The Cold Start Problem

Posted on by
  • When you first launch a product that depends on the Network Effect, you will have to get enough users to it quickly otherwise the product will die.
  • A network effect describes the value that is associated with a Product as more people use it
  • Telephone was the first network-effect product way before Uber
  • Software companies that capitalize on this network effect typically do not own the underlying assets – think ABNB and UBER.
  • Early on in the life of Slack, the head of Customer Experience would personally respond to support tickets and tweets
  • Slack incentivized Enterprise clients to adopt the app org wide by offering enterprise-level features and discounts for companies that have BUs that adopted the app in silos
  • Slack tested the app and new features internally and perfected it before rolling it out to external users
  • You need to appeal to the core part of the network by solving a hard problem. The core part is the set of users that are responsible the most for proliferating the usage of the product across the network
  • In order to help make a network effect positive for the adoption of a Product, you need to determine the minimum number of users needed for the success of the Product and also the threshold for Product usage in order for it to become sticky in the network.
    • For Slack, they found that a network would need a minimum of 3 users in order for it to be long-lasting. They have also found that once a team has sent 2000+ messages through Slack, they are 93% likely to keep Slack moving forward.
    • The same idea could be generalized across other startups by plotting the number of users on the x axis and an important engagement metrics on the other axis.
    • Airbnb and Uber require a much bigger network because choice matters. These are marketplaces where a user might want to see a dozen properties before making a decision. Airbnb co-founder was very data-driven and found that in order for a new market to take off, you’ll need at least 300 listings with a 100 reviewed ones.
    • Uber tries to optimize based on how fast they can get a person a car – the ETA. Should be 3mins. When Uber enters a new market, the ETA could be 15mins and Uber focuses on bringing this down as fast as possible. This means that in markets like LA, you have to pour in a lot of effort to provide that kind of ETA, but in the long term you are also establishing market dominance. New startups that rival Uber cannot compete with them in the dominant markets by solving for the same cold start problem – focusing on the ETA too
  • Every Product has a threshold that it needs to reach in order for it to become sticky and for the network to become healthy. FIguring out what that threshold is becomes important from a product launch strategy perspective.
  • Successfully creating a network and solving for the Cold Start problem requires a team that focuses on enhancing the depth and breadth of a network. You also need to focus on the right set of users, not just any users – quantity. These users need to create an atomic network that can subsequently grow.
  • If you can create one self-sustaining and stable network, this means that you can create a second.and the 10th. 100th etc.
  • Case in point: the first credit card network. Launch by BAC in 1958 in a small city in California where BAC had a big client penetration. BAC decided to mail the card to 60000 people – unsolicited. This helped create momentum to help sustain the network. At the same time, BAC wanted to sign merchants in the downtown corridor of that city. Focused on small merchants cz large ones like Sears had their own credit program back then.
  • Components of a successful Atomic Network
    • The networked product should be launched in its simplest form – not fully featured, so that it has a dead-simple value proposition
    • Target should be building a tiny network – the smallest self-sustaining network possible. Ignore size and focus on depth at this phase.
    • Attitude towards launch should be: do whatever it takes to make it scalable or profitable, regardless of whether it is unscalable or unprofitable, without worrying how to scale.
    • Growth hacks – strategies that help create the network, can help launch the network. Examples include: referral fee – PayPal, invite-only sign-ups – Slack, free ice cream – Uber rideshare
  • The size of the atomic network: should be in the magnitude of hundreds to start. The smaller the size, the easier it becomes to create and manage the network. That is how Uber started. For other products though like Workday, an entire company needs to adopt the product in order for you to find your friends or other people you connect with on it. Thus, the strategy here becomes different: you’ll rely on the company to mandate it.
  • Importance of atomic networks: It’s simple: once you have created the first successful atomic network, you can create another, and so on.
  • Acquiring the hard side of the network: This refers to the set of users on a network who bring a disproportionately higher number of users to the network. Each network usually has two sides, and these users has the propensity to bring a high number of users on the other side
    • Uber drivers drive down prices and ETAs and will bring more riders. And vice versa
    • Content creators in social media websites will bring more visitors to the site
    • Wikipedia has billions of users but less than .02% of the users make the majority of edits and content creation on a monthly basis.
  • Questions on the hard side of the network: important to think about these questions as you launch your network
    • Who is the hard side of the network?
    • Why would the network be appealing to them?
    • how will they hear first about the network?
    • Why will they stay on the network as it grows?
    • Why will they not jump ship as the network grows?
  • In a network, the (1,10,100) rule applies: where 1% of the users create the group, 10% participate actively and a 100% benefit from the 1% and 10% groups.
    • This phenomena also holds true in a domain like YouTube where 20% of content creators get the majority of activity.
    • The SEO of Snap has a slightly different take on it: You can think of technology and communications networks as a pyramid
      • At the bottom of it is people wanting to communicate with each other. This is what Snap enables people to do: people are comfortable communicating with friends
      • Second layer is about Status. this is what social media enable influencers to exhibit.
      • Third layer is about Talent. Those who possess the talent in a dance or another field and can communicate that effectively to audience and in a fun way – TikTok
  • The more difficult the work needed to be on the hard side of the network, the less the number of people who will do it
  • In a network, it is important to focus on the hard side of the network so that messaging, product functionality and business models are all aligned to serve them
  • Dating has a network effect phenomena. Early dating services did not work because they shared profiles like classifieds. Then came the generation of eHarmyony and OkCupid. They worked better but slowly started losing appeal with the advent of dating apps. CEO of Tinder describes below some of the things they do differently:
    • gamifying the experience: you can do it while you’re waiting for the bus. Before, you had to compose messages and send them to prospective dates, often from a computer
    • Action required was simple: swype
    • worked quite well for the core group, the hard side of the network, which is attractive men and women.
  • Hard side for marketplaces is usually the supply side.
  • for Uber, 20% of the drivers did 60% of the trips. They were called Power Drivers.
  • The Key learning from Tinder – where they have disrupted existing solutions, is that you’ll need to find the hard side of the network, and look closer to see if it is – or a subsegement of it, is underserved, and create an atomic network based on that.
  • One of the major keys to determining where atomic networks can be found is to look at underutilization – of time, of assets, or both. Look at how people spend their nights and weekends – underutilized time.
    • Uber capitalized on the underutilization of cars
    • Craigslist and eBay
    • ABNB: underutilized homes
  • One of they important keys is to examine the existing network effects and determine whether the hard side is being underserved. Example: You can say that in order for a dating network to be effective, you’ll need men and women. However, upon a closer examination, you’ll realize that women aren’t just looking for any man and vice versa. There needs to be more specificity in determining what the real hard side is – attractive men and women who also fit certain criteria.
  • Atomic networks often start at the low end, and as the hard side gets more engaged, they will start offering a more premium product/service. Airbnb might have started with airbeds, but homeowners were soon willing to offer a room, then a whole apartment.
  • Zoom: a successful networked killer product. Offered a frictionless user experience – user can join in a single click, and high quality video. No need for meeting codes
  • Networked products facilitate interaction between users while traditional software emphasize the interaction between the user and the software. Networked products grow by adding more users while traditional software grows by adding more features that support more use cases for the users
  • The most important feature of networked products revolve around how users find and connect with each other – whether it be people you may know feature, uber, buyers and sellers, etc
  • The ideal product to drive network effects would bring two dimensions: a simple concept that is easy for users to quickly understand and also ability to bring an infinite, rich group of users that would be hard to replicate.
  • In the initial days of Tinder, the founders turbo-charnged the growth of users by throwing a lavish birthday party for one of the co-founders nephew who attended the school at the time. The person was hyperconnected and very social. The catch to attend the party was to download the app – a bouncer checked for that at the entrance door. As students – 18 to 21, mingled during the party, they realized the day after that they had another way to continue the conversation with those who they met the day before: Tinder. Tinder started replicating this successful model – and more easily, across campuses.
  • LinkedIn: Started as an invite-only network. Seems counterintuitive as you’re trying to get as many users initially as you can – validate the idea and turbocharge growth. While FOMO or Product iteration or scalability reasons could be cited as reasons why an inivite-only initial approach – which worked for Gmail, the core reason would be to replicate the network over and over again by having people invite people they know – people tend to be similar to them.
    • Linked founders and employees could invite anyone they know. Shortly afterwards, the network opened to the wider society. This also helped focus the initial launch on mid-tier category of people: those who have time to network but was also well-connected.
    • LinkedIn did not advertise the network initially as a job-search network. It advertised it as a professional network – with the ability to look for jobs just like the many other features available.
    • Other reasons why the invite-only model works: users who join feel that they’re in a familiar environment. Also, the most socially-connected users join initially, and they bring other socially-connected users too.
