It’s amazing to see my worlds collide here… So I thought I would transparently write the backstory and a personal reflection of my journey over the last ~5 years from slightly before when Agara was started to today’s acquisition. However, most readers in a hurry might want to skip the long winding twists and turns, and jump to the bottom section on some of my reflections and life learnings from this journey.

From The Ashes of Stayzilla

Many of the readers in India might remember a startup called Stayzilla (SZ) led by Yogendra Vasupal, better known as Yogi, the gentle, smiling, maverick co-founder and CEO of the startup from Chennai. SZ had a long 11-year old history of doing online hotel bookings. From 2016 onwards, it aimed to be sort of an Airbnb of India, with its focus on homestays and alternative stays. When I first came in touch with Yogi in early 2016, I remember liking him a lot. The startup seemed to be doing pretty well, having just raised a Series C by well known VCs — Matrix Partners and Nexus Venture Partners. I had recently moved with my family to Bangalore from San Francisco in the summer of 2016 with the primary intention of traveling in India and surrounding countries and returning in a few years, but joined as SZ’s CPTO — Chief Product and Technology Officer.

SZ was initially a bit of a rude shock. There was too much infighting in the team, the morale was terrible and the startup hardly had any process or structure. But I was a bit on a high those days — with the novelty of a recent move back to India and the freshness one has after a long work break of more than a year (my previous work stint at Twitter in San Francisco was over more than a year back). I quickly got to work, had to let go almost a quarter of the team, reorganized the rest and within a quarter, the team morale roared back and execution was on track. Credit to Yogi that he had full faith in me and allowed me to do all these changes in quick time.

After the team execution issues seemed to have been resolved and I raised my head above water, I realized there were deeper problems at SZ. The business wasn’t as good as I had thought it was. Then the news came that the number one company in the India online travel market (MakeMyTrip) had just acquired its close number 2 rival (Goibibo). That shut out almost all other travel startups out of the India travel business. SZ’s homestay business was OK — it threatened to take off but never did.

We needed to raise our next round of funding and Yogi and I started doing the investor rounds. As the late stage VC and PE rejections piled on from Bangalore, Mumbai, NYC and Sand Hill Road in the winter of 2017, we tried to sell the company but nothing happened there either, and the writing was on the wall. SZ would need to shut down.

In a hurried blog post, Yogi announced Stayzilla is shutting down its operations in late February 2017. But what happened thereafter was worse and completely unexpected— I or anybody working in the company did not know this but it turned out that the startup had allegedly some unpaid pending bills from years ago from a prior vendor who went after Yogi, and it got messy.

Startups fail all the time, and in hindsight I can say that Stayzilla deserved to fail for its deep structural faults, but the vendor incident made the failure unfortunately ugly for Yogi and the other co-founders. (We have exchanged messages since but I haven’t met Yogi since February 2017. I called him this Sunday. He recognized my voice as soon I said Hello. It felt good talking to him. I am looking forward to catching up with him in person soon.)

But the tech team at Stayzilla was really good and loyal— as solid as any I had worked with in my career. One of these PMs in my team was Abhimanyu Singh, VP product for the host side of the stays business.

OMG. What should I do now?🤦🏾

In January 2017, it had been barely 7 months since we had moved to India and now the startup I had so enthusiastically poured myself into was dying. It had all happened too quickly. What was I to do now? My wife and I thought about whether we should just return to the US right then — some recent Indian execs at Flipkart who had come from US had not had it good here and gone back to US in 1–2 years. But my wife wasn’t too happy with doing another big move that soon. She was still “moving in” and our young kids were just settling in their school. We had not made the move to India because of Stayzilla, so we decided to stay.

I did some soul searching to think about what I wanted to work on now. Somehow, despite the SZ setback, I was still in a mood to take risks. And whenever I have been in that mood in my career, I have done a startup. It never once occurred to me to look for another job. The only problem was that I didn’t have any well formed startup idea — instead I had 2-3 half-baked ones. I had developed a rule of thumb from my past startup experience — that it takes about a year from when an entrepreneur decides “I want to do a startup” to lining up all the ducks to start execution on one singularly focused path. But I didn’t have a year. I didn’t even have two weeks. Time was running out as SZ was breaking apart.

Finally the decision was made. I didn’t know what exactly the startup would be doing, but I knew I will need builders — specifically engineers and designers — and I didn’t want to lose the best ones I had run into at SZ. In the very last days at SZ, I made a simple pitch to the top 7 engineers and 2 others in SZ. I was transparent with all of them — I didn’t yet know what we would do but they would get the same uninterrupted salary as they were getting earlier. To their credit and a little bit of my surprise, all but one of them said yes! And so I had a startup team of 5 backend and 2 frontend engineers, 1 designer, and myself, but without a clear idea of what we would do in the startup. Still, it felt like a much needed small victory at the time as I had experienced the pains of even hiring one engineer for my previous startups in Silicon Valley — here I had 7. Three days after SZ‘s final blog post, I announced what I was going to do next.

However, I remain a builder first and have luckily been able to convince a small team of awesome engineers to come with me to hack on some really raw ideas.

And so I took up a few desks in a small co-working space in Koramangala in Bangalore and incorporated a startup called Halli Labs Pvt Ltd (“Halli” means “village” in Kannada). One of the meta ideas I had always nursed was to work on multiple parallel projects like a startup factory or a research lab, so our first product would be branded by a different name — Agara Labs, named after the Agara Lake that runs between HSR layout and Koramangala, the adjoining neighborhoods in South Bangalore that were the homes of my past and new companies respectively.

