Srikanth Velamakanni built an AI company before most people knew the A in AI.
In 2000, long before ChatGPT, he and five co-founders started Fractal Analytics with a simple bet: that math could predict human behaviour and help companies make better decisions.
The market wasn't ready. HDFC took 18 months to convince. Citibank made them prove themselves against 100 in-house PhDs. They won on accuracy, stayed profitable, and 26 years later Fractal became India's first AI company to go public.
In this episode of Unstarted, Srikanth sits down with Avnish to talk about the story behind the story. Growing up middle class in a small oil town in Assam. Why he never thought he was founder material, and the Narayana Murthy lecture that changed his mind.
If you are building, or still unstarted, this one is full of hard-won lessons on conviction, timing, co-founders, and staying relevant across 26 years of technological change.
Chapers
0:00 Intro
1:15 From a small town in Assam to Fractal
2:47 Math plus human behaviour: an AI company before AI
3:34 "I never thought I would be an entrepreneur"
5:01 The Narayana Murthy moment
5:54 Not money: building a 100 year company
6:57 Conviction vs market readiness
9:32 When the market is not ready (ICICI, HDFC)
10:27 The Citibank bake-off against 100 PhDs
12:00 Compounding your core passion
15:44 The two co-founder types you actually need
18:39 Finding your ICP without chasing noise
19:16 26 years of AI, from regression to transformers
23:08 Strategy is what you refuse to do
26:24 Elected CEO, then asked to leave in 6 months
28:02 Defining the company's values after crisis
29:19 Raising money and investor expectations
31:34 Will AI make Fractal obsolete?
33:58 Why Fractal was built in Mumbai
36:30 Advice: build what OpenAI cannot kill