Prashant Singh is one of the most original thinkers I know in the Indian startup ecosystem. An ideal career for him would have been writing fiction, but he is, disappointingly for fiction readers, delightfully for us in the startup world, now golden handcuffed to product management! I think of him as a Product Manager’s Product Manager; conversations with him always lead to interesting insights and perspectives, a different take on the world, or a different way to look at the world. For those of you who want more of Prashant, after reading this transcript, check out these two threads on the next billion users of India – one, two. These were written when Prashant moved to his hometown of Bikaner in Rajasthan during the COVID-led lockdown in India in 2020. He is @pacificleo on twitter – if you are curious why that moniker it is because Prashant means peace / peaceful (hence ‘Pacific’) and Singh means lion (and thus ‘Leo’). Well!
In this conversation, Prashant shares his take on PMF – a need is being served and in a profitable way; it may or may not have scale, but there is ideally a sublinear correlation between growth and a key parameter, like cost of acquisition of customers, that is, when you are growing the cost of acquiring customers comes down. We then move to look at his startup Shifu and why that didn’t achieve PMF. This is a fascinating case study. TLDR – he essentially focused on the wrong persona (consumer instead of OEMs), and built too far along that axis to be able to pivot easily. He goes on to share his views on the key metrics that founders should track – app open rate and time spent are key for him. The final sections are where Prashant opens up to to share his distinctive takes and views on a bunch of stuff – in particular i loved the metaphor of waste paper basket (or rejected folders) he uses to illustrate why as much as looking at the final output of a great initiative or final rendering of a great product, you should look at all the rejections / thrown-away scribbles and see the trade offs made and understand why the decisions were taken. (There is perhaps a great podcast idea here).
Please find below the (edited) transcript of my conversation with Prashant Singh. Enjoy!
(The transcript has been edited to make it more readable. There will of course be errors; please excuse me for them. The interview took place on Zoom on 10th November 2022.)
Sajith: I typically start these conversations off by asking, what is your definition of product market fit? Do you have a specific term? Do you even think about it? In the context of Jar and also before that, with Paytm, initiatives that you led there, at Shifu, what is your definition of PMF?
Prashant: The way I think of product-market fit is an event where a product, a service or a company stumbled upon a solution, which serves the articulated or unarticulated needs of a consumer.
For example, is there a product-market fit for pet food? Now, it can be a very small product, it can be a profitable niche. Are there enough people looking for it? Like if you go to Japan, there is a product-market fit for repairing (pen) nibs. People can make a decent living repairing the golden nibs of a fountain pen.
Sometimes people confuse it with scaling. Scaling is a function of TAM and everything else but product market fit – it is a solution made for unserved needs. When you apply it to the startup context, it means that a need is being served and it is also being served in a somewhat profitable way. So when we say profitability and I don’t necessarily mean it has to be revenue-generating, I mean that some cost decreases as you do more of the same.
For example, when WhatsApp signs up a new user, the cost of acquiring the next user, because it’s a fundamentally network-based product, goes down. I sign up and send an invite that says “Hey, I’m using WhatsApp” to my family members, so just by acquiring me, they acquire 10 more people.
So if I were to put a numerical filter to this, I’ll say there should be a sublinear correlation between two parameters in your business. It can be with every user who comes, the cost of acquiring new users is reduced. So like in a linear correlation, if x increases by delta, the y also increases by delta. In a sublinear correlation, if x increases by delta, y increases by less than the delta.
Let’s say my cost of acquiring customers has gone down, which is a good thing. In the same way, if I’m DropBox, with every customer I acquire, my cost of servicing a customer also goes down because I can buy bulk storage, negotiate better bandwidth prices, etc.
So when a good PMF is there, some parameters are sublinearly correlated and then it is beneficial that we pick up one parameter that increases and everything else which goes in a positive spin with that. For example, if you look at food delivery, by acquiring every new customer, does the cost of food delivery goes down? Not necessarily, maybe or maybe not, but the overall volume makes up for it. Then you think in a sense – okay, if I acquire a customer and he starts ordering food online, it will still burn as much fuel and it will take as much time for a delivery guy.
Sajith: I think the definition is very interesting and nobody has defined it like this that PMF has a sublinear correlation. Now, let’s say you have founders who come to you and ask you, I don’t know if I’ve hit PMF or not. What would you say are 2-3 quick and easy ways to check? How do founders know if they’ve hit PMF?