Salaries at SZ were low and matching them turned out to be not too difficult for me— I called up some friends who were doing startups in the US. They were all desperately looking to hire engineers for their startups, so I took on a couple of consulting projects from these valley startups, only to keep half of the team busy doing something while we zone in on one startup idea. The projects paid more than enough for everyone’s salaries. They were still inexpensive from a Silicon Valley founder’s POV and these engineers were really good so they didn’t need much management. With most of the team I had assembled now keeping busy, I turned back my attention to looking for exactly what the startup would do :)

Assembling the founder team and raising seed from Kleiner Perkins

As mentioned above, Abhimanyu Singh headed the seller-side product at SZ, and has the background of starting several startups. He was not the type to look for another job after SZ. He was also thinking about various startup ideas. He would sometimes come over to our co-working space and we would chat and explore ideas. He was not yet part of the Halli Labs team.

Meanwhile, an idea started to take hold. The co-conspirator of this idea was Michael Abbott who was now a General Partner at the legendary VC firm, Kleiner Perkins (KP). I had first met Mike when he had come on as VP engineering at Twitter in 2010 — he was my direct manager for a while and after that stint, had become a General Partner at KP. After I left Twitter, he had invited me to be an Entrepreneur-in-residence at KP, and so I hung around in 2015–16 for a few months at KP’s Sand Hill and South Park offices. I didn’t like listening to startups’ pitches, so never got much into the EIR thing at all. But the occasional chats with the likes of veteran VCs like John Doerr, Mary Meeker, Randy Komisar and Bing Gordon were cool, and the Monday morning partner meetings where I was a fly on the wall were especially interesting. Mostly I was there because KP’s South Park office in San Francisco was hardly utilized and it was convenient for me to work out of there for oddball things I was doing (working on some open source projects, investing via Specialized Types, a co-investing technical group I co-founded) than working out of home or various SF coffee shops.

Mike and I were in touch after I moved to India. I told him that I have a built out a team of engineers and am thinking about various startup ideas. KP used to convene enterprise CIOs often and one of these — Proctor & Gamble (P&G) — were looking for an ML solution to improve their customer experience across their global products. An ML startup was right up my alley but I was initially hesitant to do it… An ML startup needs data and most ML startups don’t have customers to have that data. But here was a large customer with years of data so we settled on this idea. Abhimanyu liked the space too and got on board as co-founder. Mike did the seed from KP. We exited the consulting projects to recover the engineers and went about building a deep learning based CX assistive solution together with a large customer with real data and a real problem to solve.

Meanwhile, a friend reconnected me with Arjun Maheswaran, a soft-spoken, hardcore ML research scientist I had overlapped with at Twitter, as he was moving back from US to India too. Arjun was a natural fit for Agara, completed the set of co-founders and came on board in summer of 2017. We were on our way.

Google Tez

But there was another parallel story developing on the side. After SZ shut down, I got a call from my close personal friend of nearly 20 years, Caesar Sengupta, who had started his career in Google Mountain View and was now leading Google’s Next Billion Users org from Singapore. Caesar was in Kormangala in Bangalore in April 2017 to give a preview demo to the Indian tech celebrity Nandan Nilekani of a new payments app his NBU team was building, called Tez. Tez looked like a beautiful app in the making. I met him in his hotel and we chatted about NBU etc.

He and I met again in Cafe Coffee Day by my co-working space on an afternoon, and he asked me if I want to come and lead the Tez eng teams across India and Singapore at Google. I didn’t take it seriously — twice before in last 10 years I had had an offer from Google Mountain View and not taken it (I always felt it was too big a company for me, and the last time I chose Twitter over Google which worked out great)— and my mind was sort of set on doing a startup, which was doing well.

But Caesar was persistent. As I had not yet zeroed in on exactly the idea that the startup would do, somehow the Google thread remained open, in his and my minds. After a few months of occasional chats, that thread started becoming more and more attractive. Caesar is one of those flamboyant, disarming and flexible leaders who know how to close, who know how to not get hung up about the letter of the process in even as big a company as Google. And so he continued to sweeten the deal until it was too attractive to ignore — he would not just hire me, but if hiring me would take that, his NBU would acquihire the startup and its employees. I dilly-dallied. I still wanted to do the startup and we had even raised the seed from KP by now… but the deal from him continued to get sweeter until it was just not ignore-able. The idea of acquihire by Google actually seemed good on paper anyway. Besides, we had been long-time friends and liked each other. I had zero idea about Payments or Fintech at the time, but if there was one FAANG company I would ever join in my life, Google was the least unlikely in my mind because of its solid engineering-driven culture.

So at long last I made up my mind that we will get acquihired by Google.

Caesar Sengupta on Twitter: "Welcome @Pankaj and the team at @halli_labs to Google. Looking forward to building some cool stuff together. https://t.co/wiBP1aQxE9 / Twitter"

Welcome @Pankaj and the team at @halli_labs to Google. Looking forward to building some cool stuff together. https://t.co/wiBP1aQxE9

But it turned out that not everyone in the startup wanted to go to Google. Abhimanyu and Arjun explicitly did not want to join the big behemoth. The startup was actually doing really well by now. The ML work was really interesting, cutting-edge and our product was looking good to P&G. Some of the other engineers weren’t sure about Google either. Google was fine with whoever wanted to come or not come with me. The acquihire was really just a glorified hire with bonuses, with nothing to do with what the startup was doing. So we split — those who wanted to continue with the startup would carry it forward as Agara— while some of the team at Halli Labs would join Google’s NBU working on Tez. And so I left the CEO duties for Abhimanyu and joined Google.

In the next 3.5 years, I remained on the board at Agara. Abhimanyu has since been the CEO and Arjun the CTO. They have taken the startup from strength to strength, building out one of the strongest ML teams I have seen anywhere globally, and delivering an exciting product roadmap from an email assist product, to chat, and then a magical autonomous voice based product. They have raised multiple rounds of funding after KP’s seed (first from Blume, then from RTP and then from UTEC as recent as in early 2021) and have multiple paying large enterprise customers for their ML based CX products. The startup has been blossoming.

