Level3 AI has closed a $13 million Seed Round round led by Sovereigns Capital, with participation from existing seed investors. The funding arrives roughly 18 months after our seed round and reflects a period of operational results that were, frankly, better than our own projections.
Why Sovereigns Capital
We spoke with 22 funds during this raise. Sovereigns Capital stood out for one reason that had nothing to do with valuation: they'd already backed three enterprise software companies operating in Southeast Asian markets, and their feedback on our product wasn't generic. When investors ask about Thai language support and then cite the specific challenges of polite register in customer service interactions, you know they've done real work in the space.
Their faith-driven investment philosophy also matters to how we build. We're not optimizing for short-term growth metrics at the expense of the customers we're supposed to be serving. Sovereigns Capital's portfolio companies tend to run at higher operating margins and lower churn than comparable SaaS businesses — which tells you something about the incentive alignment they bring.
What the Numbers Looked Like When We Started This Round
When Sovereigns Capital first looked at Level3 AI in Q3 2024, we had 14 enterprise customers across Singapore, Malaysia, and Vietnam. Our median customer was resolving 61% of inbound support queries without human intervention. The best customer — a regional logistics company managing last-mile delivery exceptions — was at 84% automated resolution, with average handle time down from 8.2 minutes to under 90 seconds.
We'd processed 4.1 million conversations by that point. The data let us say things in investor meetings that most early-stage AI companies can't: here's resolution rate by query category, here's where the agent still fails, here's how accuracy changes with conversation length, and here's what we changed in model version 2.4 that fixed the refund-request misclassification we saw in Bahasa Indonesia conversations.
Where the $13M Goes
The allocation breaks down into three areas. The largest share — roughly $6M — goes to engineering. We're hiring 18 engineers over the next 12 months, split between our Singapore headquarters and a new office in Hanoi, where we've already identified the first four hires from the VinAI and VNG alumni network. The Hanoi team will own the Vietnamese and Thai language models, where we have the most improvement headroom.
The second allocation, approximately $4.5M, covers the Indonesia and Philippines market launches. Both markets have distinct regulatory considerations around customer data handling — Indonesia's UU PDP came into effect in 2024, and the Philippines NPC has been actively issuing guidance on AI-assisted data processing. We're budgeting for local legal counsel and government engagement, not just sales and marketing.
The remaining $2.5M goes to infrastructure. We're migrating from a multi-tenant AWS architecture to per-customer isolated environments for enterprise accounts above a certain conversation volume. That's more expensive to run, but it's what the regulated sectors — banking, insurance, healthcare — require before they can sign.
The Market Bet We're Making
There's a version of the APAC conversational AI market where global players like Intercom and Zendesk's AI features are good enough for most enterprise customers in the region. We don't think that version exists, at least not for the next four to five years.
The reason is structural. A company serving 40 million customers across Indonesia, where 70% of queries arrive in Bahasa Indonesia mixed with regional slang and occasional Javanese phrases, cannot be served by a model trained primarily on English-language customer service data. The intent classification problem in code-switched APAC languages is genuinely unsolved by the general-purpose models, and the business consequences of misclassification — issuing a refund when a customer asked a product question, or escalating a simple tracking inquiry to a human agent — are significant at scale.
We've been building specifically for this problem since 2022. The Seed Round gives us the runway to do it properly, in more markets, before the window narrows.
What We're Not Changing
We're keeping our 14-day deployment commitment. Some investors suggested we allow more time for enterprise sales cycles and onboarding. We declined. The 14-day timeline is a forcing function that keeps our product team honest — if deployment takes longer, it means something in the product needs to get simpler. We've been shipping easier setup flows specifically because of that constraint.
We're also keeping our data-in-region policy. Customer conversation data stays within the AWS Singapore region unless a customer explicitly requests cross-border processing. Several larger enterprise deals in 2024 closed specifically because we could say that without caveats. As data sovereignty legislation tightens across APAC, this will matter more, not less.
A Note on the Timing
Closing a Seed Round in Q1 2025 means competing for investor attention in a market where AI companies are raising at multiples that require extraordinary growth to justify. We chose not to participate in that dynamic. Our valuation reflects our actual revenue, actual retention, and a realistic model of what we can build with $13M. We think that's the right foundation for a company that plans to be operating in this region for the next decade.
If you're an enterprise in Southeast Asia dealing with high support volume and limited multilingual coverage, we want to hear from you. Reach out to hello@level3-ai.com or book a demo on our contact page.