Prashant: They have to ask themselves what does growth mean to me. Like if you are a fitness freak, what does growth mean to me? It means BMI ratio. If I’m a ride-sharing company, the number of rides going at any point in time. For a bike-sharing company, it is essentially the cost of servicing a bike. So you have to see which parameter is the best proxy or best KPI to show the system health and then can I basically increase that parameter? What are the levers?
Let’s say I’m Yulu. I can increase the uptake of my ride by sprinkling all my bikes near some college campuses. The number will go up. Is it PMF? Maybe, or maybe not, but will they come again? If there is a real need being served, they will come up again and pick it up. For that, I have to service it again, recharge it in the night and then I’ll figure out what’s my cost and then this cost of servicing and running the operation is a net positive. That’s where you figure out that I have hit a PMF when users are returning to use my service without me having to entice them. Whether you fulfill that service with quality or in a profitable manner is secondary. That’s more of an operational and organizational issue.
Product market fit is basically the market situation. Market ko iski demand hai ya nahi. (Does the market demand your product or not?) This is a very wrong notion that a company attains product market fit. A use case attains PMF, the company never does. Most likely the company will fuck up the product-market fit. Like short-form video, like UPI, they have attained PMF. If the Paytm server is down, users will go and use any other UPI app unless there is a lock-in.
Sajith: Per your definition, did Shifu attain PMF?
Prashant: Yeah, we failed to scale because of a variety of reasons, but in the sense, some people were using it in ways which we never really envisioned and we were not even expecting or even imagining them to use it.
Sajith: Can you take five minutes and take through the Shifu story, which is not as well known, as I’m also collecting case studies.
Prashant: Grand vision for Shifu was as delusional a use case as a first-time founder can come up with. So, I used to have the latest smartphone because of the industry I was in, we used to get sample units and I used to play with it and I was very frustrated because every day I was doing the same thing. Every day, I woke up and there was a neighbourhood autowala back then, I used to call him, he came to my house and then I used to call my colleague – “kahan ho” (Where are you?), I used to call my boss – “Mai late ho jaunga.” (I will be late.) Every weekend, I used to call my mom. Every time I went to Gurgaon, I used to call a certain friend, but my phone book was always alphanumerically sorted. I used to get a lot of phone calls and it used to ring in the middle of my meetings and I was so frustrated Yaar isko dikh raha hai na calendar mein ki mai busy hoon. (Why can’t my colleague understand that I am busy from my calendar and call me OR why is my phone allowing it to ring when it knows from my calendar that I am in a meeting?) Why is this phone ringing and can’t I set a rule that it should ring only if it is a phone book number or it should ring only when Vijay Shekhar Sharma calls or some investor calls. So the idea is – can you build a system that may learn from your behaviour? It’s a smartphone for all the smart in the name but not smart because it is not learning from my usage pattern. When we dug deeper, we realized that it was an interface problem – the interface was designed at a time when there was no AI and there was no persistent network connection.
So we thought, how do you design a phone today if there is an AI, if there is a continuous connection with the network and you are mining all your data? So what will your phone book look like? You will open the phone book. Will it show you an alphanumerically sorted list of contacts or will it show you that out of 2000 contacts, based on your history, these are the 10 people you are most likely to call? If you are saying you are turning the phone silent, don’t turn the phone silent, I’ll make that judgment call, if the call is coming, whether you want to accept this call or not and then if your mom is calling, maybe we’ll turn it on silent but if the boss is calling we will blow the horn in the middle of the meeting. In that case, you will trust the judgment of the phone.
So the phone should be able to and there’s enough data available on the phone to make those smart decisions and we proved it. We built phone book integration. Let’s say I get a lot of missed calls, I am sitting idle in the Uber in the evening. It used to remind hey why don’t you return this call. There’s this phone book number, you have a history of talking to this guy, this uber ride is of 15 minutes, why don’t you call? So you’re trying to guess those things and the phone becomes more like a very intelligent personal assistant and there’s enough data to do. These are automatic algorithmic features, but algorithmic features requires some data to begin with. To overcome the cold start problem, there were features like can you assign some task to your personal assistant – remind me when I have 10 minutes, I need to make a call to Sajith or remind me to buy a deodorant when I am at a mall or remind me to eat biryani at certain place when I am in Hyderabad or remind me to call my friend from college when I am in San Francisco. You set this thing and you forget and then next time when you are in San Francisco, Shifu will remind you to call this guy. So those are the context-sensitive alerts we built and both of these things were very popular. We got a lot of traction in Brazil, we got a lot of traction in Russia and we were featured by Time magazine as an innovative app. Even Google listed us as the most innovative app on the play store. We got investment from the person who used to head Search at Google – Amit Singhal – he invested in us. We had a good run.