The Google Tez Rocketship

𝚙𝚗𝚔𝚓.𝚎𝚝𝚑 on Twitter: "Just came across this (slightly dated) graph of UPI growth while making some slides. @GooglePayIndia was launched on Aug 17, 2017 and dramatically changed the slope of UPI adoption.Source of the underlying graph: https://t.co/TFardwQmaq pic.twitter.com/gr8fofxxCS / Twitter"

Just came across this (slightly dated) graph of UPI growth while making some slides. @GooglePayIndia was launched on Aug 17, 2017 and dramatically changed the slope of UPI adoption.Source of the underlying graph: https://t.co/TFardwQmaq pic.twitter.com/gr8fofxxCS

I had no inkling about what Google Tez (“Tez” literally means Fast in Hindi-Urdu) would become when I joined. I had no knowledge about Payments and Fintech. It even took me a few months to unlearn my scruffy startup ways and adjust to a large company like Google, in particular leading a large org of 100+ engineers that was rapidly growing organically as Tez was growing.

I had zero idea that Tez would be the rocketship it would become, propelling UPI’s jaw dropping growth, changing the game completely for digital payments in India, and making Google Pay a household name in the remotest parts of India. I am truly thankful to the Gods of Chance for me managing to be somehow in the very center of that revolution.

A Crypto Calling

I fell completely in love with Crypto sometime in 2018–2019. I had a Coinbase account since 2014 and even did a personal side project analyzing the Bitcoin blockchain graph when I was a lot into graph algorithms and social network analysis at Twitter. I had not even scratched the surface of crypto though.

I was loving working in Payments and Fintech and was overtaken by the desire to read and learn all I could about this field that was new and fascinating to me. Over a stretch of 2 years — I read and learned about the topics of money, banking, global payment systems, etc. Running into crypto was inevitable. Soon, studying, learning and talking about crypto became a personal hobby. I came to be known as the crypto person in the whole Payments org at Google. I started doing some serious research into some new cryptographic primitives and started a couple of CBDC related projects at Google. At some point I decided if I ever was going to leave Google Pay, it could only be to work in Crypto.

In March this year, a few unexpected events took place in Google and outside that eventually led me to join Coinbase as VP Engineering and Coinbase India Site Lead. Starting stuff from scratch always excites me, and I was by now a big believer in crypto. So I transitioned from leading a large org across India, Singapore, China and US at Google to being the first employee of Coinbase India. The day I announced internally was April 15 — coincidentally the day that Coinbase went public.

Customer Focus

Soon after I joined Coinbase, I learnt that Coinbase wanted to invest a lot more in customer experience. We were hiring fast, but not fast enough to work on building out all we wanted to in this area. I was thrilled then when Coinbase expressed interest in acquiring my ex-startup Agara. And, given the overlap of interest, I was fully transparent with Coinbase and did not participate in any further evaluation of the startup.

Old friends in new settings

It turned out that there was indeed a great fit for both sides, leading up to today’s acquisition announcement. I am thrilled that this has happened — it is a clear win-win for both companies. Agara has some of the smartest folks I have worked with in my career and they will bring their deep expertise in ML and in the customer service domain to Coinbase.

𝚙𝚗𝚔𝚓.𝚎𝚝𝚑 on Twitter: "Welcome @quartermiler, @segmenta and @AgaraLabs to @Coinbase. Looking forward to building some cool stuff together! https://t.co/ahShzWq2X6 / Twitter"

Welcome @quartermiler, @segmenta and @AgaraLabs to @Coinbase. Looking forward to building some cool stuff together! https://t.co/ahShzWq2X6

Personally I am happiest about reuniting and working together with some of my old buddies. I don’t know what it is but there is something special about working again with the people you have worked closely with in the past. It is almost as if long lost friends are coming together again. All my ex-colleagues have a special place in my heart ❤️ and luckily for me, they more often than not want to work with me again too.🙏🏾

Some Reflections

𝚙𝚗𝚔𝚓.𝚎𝚝𝚑 on Twitter: ""Wanderer, your footsteps arethe road, and nothing more;Wanderer, there is no road, the road is made by walkingBy walking one makes the road, and upon glancing behindone sees the path that will never be trod again.Wanderer, there is no road - Only wakes upon the sea" / Twitter"

"Wanderer, your footsteps arethe road, and nothing more;Wanderer, there is no road, the road is made by walkingBy walking one makes the road, and upon glancing behindone sees the path that will never be trod again.Wanderer, there is no road - Only wakes upon the sea"

As I wrote the draft of this blog post last weekend, I reflected on this personal journey with its twists and turns, lows and highs. Why is there so much stuff happening here? Why is there so much overlap and intersection of old friends and ex-colleagues coming together at various points? Multiple parallel arcs seem to stretch across time driven by their own destinies but then also mysteriously intersect and converge. I have two meta learnings:

1. A lot of these events are linked to each other in a web of strong human relationships. So much of this story would never happen unless the people involved didn’t have a prior history. The learning? Whatever you do, give your best and treat your peers with the highest respect. There can be only one lasting reason to work hard and give your best — no, not to get promoted or get a raise or look good — but to hopefully earn the respect of those peers forever. You will surely meet some of them at some point later in your life — you just don’t know who and when. But if you have been sincere and given your best and treated them well, they will remember you fondly, they will want to work with you, years of context will not need to be re-developed and they might change the course of your career.
2. Don’t hesitate to take risk when the time has come to you to take it. In order for interesting compounds to emerge, one has to mix it up, move around a bit and not be afraid to fail. In fact I will contend that many of us are so comfortable that we don’t even have a sense of what it means to fail. Stayzilla’s failure was in fact not a failure at all. Most of the folks I know from there are now doing wonderfully well at places like Google and Agara and other companies and their own startups.

So I wish and hope for all of you to not get too comfortable. We delude ourselves in thinking of a million reasons on not changing anything in our lives. But I hope you will periodically seek out new environments and new areas to explore. Whether you believe in Karma or the circle of life or whatever, good stuff won’t happen unless you let stuff happen to you.

PS: Come work with me at Coinbase. We work hard and play hard!

𝚙𝚗𝚔𝚓.𝚎𝚝𝚑 on Twitter: "PSA: We are hiring across all roles & functions at @Coinbase India1. Browse roles & apply here: https://t.co/1eNaxia3242. Blog post on our ambitious 🇮🇳 plans: https://t.co/bvQbJipnOJ3. Fireside chat video discussing (2): https://t.co/2fGLIWfei74. We like to have fun!🔥⚡️ https://t.co/Tj0ZS7Ap1K / Twitter"

PSA: We are hiring across all roles & functions at @Coinbase India1. Browse roles & apply here: https://t.co/1eNaxia3242. Blog post on our ambitious 🇮🇳 plans: https://t.co/bvQbJipnOJ3. Fireside chat video discussing (2): https://t.co/2fGLIWfei74. We like to have fun!🔥⚡️ https://t.co/Tj0ZS7Ap1K

Coinbase is innovating at the frontier of crypto and blockchains — some of the most exciting trends in the world today. In my conversations with the Coinbase executive leadership, it became clear that this is an ambitious and mission focused company for builders like me :)

India has world-class tech expertise and some of the brightest minds in the industry. Coinbase plans to tap and develop this local talent. Specifically, the plan is to hire hundreds of employees across all levels in engineering, product management, UX design, research and program management within the next 1–2 years to build out a full tech hub in India. Together, we’ll work on some of the most interesting challenges spanning the full modern tech stack — including deep tech areas like blockchains, data engineering, infrastructure, machine learning and more.

To support our ambitious growth plans in India, we will also explore startup acquisitions and acquihires. Founders who might be interested in joining Coinbase’s journey and mission, please contact me.

Please review the open positions here at the India Careers site. My DMs are open for any questions or comments.

Looking forward to building together the future of the global cryptoeconomy from India 🇮🇳!

[My farewell note in Google, sent April 15, 2021. Slightly modified to redact sensitive information.]

And it seems like only yesterday

Gee, ain’t it funny?

How time just slips away¹

Comrades!

Life is short, they say, though Seneca might disagree. Time however always passes. At the exact same rate for each one of us — one second per second. But whether it spans split seconds or years, sometimes it gets etched in our minds forever. I still remember when I came into the welcoming lobby of Google Bangalore’s RMZ office on July 17, 2017. Well, if life is made of phases — segments of time — this lovely segment of my life labeled Google will end on April 28th which will be my last day.

Isn’t it amazing that — in all of space and in all of time — a path of my life’s journey crossed with yours? And that too while we together built a product that has surely put a little dent in some small corner of the universe.

I have a great sense of pride and gratitude to have been a part of the Tez rocketship aka Google Pay alongside you from zero to nnnM+ users today now sending annually more than $mmm Billion to each other! I hope and wish for you that a corner of your heart feels that way too. I have full faith that you all are going to make this dent bigger, much bigger. Thank you from the bottom of my heart for your comradeship.

I have learnt so much from you. I came as a data-hungry ML practitioner and tried to inflict you all with a custom co-designed ML course². I leave as a data-shy Privacy fanatic. I came as a total noob in FinTech. I leave as someone who got the opportunity to go deep into real time payment systems around the world and the very definition of money. I now converse with central banks around the world and co-architected the payment rails of our NUE application. I came in not having thought too much in the past about the critical issues of Diversity, Equity and Inclusion, but got the opportunity to lead our first Google India Tech DEI Council and help shape our DEI mission and agenda. I came in as an Apple fan-boy but leave wowed by the unimaginable reach of Android. We even worked closely with Android Security Platform recently to include new Keystore/Keymaster APIs which will bring new cryptographic primitives to upcoming Android 12 later this year to help anyone build offline, networkless payments with minimal changes to the Trusted Execution Environment.

Of all the memories flashing in my mind right now though, those that will endure will be about all of you. Perhaps that is what it is all about when all is said and done. To work with people that you come to respect and to give your best so that hopefully they come to respect you too.

<Pics with colleagues in parties redacted>

What next? Many of you know that I have been enamored about all things crypto and blockchain for the last few years³. So I am going to follow my passion to work at a crypto company in the news these days and get immersed in this space⁴. I will continue to be based in Bangalore 🇮🇳. There is never a good time to leave, but we have an amazing senior team right now and I have the satisfaction of knowing that you all are in excellent hands.

A special thanks to my friend of 20+ years, Caesar, who took a bet on me ~4 years ago starting from a meeting in a cafe in Koramangala, Bangalore. He showed me a not-fully-baked-yet but stunningly beautiful app called “Tez dogfood”. It completely went over my head at the time but I somehow managed to keep my mouth shut than remove all doubt of my incompetence in the area, which resulted in this tweet a few months later. Well, as his tweet predicted, we did build some cool stuff together!

Caesar Sengupta on Twitter: "Welcome @Pankaj and the team at @halli_labs to Google. Looking forward to building some cool stuff together. https://t.co/wiBP1aQxE9 / Twitter"

Welcome @Pankaj and the team at @halli_labs to Google. Looking forward to building some cool stuff together. https://t.co/wiBP1aQxE9

I’ll miss everyone so much. I will be proudly watching you take this nation-impacting and hopefully world-changing product Google Pay to new heights.

Please do not hesitate to ping me if I can be of any help whatsoever at any point in your life — no matter how many years or decades would have passed by since we next see or talk to each other, I would love to hear from you! It is the ultimate validation of my life’s work if I were to be remembered by you and to be of any value to my friends and colleagues I ever crossed paths with⁵. Of course please feel free to ping me too while I am here in the next few weeks and we will have a 1:1 chat about anything!

Thank you for all your love,

Chalo Phir…⁶

Pankaj

“Gupta Ji Fruit Wale” with GPay QR!

