, because there’s never a perfect time to start your business.

“If you feel like you have this burning desire and you want to go start a company and work crazy hours — if that sounds like it’s gonna be fun for you, because you’re really passionate about something — then go do it.”

Other AI news this week:

Ed Husic ‘announced’ a $17 million AI Adopt program for SMEs at an industry event. The only problem was that it was already announced back in December. It’s also very similar to a program announced by the coalition back in 2022… but theirs was for $44 million.
Australia just got its first AI board advisor, Alice Ing (get it?), over at the Real Estate Institute of NSW. Let’s hope the gimmick is a little less overt than that Polish AI CEO who was created to help sell luxury rum NFTs.
Sapia.ai is opening up its own API program on AWS marketplace
Former OpenAI board member Helen Toner has dished on why Sam Altman was fired as CEO in 2023. She wouldn’t get into everything, but one of the things she did allege was that Altman didn’t tell the board that he owned the OpenAI startup fund, despite claiming to have no financial interest in the company.
Speaking of Y Combinator and firing Altman — the accelerator’s co-founder took to X this week to refute claims that the OpenAI CEO was once fired from Y Combinator.
PwC just became OpenAI’s first resale partner.