Sajith: What was the use case which got the most appeal?
Prashant: Miss call reminder was a hero thing and then basically location-specific reminder and the third one was, for example, if I’m talking to Sajith, remind me that he owes me money. There’s another version we’re building that if both of you have Shifu installed on your phone and you’re in close proximity – it will tell these are three things you wanted to talk to him about. There was another feature that got popular in Southeast Asia. It was a feature that reminded you to charge your phone even if it is at 70-80% as you would want to play a game or listen to music or watch a video on the ride back home.
We actually started Shifu with a very small team of 5 folks. On day one, we wanted that all 832 screens on the phone should be AI-powered but that was a big herculean task. We started building with miss-call, location and phone book features, but they became a patchwork of solutions. I guess, like every startup, we were ahead of our time.
Sajith: When you were building Shifu, did you formally think about the PMF concept? Did PMF ever come up formally in a sense that you sit and work towards PMF?
Prashant: I had read Steve Blank’s book, which gave me some idea about PMF, but we were in the middle of the Facebook growth hacking days, and the month’s flavor was retention numbers. PMF as an umbrella term was not like a go-to question.
Sajith: Would you say that you got PMF in Shifu because of the growth and retention numbers and traction?
Prashant: We never spent a single dollar on user acquisition. People were coming on their own and they were figuring out the use case themselves. For example, our users in the Middle East liked the feature of turning the phone silent because five times a day, they have to go for namaz. Our whole business plan was that we’d figure out a complete solution with the data set. Then we’ll go to OEMs and say, look, here’s the data. Let’s build a phone which is customized for Middle East requirements, which is customized for an urban commuter in Bangalore, which is customized for a high-profile executive and let’s build it. There was some structural mistake we made that decision-making was being done either in Korea or China. In India, OEMs embed whatever they get from Korea. No one goes at the firmware level.
We came very close to being acquired by DropBox at one point and we came very close to being acquired by another company you might have heard of – CyanogenMod but both fizzled out primarily because we never had (U.S.) work permits. It was a time when they didn’t really acquire companies because you and your team didn’t have work permits. Now, they have this process that you acquire companies anywhere, put employees on Facebook London’s payroll for one year, you work at Facebook London and then you have an intercompany transfer. One thing I would do differently is the moment I get my first seed funding, I would have processed L1 for my team and then you would be ready for acquisition. We were counting and thought we would get $10-15m easily.
Sajith: So you didn’t get consistency with the whole idea and created a lot of patchwork solutions which didn’t embody well with your initial thought process, correct?
Prashant: The phone is a very personal thing. There are spaces in the phone, there are nodes in the phone which are customizable today. You can customize your wallpaper; for example, you can put your kids’ photo, your dog’s photo or Lord Krishna’s photo. Your phone is very similar to mine, right? But if I start changing that and I don’t change it across the world. What to expect, then? Now I’m in the phone book and I expect to have an alphanumeric listing, but you realize it’s the relevance-based listing and then suddenly shift the expectation. Now, if you will open settings or something on your phone, you will not get the same experience because we have to do something incrementally. So one way to do it is – you basically have six different applications or you keep on adding new screens to the existing app.
Sajith: If you could redo it today, what would you do? If the Prashant of today could go back and launch in that era, what would he do?
Prashant: The thing is that people usually find some other way of doing the same thing. For example, when FourSquare fizzled out, nobody came to take over FourSquare(‘s idea). It was Facebook that integrated all of those things. This time, I will not focus on user acquisition. I will focus more on building technology and refining my algorithm and working more closely with the OEM partners. Approaching GTM in a customer-centric way was not beneficial unless you are Carl Pei or Andy Rubin, who can build their phone because, at the end of the day, it has to be a deeper integration with the phone.
Sajith: On PMF, do you feel there is a specific framework you have in mind that you can readily give to other founders to think about PMF?