[1] Willie Nelson — 1997 — Funny How Time Slips Away

[2] <Internal link to “Machine Learning Literacy Course”>. Thank God while thinking later about doing an experimentation course, I got wiser and instead of doing any work myself, persuaded some of our engineers more capable than me to design and teach it :)

[3] A multi-part tutorial on bitcoin and blockchain that I did a couple of years ago during our tech talk series. Also see <internal link to all things CBDC>

[4] “We are like the spider. We weave our life and then move along in it. We are like the dreamer who dreams and then lives in the dream. This is true for the entire universe.” — Loosely translated from the Mundaka Upanishad, also popularised by David Lynch in the movie Inland Empire and in the TV series Twin Peaks.

[5] जब दिलों को जीतने की कला आ गई

तब समझना कामयाब हो तुम।

(When you have learnt the art of winning hearts / Consider yourself successful)

[6] “So long” or “Till we meet again”

Active

• My Twitter — where I post the most
• This medium blog — Active for all personal blog posts
• A Zero Knowledge blog — For technical and semi-technical posts

Old and Not active:

• Idle Thoughts — Old personal blog
• Tumblr — Misc quotes and tidbits I came across. Now superseded by my Twitter
• Mildly Interesting — Old technical / work related blog
• Old Running blog

This blog post tells it all. https://medium.com/@halli_labs/halli-labs-is-joining-google-feeada989cd9

Stayzilla, as I wrote at the time of joining, was the smallest, least well known and the riskiest of my four choices at the time. I took it because — among other reasons, I have a bias for alpha bets (which I guess I should correct now 😀). That bias had paid off handsomely last time I took it in Twitter. Of course, in hindsight, this was clearly not my best decision. But it does not take away from the fact that I thoroughly enjoyed my time at the company. For that, I must thank the founders (Yogi, Sachit and Rupal), my direct reports — Santanu (VP engineering), Mouli (VP product, guest, and design) and Abhimanyu (VP product, host) — and most of all, my awesome team.

This team is good

From my past experience at Twitter, I had developed some models of team building — vertical, autonomous small product teams who are as self-sufficient as possible for delivering their own metrics, are staffed with their own dedicated functional resources, and are supported by strong infrastructure and analytics. I put these into practice at Stayzilla. Under competent and motivated team leaders, I would like to think that this led to a pretty incredible culture, productivity, team dynamics and overall execution efficiency.

I am proud of this team — they will do enormously well wherever they go. For me too, it was an enormous learning to build and organize a mid-size product, engineering and design team in India and I have innumerable lessons to take forward.

Next Play

I am becoming an advisor to the two firms who were on Stayzilla’s board — Matrix Partners India and Nexus Venture Partners India and look forward to being of help. I am also an individual investor in some remarkable early stage Indian founders and startups — Shuttl, Myra, Belong and Mihup — and will look to be of more help. However, I remain a builder first and have luckily been able to convince a small team of awesome engineers to come with me to hack on some really raw ideas.

On Startups and the Indian startup ecosystem

Startups are about taking chances, and some startups will fail. This is almost a mathematical statement and independent of a particular startup’s market, timing, or execution. But sometimes we tend to forget this. Recently, there have been instances of layoffs at a few large startups in India. And everyone seems suddenly worried. I am no expert, but startup ecosystems go through such phases. What the Indian startup ecosystem needs — as always — are even more dreamers and builders.

I love San Francisco

When someone from outside the city asks about living here, I like to quote Sander Daniels, co-founder of Thumbtack who has summed it up nicely: “…living [in San Francisco]… is like living in Florence at the time of Renaissance”!

I have participated as an employee, a founder and recently as a startup investor and advisor. I have founded two startups in the Valley. I am fortunate enough to have been an early, active participant in the formation and stunning global growth of a household startup name, Twitter. There’s no mistaking that San Francisco is the center of the tech startup Universe. My journey from Stanford to Palo Alto to San Francisco has been right through the center of it. Aided by extremely good luck and an amazing set of friends and mentors, I thoroughly enjoy the fruits of living here. I am in love with the San Francisco life style — its Golden Gate Bridge, its California Academy of Sciences, its IMAX theatre, its numerous parks, its weekend jaunts to the Tahoe slopes in winters and to Napa in other months, its Philz Ether, its startups in seemingly every nook and corner of SOMA, its motorized scooter riding brogrammers on 2nd St. … never ever seem to grow old.

Why move then?

Some friends have asked me why are you making this “big” move? The unwritten expectation of an answer is that there is some burning issue — perhaps an unwell parent or in-law, or may be an inability to adjust to the American life, or perhaps a burst of patriotism fueled by years away from the country of birth, or may be an early onset of midlife crisis? I assure you all that in my case there is none of those. It just so happened that an opportunity together with a probing question presented itself, one thing led to another, and my wife and I decided that if we have to explore, this is the right time for us to do it. The question then quickly transformed itself from “why” to “why not?” And that was it. So, yeah, why not?

India and its startups

Much like in Silicon Valley, the startup mania of the past few years in India has somewhat cooled with realization of unsustainable burn rates, dropping valuations, and rediscovery of priorities related to making money or at least losing less of it. Still, startup ecosystems once built never just go away. And the one in India, especially in Bangalore, is alive and kicking. In the last few months, I got a chance to talk to many players of the ecosystem there, including angels, VCs, private equity investors, and most importantly founders of startups of all animal shapes and sizes. I found an ecosystem brimming with energy, potential and ambition.

Several intriguing opportunities presented themselves organically as a result of these discussions. Of them, one stood out for me — a rapidly growing mid-stage startup called Stayzilla. It offers peer to peer accommodations — aka homestays and alternative stays — much like Airbnb, which also has a small presence in India. I can obviously not divulge much about the rising time-series graphs at Stayzilla, but just as one data point, it is adding one Airbnb in India every month. It is led by Yogi, co-founder and CEO, who possesses a rare combination of clarity of vision, humility, integrity and burning ambition. Yogi and his team find themselves in the midst of a potentially explosive market and space in the years to come, and not just in India.

