EN
Back to the archive

The archive · AI & Models · Product decision · 2023–2025

Leaping AI's self-improving voice bet: $4.7M seed after YC W25

German-founded YC W25 voice AI reports 10,000 calls a day and $1M ARR by betting agents that learn from every call, not scripted bots.

Leaping AI

The betThat call centers will trust voice AI only if it demonstrably improves from every call — self-improving agents, not scripted bots, take over real inbound volume.Live

What the business is

Leaping AI sells self-improving voice AI agents that answer inbound phone calls for call centers and customer-service teams across travel, household services, insurance and real estate, handling calls over phone and web.

Starting capital$500K standard YC W25 investment (early 2025), then a $4.7M seed led by Nexus Venture Partners with Y Combinator cofounder Paul Graham, Shopify COO Kaz Nejatian, Ritual Capital, Pioneer Fund and Orange Collective participating.

How it started

Kevin Wu, then a Boston Consulting Group consultant in Berlin, quit after recalling the day he spent taking calls as an Amazon intern: "It was such an ungrateful job." He founded Leaping AI in Germany in 2023 with CTO Arkadiy Telegin. German investors turned them down twice — "there is no venture capital for technical founders with unproven models in early stage," Wu said — so at the end of 2024 they applied to Y Combinator, were accepted into the Winter 2025 batch and relocated to San Francisco.

What happened

Revenue doubled in the first two months in San Francisco after a year of German growth. After the batch, Wu ran five straight days of fourteen 30-minute investor meetings and closed a $4.7M seed within a week — "here, pretty much everyone said yes" — led by Nexus Venture Partners, with Paul Graham and Shopify COO Kaz Nejatian among the angels. The company says its agents now handle about 10,000 calls a day across travel, household services, insurance and real estate; one travel client routes about 50% of repetitive booking calls without human help and reported customer satisfaction exceeds 90%. The Launch HN post on 2025-07-10 drew 73 points and 42 comments, much of the debate on whether "self-improving" voice AI can be measured.

No ending yet — it is still running.

Background

Leaping AI is a voice AI startup betting that call centers will only hand real traffic to agents that get measurably better with every conversation. Instead of scripted decision trees, its agents handle inbound calls over phone and web, log each interaction, and use what worked to improve the next call — a pitch aimed at the customer-service teams who have been burned by brittle voice bots.

The company was founded in Germany in 2023 by Kevin Wu and Arkadiy Telegin after Wu's memory of a miserable day answering calls as an Amazon intern convinced him the work was worth automating. German investors rejected them twice, which Wu attributed to a market that doesn't fund technical founders with unproven models, so the pair applied to Y Combinator, entered the Winter 2025 batch and moved to San Francisco.

The move changed the numbers: Wu says revenue doubled in the first two months in the US after a year in Germany. After the batch he ran five days of back-to-back investor meetings and closed a $4.7M seed in a week, led by Nexus Venture Partners, with Paul Graham and Shopify COO Kaz Nejatian participating. The company reports more than $1M in ARR and roughly 10,000 calls handled per day across travel, household services, insurance and real estate clients.

The bet is that voice AI wins on trust, not on raw speech quality. For one large travel client, Leaping says about half of repetitive booking calls are handled without human help with reported customer satisfaction above 90% — enough proof, the founders argue, that an agent which demonstrably improves call by call can earn the volume that makes the business real.

What has to be true

  • Wu's own call-center day showed the work was miserable and largely scripted, so automating repetitive inbound calls attacked pain buyers already budget for.
  • Self-improvement is the trust wedge: enterprises won't hand real calls to static bots, but an agent that learns from every call can be measured, tested and improved.
  • Geography changed the outcome: revenue doubled in two months in San Francisco, and investors who passed in Germany said yes within a week after YC.
  • Voice agents don't need to handle everything at once: automating roughly half of one travel client's repetitive booking calls was enough to prove value.

What can be applied

Unproven models die in cautious markets: German VCs passed twice, yet the same pitch closed in a week in Silicon Valley once YC and early revenue gave investors a signal to say yes.

Aftermath

As of the 2025-09-17 Business Insider report, Leaping AI is live and expanding: the company reports more than $1M in annual recurring revenue, roughly 10,000 calls handled per day and customer satisfaction above 90% on its agents, and it planned to spend the seed round on product, go-to-market hiring and improving agent capabilities to meet demand. No later funding or revenue figures were disclosed in the sources below.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases