Irzan Raditya

Early life and education

Raditya has been an entrepreneur since elementary school, where he sold caricature drawings, bootleg CDs and mixtapes. As a schoolboy, he built modded games and a bot engine for online games for friends. He studied computer science in Berlin. [1]

Early career

Raditya held a stint as head of mobile products at a company. He says he had around four failed pre-startups before he built a company. He says he began his journey in 2015 by building an SMS concierge platform. He returned from Germany to Indonesia with his co-founders after they saw a trend in the US, and he says he came back because he believes Indonesia is the best place to build from zero to one. [1]

Founding years

Raditya says his company knew its model would not scale without automation technologies, particularly machine learning and natural language processing. He says Google, Amazon and Microsoft did not have an NLP algorithm for the Indonesian language. His company decided to build an AI model for the Indonesian language, and the team built an Indonesian language corpus from forums and social media with linguistic students. The company raised its seed round from investors. It had two offices, in Jakarta and Manila, with almost 100 people, most of them customer service agents. [1]

Pivot to Kata.ai

Raditya says the company was running out of money in 2016, with what he recalls as zero revenue. He says he had to let go of 92 staff because the company could not raise funds. He says Unilever approached him to build the first female assistant or female friend for its consumers, especially in Gen Z. The team reasoned that if Unilever needed the product, other companies in the FMCG sector would need it too. The company made a bold pivot to Kata.ai in October 2016. [1]

Kata.ai

Raditya describes Kata.ai as an omnipresent AI company that helps businesses improve customer engagement, cut costs and increase sales. He says it focuses on enterprise clients, especially regulated ones such as banks, insurers and telecoms. Its customers together reach 70 million end users, and the company facilitates 2.5 billion US dollars of transaction volume each year. One customer is the largest national pawn brokerage in Indonesia, and a major telecommunications company is both a customer and partner. With the same product and technology, Kata.ai helps its customers generate revenue instead of only saving costs. [1]

Raditya says Kata.ai mixes models, using smaller models for simple tasks and reasoning models for agentic functions. He says building a large language model to compete with frontier labs takes a toll, and the company does not have the luxury. His mantra to his team is never to marry your models. About 70% of Kata.ai's usage still comes from US frontier labs. The company packages its product in containers on Kubernetes for customers who run it on their own cluster, and it partners with a local data center that provides a token factory for data residency in Indonesia. [1]

About 95% of Kata.ai's deployments run on WhatsApp. Its memory combines persistent memory tied to CRM and CDP systems with lightweight session context. Raditya says the company can update its technology as it goes, as long as it does not break things. He says its principle is to put customer experience first and work backwards to the technology, and that it measures everything at the gross margin level. [1]

Growth and team

Raditya says Kata.ai's top line doubled and its bottom line tripled in financial year 2025. As of mid-2026, he says it is growing three times over last year on revenue. He says demand grew three to four times over two years on the same marketing budget, organically, and he credits Kata.ai's early AI leadership in Indonesia for its brand recognition with enterprise buyers. Kata.ai has around 60 people, who he says are much more productive than the 100 people it had before. He says the company has been doing forward deployed engineering for 10 years, estimates that about 80% of its deployments are equipped by forward deployed engineers, and says it once helped a customer reduce its headcount by 70%. [1]

Views

Raditya believes the future of AI business is outcome-based, not charging by tokens or usage. He says customers care about outcomes, not which models are used or how many. He says human-in-the-loop is still needed, even with the most advanced models, and estimates that in the next three to five years work will be about half AI and half human control. He says real results depend on two things: ambition and tokens. [1]

He says Indonesia is a land of opportunity, where problems are opportunities, and that cost still matters in developing countries like Indonesia. He says sovereign AI is existential for Indonesia, not a nice-to-have. In his view, AI built by Indonesians for Bahasa Indonesia performs better in Indonesian markets, and sovereignty means Indonesia captures the value, controls its data and shapes its technology on its own terms. [1]

He says enterprises increasingly want AI to reside in their own servers or cloud clusters, so their data stays put. He adds that some enterprises feel safer with AI in-house, for peace of mind about frontier lab terms. [1]

He says about 70% of white-collar workers in Indonesia bring their own AI tools to work. He says corporate bureaucracy and decision-making slow AI adoption in Indonesia. He also says Indonesia does not yet feel the urgency to use AI to survive. [1]

Raditya believes AI exposes which people, organizations and countries can adapt and which are not willing to, and that society will split into two camps, those who benefit from AI and those left behind. He says everybody will have their own AI, even running on devices, and that there should not be a monopoly from the big AI labs. He says his lighthouse is the success of Kata.ai's customers. [1]

Timeline

2015

Raditya begins his journey by building an SMS concierge platform.[1]

2016

The company runs low on money, and in October it makes a bold pivot to Kata.ai.[1]

2025

Kata.ai's top line doubles and its bottom line triples in the financial year.[1]

2026

Kata.ai reports growing revenue three times over the previous year as of mid-year.[1]

Appearances

Frequently asked questions

References

1. Never marry your models: Kata.ai's decade of outcome pricing . Beyond the Bubble , 2026-08-14.