In the role of Chief Product Officer and President of Technology, I will be helping Yogi build world class products and technologies. I eventually chose Stayzilla over a few other, much bigger startups in India simply because the Stayzilla opportunity falls in that sweetest of spots at the intersection of a rapidly growing market, a fine team, and a massive growth ahead of it — the perfect fuel for a rocket ship in the making, at least in my humble opinion. When faced with a choice of various sizes of companies, I have always sided with smaller ones before. And the last time I did it, that high alpha bet worked out quite nicely for me :)

Specialized Types

Last year, I co-founded a first of its kind technical advisory group for startups, called Specialized Types, with the mission to connect the world of specialized, deeply technical folks with that of startup founders. I am happy to share that this group, with its novel model of working together with world-class VCs, has now established its own product-market fit. My co-founders, Ram and Babak, are now scaling it multiple notches up. I will remain a member of and an advisor to the group here and will explore how a version of it could be brought to the startup ecosystem in India.

Why to… Why not?

So now I am doing this thing in the 42nd year of my life. 42 may or may not be the answer to the ultimate question of life, the universe, and everything. But hopelessly habitual pattern finders that we humans are, I can’t help myself note that coincidentally I was just short of 21 when I first landed in the US from India as an incoming Stanford CS PhD student. That makes it almost half and half in nearly diametrically opposite points of the globe. So does this move set the stage for the rest of my life? *Rolls eyes*.

(Image credits: Drawings made by my son’s fellow first grade students for him as they learned about his move.)

Different things work for different people, but my only resolution of 2016 is to master the daily practice of the above.

The biggest thing in the above is to focus. Peter Drucker said it best:

Nothing is less productive than to make more efficient what should not be done at all.

Related reading:

1. Essentialism by Greg McKeown is a wonderful book that makes you think about what is important and what is noise.
2. This video by Jony Ive where he talks about what he learned from Steve Jobs about focus.

It finally happened. I checked moments when I meant to check notifications. I'm currently recovering & will be ok. Thanks for your concern.

Last Friday the Twitterverse detected that the location of the Moments icon on Twitter’s Android and Web apps had moved leftwards closer to the Home icon, swapping places with the Notifications icon. I find nothing wrong with the company’s attempts to get more users to click on Moments — if it leads to sustained higher engagement, the temporary dissonance to muscle memory will be quickly forgotten. It might also help add incremental revenue via Promoted Moments vs the un-monetized Notifications tab.

More concerning to me wrt Moments is that certain lessons from history have seemingly not been learnt.

History

There was this quite reasonable insight inside the company, first articulated by Jack in early 2011 IIRC, that there is great content flowing on Twitter all the time — and the product should bring that content to everybody (with the unsaid but hoped-for implication on new user growth and retention). The first full horizontal implementation of that vision was a feature called #Discover, released in December 2011, but there were many more “vertical”

attempts during sports events such as World cups and Olympics, as well as the manually curated Twitter Stories. All of these efforts have now been abandoned, replaced by Moments that was launched two months ago.

Obviously Moments represents the state of the art in company’s execution of that insight. When it was first launched, I was confused by how it could have not learnt from the lessons that were painfully learned during the 3+ #Discover tab years. May be I should just wait a few more weeks, I thought… It has been exactly two months now. I fear that what the icon swap could do, somewhat counter-intuitively, is to lead to lower overall engagement over the medium term as more users get exposure to it quicker and get trained quicker on ignoring the tab altogether.

The lessons of the “#Discover tab”

The Discover tab was officially abandoned in April 2015, but was on crutches since 2013. It was built and owned by a highly competent peer engineering team of mine in NYC, but I was closely involved in key aspects as my (ex-) team provided the personalization — both interest and network based — to the product. In other words, it was using a couple of services that were used to ensure that the content shown in the tab to a user is relevant to the user’s interests and/or originated from user’s extended network. There were many painful lessons learnt by the company in the multi-year process of trying to make Discover more useful to users. Let’s recap some of those and see which ones seem to have been applied already by Moments. (Thanks to Ruslan Belkin, my and Discover team’s boss for much of the Discover years, for his useful input on these — here’s hoping he will write a blog post himself one of these days.)

The key lesson that seems to have been learnt well by Moments is that of Organization and Categorization: Discover was meant to look and feel like the home timeline and was therefore a collection of tweets ordered reverse chronologically. This was later found to be jarring to users and the reason was traced to the fact that unlike the home timeline, these tweet authors are strangers, plus the tweets intermix topics unfamiliar to the user and this combination leads to total lack of context. Moments has as its starting point a clear organization backed by excellent design (the best part of Moments IMHO) and categorization backed by a taxonomy, currently shallow which I hope will get much deeper over time.

The two key lessons that have been unfortunately ignored by Moments are:

1. Users not only do not care about stuff that is not relevant to them but are actively pissed off by clearly irrelevant content. Interest-based relevance is important, and irrelevance is criminal. This is true btw in general about all recommendation and discovery products —one’s annoyance threshold becomes much lower for stuff that they did not explicitly ask for.

Tren Griffin on Twitter

I suddenly had an urge to create a Venn diagram. It was an uncomfortable feeling, so I came up with this. pic.twitter.com/aVdvTOHlsq

Marc Andreessen on Twitter

Twitter Moments could tell me what's hot among the topics in which I am interested. Instead it tells me what's hot on the Yahoo home page.

2. Content without social proof is at a big disadvantage because most other content in a social network is flowing through social pipes. On the flip side, when a user’s local Twitter social structure is overlayed with this content, it is magical. This was a clear observation from significant outperformance in a/b tests on different kinds of tweets and different kinds of presentations. Once low performing tweets were eliminated, almost all of tweets in Discover were good quality tweets in that user’s extended second degreee network (tweets from not just that user’s followings, but also that user’s followings of followings) filtered by the user’s interests.