Prashant: I’ll first ask what activity metrics you are tracking and people will always tell you that certain metrics are very important. For example, if you are launching Evernote or Dropbox, the first question people will ask is how many paid customers you have, but I don’t think that’s a good PMF sign. That is an optimization problem. The PMF sign is how many people still retain your app on their phones. As long as the app is on the phone and new data is being added to it, it is a good sign. Eventually, that data will become important and somebody in a month or six months will convert. So, I believe some statistical correlations happen no matter whether we engineer it or not. If enough people are opening the app, then you can put anything there and sell it. There is a non-zero chance that you won’t be able to sell your product/service on that app. If I have a million users opening my app for some reason I don’t understand, 100K of them will add to the cart. If 100k of them add to the cart for some reason I don’t understand, 10K will go to the payment page and then 2K will successfully order stuff – I don’t understand why it is happening. Now, this is an optimization problem. But the question here is – is there a need you are serving?
I think a very good metaphor to look at it is a shopping mall. Most people think of their app as a standalone shop where users would come on their own and buy things. Unless you are Prada or something (which is very rare), you won’t get the traction that you want. Most consumer products are like a shopping mall, where you have everything ranging from a movie theatre to food court to gaming arena. People will go for one activity but will buy a few other things too. Even after that, you have to do pop-ups every day or every weekend, you have to throw some flea market because they know if you put on some dance show, people will come to watch it and some of them will go into the mall and see a shoe on the display window, some of them will end up buying that. The Law of large numbers will work. When everyone is trying to become a super app, I think somebody needs to understand and study the function of a shopping mall which is not studied properly.
Sajith: What are the metrics that you would say one should track? Retention seems to be one of them.
Prashant: App open rate or time spent – these are the leading indicators. How many units are being browsed and then how many are actually added to the cart – these are also interesting metrics. If they are not buying, then it is a separate problem. I talked to this guy – a salesman at one of those cosmetic shops. One thing they continuously observe is that some ladies pick expensive transparent lipsticks, which seem to be popular these days. The next day, 80% of the people coming are asking the price for the same, but they obviously don’t buy because it is too expensive. Now, store owners would place them such that those expensive lipsticks are visible more and induce people to at least try them. Now, people see themselves in that expensive lipstick and it looks good and they have a credit card, so the empowerment is already there. Suddenly their original budget of ₹450 is increased to ₹1000. They consciously engineer this. So there’s a lot of elasticity in the purchasing power if you basically put the right FOMO-inducing hacks.
We don’t give enough credit to physical retailers. They have a conversion factor of 30%. If 100 people walk into the shop, there are 30% of people end up buying something. In an online setup, if a million users come, maybe only 10K will buy.
Sajith: How do you track retention or churn?
Prashant: Depends on the category. If you are a car seller, what does retention even mean? It’s not going to come back. If you are a sneaker seller. I haven’t bought a sneaker in like three years. Maybe a shoe cleaning kit will dilute my brand. So you have to understand what is the approximate repeat frequency. One mistake a lot of people make is they think everyone is obsessed with their category. If I am buying a shoe, I want to buy six shoes. There are sneaker ads in which the actor buys six shoes. No sir, I bought one shoe.
For retention, I think a good way is to see whether retention should be in the same category or should be basically cross-selling in adjacent categories. If I sell somebody a shoe, maybe next time, should I sell them warm socks or ankle socks or something like that? Selling shoes is the most important thing for you. One company which do it very well is Allbirds. They are in the business of selling shoes, but they have gotten into other categories also. Because the person who is buying culturally or climate-conscious shoes, would also like to buy climate-conscious t-shirts.
Sajith: The people who got PMF or have grown fast. Is there a commonality between them? Are there certain things they do better than the people who don’t grow fast or don’t get to PMF?
Prashant: Are you saying commonality in terms of organization or commonality in terms of the founder?
Sajith: Anything. Both. What gets someone to succeed faster?
Prashant: I think adaptability. Having the humility to see that this is working or this is not working. The thing is, it’s very counterintuitive that when you start a company, you think this is the thing that the world needs and if I will not do it, it will not happen. Then you go out and realize that the world needs are slightly different and your life’s purpose is slightly contaminated. There are two ways – both ways are perfectly valid. Steve Jobs’s way – no separation between hardware and software or Bill Gates’s way, which asks for a separation. You have to know what is your extent of adaptability and then that will dictate who you serve and what price point you operate upon. You can’t operate an Apple business at a Microsoft price point, and you will never have 100% market share. You make peace with this.