  • Come for the product but stay for the network: getting users to join because a product is great but then pivoting their usage of it to ways that capitalizat on communicating and interacting with other people on the network
  • Instagram: Initially, Hipstmatic was the app that validated the demand for modifying mobile photos and sharing them on social networks. Among its advantages included the opportunity to give a retro feel for pictures – something that proved very popular. However, the user experience was clunky: you had to wait for a few seconds to see how your picture would look like after you apply a certain effect to it. You also had to share the photo from your saved photos library – app offered no feature to do that. This is where Instagram came to the picture: initially it had too many features such as sharing location with friends and capturing photos. It quickly became evident for the co-founders that they need to refocus on one thing and do it well. Result was Instagram
  • Pulling off a “come for the product and stay for the network” is not easy. Users might not seamlessly adjust to the change in behavior this requires.
  • “If you have a chicken and an egg problem, buy the chicken” this is a good analogy of what Uber did in its early launch: they put ads on Craigslist guaranteeing the payment of the 30/hr regardless of how many riders a driver got, as long as they have kept the app on. Historically, companies used coupons as a way to create a network effect: subsidize the presence of your new product at a grocery story so that customers see it and try it. Becuase if customers do not see it, they will not buy it, and if they don’t buy it, you won’t be able to sell it – classic chicken and an egg problem
    • This strategy became quite expensive though for Uber and they had to switch to other strategies to incentive the fee-based model: encourage drivers to have their friends sign up for a 200-referral bonus each, surge pricing, get discounts on trips, etc
  • You will need to create a killer product and an atomic network before reaching for the financial lever. Startups that do not have a lot money can find a subsidization strategy to be quite costly
    • But once a company can successfully launch its first atomic network, financial levers can accelerate the journey to the tipping point.
  • Using financial levers can be a powerful tool in getting a networked product to reach the Tipping Point faster. This is the case with Bitcoin, and Microsoft. These financial levers are particularly effective in networks that are close to money
  • Bitcoin: one of the biggest value networks that has been created in decades. A deflationary currency. Its supply will not grow regardless by what financial entities decide. Miners get rewarded for preserving the security of the system through receiving bitcoins for dedicating computer power to solving mathematical equations.
  • MSFT: started as a company that created applications for BASIC. Then, they partner up with IBM. IBM created personal computers in 1981 and signed an agreement with MSFT to create an OS for it. The agreement gave MSFT exclusive rights to install DOS – later Windows, on these machines while allowing MSFT to also create versions of the OS compatible with other PCs. Result is market dominance by MSFT thanks to the partnership with IBM – which allowed a networked effect of developers, PC manufacturers and Users.
  • Forking money initially to create a Netowrked effect might seem like spending a dollar to earn 90cents but it can also be critical to accelerating the path towards reaching the tipping point. It’s why startups do it: sometimes the path without it will be very slow or unsuccessful.
  • “It’s a huge advantage to have a strong personal network in B2B, which you can also build by bringing a connector investor or joining an incubator such as YC.”
  • Startups have to do things manually at the beginning in a process that does not scale. That is essential though to the success of the startup. Founders might not want to to do these out of shyness and laziness. They prefer to sit and write code as opposed to talking to people to join their product when they’re going to get rejected most of the time.
  • B2B startups should build functionality to a limited set of clients first and treat that limited set of clients as if they were the company’s consultant, taking their feedback and improving the product before they generalize it to the wider client set.
  • The Gray zone: This references the situation in which a startup might find itself with respect to laws. Examples:
    • YouTube: initially there were lots of copy-righted material that were uploaded to the network
    • PayPal: Scammers took advantage of the service
    • Uber: was quickly declared as operating outside of the legal framework with their rideshare service
  • This raises some hard questions to contend with: should a startup fix it immediately and risk losing users? or should you nudge users overtime with better alignment with the legal framework?
    • Gradually, YouTube restricted copy-righted content and partnered up with some content providers
    • PayPal introduced the first captcha
    • Uber obtained licences to operate
  • All these companies chose to gradually shift gears instead of any abrupt changes
  • Dropbox convened a Revenue Growth team to deal with rising infrastructure costs that ensured as a result of massive user growth. The team targeted low-hanging fruit by redesigning some webpages and offering CTAs for customer to up their subscriptions, and also nudging them towards doing that by showing alerts that they’re about to hit their storage limit. They also mined the data and found that they could divide the userbase into two types: High-value who collaborate on documents shared through Dropbox and low-value that just share documents. This allowed Dropbox to shift its strategy from focusing on everyone to just HVA. Also, the growth team had to overcome some internal obstacles: why duplicate marketing job, why have engineers spend time on finding revenue opportunities instead of building great products?