Feature Friday: #Discover - AVC

What Twitter has that no other product has is its unique graph — asymmetric, public, semantically rich. The graph’s edges neatly and succinctly encapsulate everything about what a twitter user is and cares about — both by way of social and interest connections. This graph has time and again proved inside Twitter to be effective, “unreasonably” so as I liked to say. Furthermore, when a user invests even a little bit in building their graph, the chances they will come back get much brighter.

“But Moments is meant for new and casual users…”

This is an often cited argument, which IMHO is somewhat lazy. Moments is not a separate app — a new user has to at the minimum go through the Twitter signup flow, leaving enough crumbs along the way of what they care about. For example, even a few followings are enough as a seed — standard extension techniques like following edges from that seed and looking up similar nodes in the graph work quite well in practice. New and casual users also browse other people’s profiles, leaving more hints on what and whom they care about. In the very worst case when there is absolutely nothing known about a user, the right strategy is to accept low recall but to still not compromise on precision (i.e., show just the really globally popular content and it is okay to show nothing else).

You have to wonder why these lessons may not have been learned or applied. May be they were deliberately ignored for now just to do a quicker launch? Or perhaps, once a higher level decision was made to go with curated rather than algorithmically generated content, there is just not enough content to be personally relevant to everyone? Or perhaps the real driver is incremental ad revenue which might be easier from top news style content vs niche, personalized content? That is in fact the only conclusion I can come to — Twitter really wants Moments to be a news app. Unfortunately, this sells Twitter so short — casual users can find daily news in too many other places. Yes there is a lot of great content unique to Twitter, but that comes to fore for casual users only during extraordinary events — such as the Paris terrorist attacks — which are far too rare to have any chance of building retention of a casual user. The real draw of Twitter is great niche content and commentary from people and sources relevant to you, the tail of the content distribution.

Without getting into this tail, without a social layer, Moments is just a meh me-too-news-app. For Moments to have a chance, it has to be one with the rest of Twitter in a mutually reinforcing relationship, for both casual and active users.

Back to Attrition

But this post is not about a lesson or two missed. Let’s get back to why those lessons from #Discover and other recommendation related product features might not have been learned or applied to Moments.

I hypothesize that there is a deeper, sadder reason. Almost all of the senior people in engineering and design and management involved with Discover have moved on. Those lessons are likely quite simply lost from institutional memory. Every new person at the helm does not only have to start from scratch, but also live with the insecurity of being on a short leash.

The meta-lesson here IMHO is for every startup that is hot today — try your absolute best to retain top talent for the long haul (5–10 years) and don’t let go of your top management every year or two.

First, what does it mean to be real-time? Let’s look at some popular examples. In terms of products: Twitter is real-time and CNN is not, Uber is and Yellow cab is not, Hipmunk is and that neighborhood travel agent is not. In terms of technologies: Storm is real-time and Hadoop is not, the serving layer of a website is but the offline analytics layer is not, etc.

But this is too broad a brush. I claim that Twitter, Uber and Hipmunk and the on-demand instant gratification startups in vogue today are not really real-time! Wait, what? Yes, but I get ahead of myself. I need to first tell my own story — the story of a feature I came up with at Twitter called MagicRecs.

A story

In my early years at Twitter, I had worked on a recommendation product feature called Who-to-follow (then and still lovingly called WTF). It was based on an offline pipeline that recomputed recommendations for a subset of users every day, and thus for all users every few days. About a year after its launch, we started noticing that it has been doing really well for two types of users: (1) new users, and (2) those users whose recommendations had been recomputed the previous day. The increase in our metrics for these cohorts was dramatic. On digging deeper, it turned out that fresher recommendations that are based on user’s recent actions are just a lot more effective. But why new users? Well, turned out that we had special code for new users. Because we did not have any prior data for new users, we had to put in code to recompute the recommendations for them in our online serving layer itself, ensuring the use of the freshest underlying signals for them.

With that data driven insight, my team quickly worked to change the pipeline to aggressively recompute the recommendations as frequently as possible, limited only by the capacity of our Hadoop cluster. Looking around the company, it was actually not difficult to notice a pattern emerging in the backend infrastructure powering a lot of Twitter product features. System after system was finding that real-time computations were much more effective in delivering product goals and were simultaneously getting disenchanted with the latencies and best-effort-ness of Hadoop.

I had become a full convert by this time. I knew that we had to come up with a way to compute recommendations not just quickly, but in real-time. This is not only a hard technical problem but almost seemed impossible, as the computation involves pretty sophisticated graph traversals and machine learning that are simply too complicated to think about doing in real-time. But I was on a mission by now, and pressed my (excellent) team. They had big concerns. Here is my recollection of how the dialogue went between me and some of (really strong) engineers in the team.

Me — We’ve seen how great fresh recs work. We got to make them even faster, perhaps real-time…

Team — Let’s somehow try to run the Hadoop job every hour instead of every day. (also mumbling: but that will take up a lot of Hadoop capacity. We are already one of the biggest users inside Twitter.)

Me — I think we need to do this in real-time. Let’s move to a Storm based pipeline?

Team — Umm, what use is that ? These are just recommendations after all. Why do they have to be computed in real-time?

Me — Well, we have seen that the fresher the recs are, the better they perform.

Team — Sure, but may be we can try first doing it every hour. Besides, what is the point of computing a rec for someone who is not even going to see it until she visits Twitter next time, say the next morning. We will be wasting so much compute resources.

Me — You have a point there…. Oh wait! Aha! We not only need to compute these in real-time, we need to push these in real-time too ☺These can be like notifications!

Team — Wait, what! That doesn’t make sense at all. I mean even if we compute them, we can enqueue them into our email queue and send by email. What’s the hurry in a user knowing right away that a rec has been generated for him?