The thing is, a lot of the time market is clearly telling you that it is not working. Sometimes people have this religious-like allegiance to data. They think data always shows the truth, but it isn’t true every time. For example, I remember reading somebody made a business case – At McDonald’s, they sell cheap burgers and give a toy as a gift. They should flip the model. They should sell expensive toys and to invite customers, they should sell free burgers. A lot of people will come and they will buy. But in reality, we all know that will never work. With Excel, you can write a lot of good fiction. One can write better fiction in Excel than in Word.
Sajith: Are there specific readings, books, or videos you recommend for founders? It could be related to PMF, growth, retention or traction.
Prashant: You may read books like Four Steps to Epiphany by Steve Blank but usse kuch nahi hoga (it is not useful). It’s like thinking that you will read Kotler and you will learn marketing, ye sabne hi padhi hai (everybody has read that). There is a lot of lousy marketing going around. So I think a better way is to find somebody who has done that journey and get them to talk to you in a non-public forum where they actually impart real knowledge. The best source of knowledge is the waste paper basket. All those plans that were discarded. To give you an example, we were trying to talk to an ad agency and they said, everyone who comes to them says that they want an ad like Cred. It’s a good thing. Good for agency business. Suddenly, every consumer internet company wants to do Cred like ads. The problem with this is, if the ad agency showed them the same ad, they would have rejected it. 90% of founders would have said, who knows Indiranagar. Instead of looking for Cred like thing, you think that it will fail most likely but we are still gonna attempt it.
Number two, look at the concept they rejected. A lot of people don’t do this. Nowhere in the world have I seen people say these are my rejected folders. That’s what working together teaches you. It’s not the final product. Final product everybody sees and deconstructs. It’s the journey that matters. You start with something totally different and then, in an iterative fashion, reach a final product. In a public domain, you can say, maybe somebody from Cred can come and say, this is how we approached this ad. These ideas were considered and rejected. This was debated and there were a lot of heated arguments.
Sajith: The metaphor is this wastepaper basket. What did not work, right?
Prashant: What did not work and why? For example, there’s a path dependency on every decision a startup takes. From the outside, what we do is basically copy the outcome. We don’t understand the underlying logic. Example-wise, when demonetization happened, we had a simple API (in Paytm) called CheckBalance. Now the CheckBalance API was designed for like a million or 10 million hits a day. Suddenly that goes 10x. Hence, it was not scaling well. It worked like this – every time a user came on the homepage, the CheckBalance API was auto-triggered. We solved it by placing a button for fetching balance which ultimately reduced the number of queries and became sustainable.
Our competitor, I’m not going to name who, saw it as a security implementation. They thought that people did not want to see their bank balances on the homepage. They changed it by copying it. They didn’t realize it was a scalability issue. People actually wanted to see it on the home screen. After like 15 days, we refactored the API and it was back up again. So we were doing it on our own. Somebody was mimicking us without knowing the context. So, thought process was necessary.
Sajith: Is anybody else I should speak to in your recommendation? Who else should I speak to? Who are the two or three folks I could speak with?
Prashant: I think you can talk to the UltraHuman guy. A lot of energy and needless energy is being spent in articulating and defining PMF but I think it’s high time, as an ecosystem, we move the discussion to the next level, how do you build an org and processes that increase the probability of you stumbling upon PMF soon. Because one thing is given – reaching a PMF is not a deterministic process. That requires a lot of experiments. Some of them will go wrong. If you see the amount of buzz UltraHuman has created, that’s one company that is not being under the scanner enough. For example, do you know that one of their content creators used to work for Stranger Things?
I was talking to somebody in your venture capital community and they think, interestingly, everyone talks about a lot of experiments but no one actually implements it in their org. The only place where they have seen a lot of experiments, a lot of iterations, some kills, some double down, and some new verticals are Swiggy and Razorpay. Machinery is being created at only those two places in India. Machinery where a lot of experimentations are being done, which works – double down, which doesn’t – scrap. What is common between these two startups? Actually, for a surprising reason, you will find those founders were not involved in day-to-day operations. Founders were able to separate their day-to-day things and people in the org were releasing, optimizing, scrapping, and doubling down on their own. The machinery was there. The founder was dealing with bigger issues. A lot of time, that’s a very interesting signal. I don’t know whether it’s the primary variable or not, but if the founder can get out of daily operation, a lot of magic starts happening for some reason. Even employee retention increases. Not that the founders are tough people to be around, but you get more ownership.
Sajith: Thank you for this. UltraHuman, I have made a note. Even Sameer of Nexus told me about this one. So thanks for this and if anything else comes to mind, send me voice notes. I’d love to get voice notes from you.