Voice AI Ventures Towards Localization in Africa and the Middle East

Customer support and service are among the hottest sectors in voice AI right now. But building a product that sounds human and responds without noticeable delay turns out to be much harder in some markets than others — and most of the major players weren’t built with Africa and the Middle East in mind.

AethexAI, a startup established last year to address this gap, has successfully secured $3 million in pre-seed funding. The investment round was led by 4DX Ventures, with support from Enza Capital, Dorm Room Fund, Mojo Ventures, and Stanford GSB 26 Fund. Notably, the funding also includes contributions from individual investors such as Stanford faculty and AI experts from Anthropic.

Instead of relying on existing orchestration platforms like Vapi and LiveKit, AethexAI took the initiative to create its own model and orchestration framework tailored for the distinct dialects of English, French, and Arabic prevalent in its target regions. This strategic choice stemmed from the specific challenges of operating in these areas.

The startup is launching its platform for enterprises so they can experience its technology firsthand and register for its services. In addition, AethexAI offers APIs and SDKs for developers eager to experiment with its models.

Co-founded by Mariama Diallo and Ayooluwa Odemuyiwa, AethexAI brings together diverse expertise. CEO Diallo previously held a role at Goldman Sachs and later joined Y Combinator-backed ModelML. CTO Odemuyiwa is an alumnus of Caltech and has experience from Meta and Stanford Business School before embarking on this venture. The duo envisioned creating solutions focused on emerging markets, prompting them to seek out viable opportunities.

Globally, businesses are rapidly adopting AI solutions to streamline operations. However, challenges arise. For instance, a call center in Egypt automated many of its calls but had to revert to traditional methods due to disappointing outcomes. Founders noted that many support centers in Africa often grapple with the task of hiring engineers capable of implementing cost-effective call automation.

“The latency and jitter encountered in automated calls in this region are substantial. If we had opted to be orchestration providers, we may have been required to use large external models, exacerbating latency issues. We understood that success necessitated creating smaller models to minimize latency throughout the process,” reflected Odemuyiwa in a conversation with ToolsMixAi.

Typically, AI labs invest millions in training large models and gathering extensive datasets. AethexAI approached this differently. Instead of pursuing the biggest models available, it created its own Kora series, which features parameters between 300 million and 1.7 billion. This size is intentionally limited compared to more extensive LLMs, which aligns with their goals.

The startup has utilized anonymized audio from a call center partnership for model training. Additionally, it distributed hard drives to various radio stations across Africa to gather further audio data. To manage costs effectively, the company engaged university students to assist in data annotation and correct pronunciation of local names. As a result, AethexAI currently handles over 17,000 calls daily.

On the business front, the startup emphasizes guiding clients unfamiliar with voice AI through a structured experience. They provide onsite demonstrations and workshops to help identify optimal automation use cases.

“We inform customers that we can’t accommodate every need right away. We’re a small team. When initiating discussions, we encourage them to select the most critical use case to focus on initially,” explained Diallo.

Although the company is open to opportunities in various sectors, it primarily addresses use cases related to debt collection, customer activation, and Know Your Customer (KYC) verification processes utilized by banks and telecommunications providers. AethexAI is hiring engineers on a contractual basis for local markets and establishing partnerships with telecom firms to facilitate telephony for voice AI calls. They have noted that off-the-shelf solutions do not suffice in this context.

Walter Baddoo, co-founder and managing partner at 4DX Ventures, indicates that the markets in Africa and the Middle East drastically differ from those targeted by most traditional voice AI firms.

“Companies in Africa and the Middle East experience approximately three times the call volume compared to those in the West, as voice communication remains the primary channel for customer interaction,” he stated. “Existing systems were designed for Western markets which feature high-end GPU capabilities, standardized English language environments, and common enterprise workflows found in the U.S. and Europe. This creates significant challenges in catering to dialects, code-switching, and informal speech, as well as working within their established telephony systems at accessible price points.”

In summary, while companies such as ElevenLabs, Deepgram, Sierra, and Cognigy expand their global presence, the markets they were originally designed for often differ from those they are entering. Startups like AethexAI are betting that the lack of tailored solutions for local dialects, combined with regional partnerships and customized infrastructure, signal a market opportunity that larger firms are either unwilling or unable to fulfill.