More arguments and discussions followed back and forth. The team had really good points. It was indeed too expensive a problem to solve without a clear product win. But I was convinced that it would be a win, and to the team’s immense credit, they were eventually game to try. We found a way to test it by creating a test Twitter account (making it initially protected to not get in the public’s eyes). Whoever followed the account would get these recs by Twitter direct messages (DMs) so it would be an opt-in feature in this testing phase. The team came up with a somewhat hacky, brute-force solution which worked fine because scalability was not yet a concern until we knew the product feature worked. I started recruiting my colleagues and one by one purely by word of mouth it started getting popular.

Somehow, Dick, Chris and the rest of the exec staff got wind of it. They started following the test account, liked it (not all of them wanted to launch it to my unpleasant surprise, but let’s leave that) and we were ready to make it a real-product. We first un-protected the Twitter account and then worked on delivering these recs as mobile push notifications (instead of by DM from an account you had to explicitly follow). The launch that happened in September 2013 opened it up to tens of millions of users. (For the technically curious, before we could do a full launch, we had to solve the problem of scalability which the team — now armed with a lot more confidence in the success of the product — did manage to crack, significantly enhancing the state of the art in real-time graph processing at scale. This was recently published as a recent paper in VLDB).

I can only say that MagicRecs has been quite successful for Twitter. The strong team behind it has taken it from strength to strength, and it has been received enthusiastically by users. The idea inspired other product features inside Twitter and seems to have at least partially inspired others in the industry as well (Nuzzel, ThinkUp, etc.). Even the phrase “MagicRecs for X” is sometimes heard. It is not hard to extrapolate how Twitter (if it wants to, as I don’t know its plans now) could take it forward in a notifications-centric mobile experience as seems to be enabled by both iOS8 and Android platforms.

Back to being real-time

MagicRecs worked especially well only because it was real-time. The median latency of delivering the recommendation after the triggering event is in seconds (some more technical details are outlined in the VLDB paper mentioned above). In the beginning of the project, I was adamant that the copy of the notification (e.g., see some here) include exactly how many seconds it took, in order to create that sense of timeliness of the notification. What was being recommended was not new content, just summarized and pushed in real-time to the user.

On the face of it, this is not new. Remember the time when you had to run a spell-checker at the end of your document versus now being told in real-time while typing? From Google to Uber, people have realized the benefits of sheer speed. Google’s findings at improving the speed of their search results page is now folklore. And that is my point — despite the need of speed being folklore, I still find big resistance to making products real-time. Perhaps this is because of an inherent bias of the curse of past knowledge in a lot of us — those on the engineering side presume that real-time processing must be hard or expensive (it generally has been) and those on the product side like to question whether something really needs to be real-time? What I have found is that it pays to give the benefit of doubt to being real-time.

The flip side of anything that is not widely understood is opportunity. We in the Internet community have clearly made a lot of strides in improving speed but it is in no sense a done deal. Some examples below: (I hope new startups will continue to work away at solving the technical and business challenges to make them a reality.)

• Uber and Lyft give instant gratification as they try to service a request in real-time. However, I currently use them about twice a week and I think would use them twice a day if the average latency from me requesting a car to sitting in one is less than a minute and not the current ~10 minutes (averaged over many locations and times in San Francisco). Small difference you might say, but I find myself preferring driving to taking Uber (when other factors like parking are less of an issue) just because I want a predictably low time to reach a certain place. The lack of predictability forces me to add on a further 5–10 minutes to my estimate.
• When Amazon introduced free two day delivery to Prime members, it seemed magic at the time. I don’t know Amazon Prime’s de-activation rate but I have been hooked since. I think free same day delivery is in our near-future and will have a similar effect on users in the long term. In fact, I hope for a world in which delivery is with in an hour (besides being free).
• Talking about deliveries in hours, the other day I tried to use Postmates to order my favorite Walnut pie from a shop 15 minutes away in SF. The delivery time I was quoted was 1 hour. That is not real-time. I did not order the pie — the urge for pie would have long gone by then. Likewise, I still can not get a cup of hot coffee from my nearby favorite cafe delivered to me at home cheaply.
• I use Hipmunk for my travel bookings, and it is of course much better than working with a travel agent. But it is still too slow (and painful) to do planning of even the most simple flight/car/hotel itineraries. My time waiting as the site contacts various airlines and tries to present me options might not sound like too much, but if you have to tweak your search multiple times, it gets annoying fast.
• On the technical side, one example is of Hadoop/MapReduce that now seems destined to be relegated to important but somewhat specialized tasks of background model building and daily analytics and charts — and even those need to be complemented by real-time versions. Another example is of a/b testing frameworks that need to be real-time. The ones prevalent in industry today take days to get useful data and that is really too slow. As a simpler example, programmers prefer using an interpreted language for small tasks that skips compiling partially or completely. Personally, I am hoping for much higher abstractions than todays’ languages provide to allow us to decrease prototype times by an order of magnitude. Alas, too few people (if any?) seem to be working on that.

Real-time’s unfair advantage

There are many more important aspects related to real-time — noise, rawness vs relevancy, expense of computation, combination with non-real-time processing — but this blog post is already too long. There is something about real-time that appeals to us humans at a fundamental, psychological level. Indeed, economists and psychologists have systematically studied Time preferences of people. In particular they have identified a cognitive bias called Hyperbolic discounting — “Given two similar rewards, humans show a preference for one that arrives sooner rather than later.” This discount in perceived utility with increasing delay is not at a constant rate, it falls much more rapidly in the beginning.

Chris Anderson wrote in his 2009 book ‘Free’ that “If it’s digital, sooner or later it’s going to be Free” and that “You can’t stop Free”. I think we can say something similar: “If it’s going to be delivered, sooner or later it’s going to be delivered in real-time” and “You can’t stop real-time”.

(Image credit: Wikipedia)