  • Any product can acquire users through Google and Facebook ads, but viral growth occurs when users tell other people they know about the product, keeping user-acquisition costs low.
  • Mobile apps experience a high degree of churn: lots of users stop using it within a month – with a significant portion using it only a single time. The author uses the following retention metrics in determining which products have the potential to be networked products: 60% after day 1, 30% after day 7, and 15% after the first month. Any product that surpasses that has the potential to be a networked product
  • One of the effective ways of addressing churns is by dividing users by levels of engagement with the product. From there, you can look into what differentiates high value from low value users. From there, you decide how to focus on high-value users
  • SaaS companies – VC’s backed, achieve a significant milestone when they reach a billion dollars in valuation. It’s argued that the path to achieving a billion dollars in revenue should take 7-10 years and can follow the following path:
    • Establishing a great product fit – takes 1 to 3 years
    • Hit 2m in ARR
    • 6m in ARR
    • 18m in ARR
    • 36m in ARR
    • 72m in ARR
    • 144m in ARR
  • You do not have to necessarily follow the path above: you could have your own goals. Perhaps you’re happy with 500million or you’re backed by an angel investor. The equation below helps you determine the growth factor needed to achieve your goal:
    Growth factor = ((target revenue – beginning revenue)/beginning revenue)^(1/years). Years being the number of years you assign to achieve your goal.
    example: ((200-1)/1)^(1/6) = 2.4x
  • Usually when a company does this type of analysis they do have a year or two’s worth of data and they use this modeling to predict how the growth should look like.
  • Each product eventually reaches a saturation point. This is a difficult challenge to address. One way to address it is by looking to adjacent networks – whose experience with the Product is not great, to the core network that has fueled the current burst of Product growth. eBay did that – starting with US markets and then moving to the international markets.
  • Just as is the case with everything in life, it gets more difficult to achieve the same level of results by doing more of the same thing. At the advent of the internet, ads used to have a great click-through rate. That rate has declined steadily over the years – ads lost their novelty. Companies responded by allocating more marketing budget but the return on investment declines as you allocate more dollars. You need to improve network effects not spend more money
  • Uber tried to get into the business of leasing cars but lost big on that. Started in 2015, the idea is to give those drivers who don’t have a car of their own access to a car. That benefitted those who could not get a car through the regular channels – recent immigrants or those with bad credit. The interest rate charged was high and Uber did not intend to making money off this program, but losses amounted very quickly with Uber offering a generous return policy – no penalty, high interest rates, automatic deduction of lease amount from weekly earnings, and theft of cars.
  • When a network becomes large, diverse and rich, it is referred to as an “Economy” – Gig Economy, Attention Economy, etc
  • Usenet is a network that was quite popular in the 80s and preceded the GUI internet. It was the network where big announcement like the creation of WWW was made. Usenet was like Reddit with groups focused on different topics. It was not hard to find quality threads in it.
    Then in 1994, a massive user growth explosion happened. Though it resulted in the introduction of some new good features such as support for new file types, it also led to unintended consequences such as the presence of pirated files, pornography and other salacious content. This made it harder to find quality discussions, and the network will never recover from this as power users who were on the network started leaving it.
  • “When you first join a social network with your close friends, it’s easy to use it a lot. You might post photos and comments all the time—full of in-jokes and shared stories. You and your friends like it so much you invite your other friends, and then their siblings, too. And so on. But eventually, photos and content meant for your close friends might attract comments from people you don’t know well. Your parents get invited, and maybe your teachers, or your boss. Those photos of a party you went to might get you in trouble.”
  • Context Collapse is a problem that networked products eventually face and happens when initial power users are driven out of the network. Each network has its own “netiquette”: a set of norms that get established early on in the life of the network, but as new users join – lots of them and from different segments than the previous set, this can start having adverse impacts on how the power users have been operating within the network
  • “This could be Craigslist’s culture of low prices and no frills, Airbnb’s early focus for unique places to rent, or Slack’s early use by the early-adopter tech community. These are three different categories—a classifieds site, a travel marketplace, and a SaaS workplace product—and yet they face the same thing. As the network grows, the hard side is often forced to participate less.”
  • “For a marketplace product, an early community of high-end sneaker enthusiasts might grow and eventually find itself inundated with casual buyers who care more about affordability. If they don’t appreciate the products as much, or say the wrong things, this could turn off the initial sellers. On the other hand, new sellers might start to list more affordable but less cool products, making it harder for the early community of buyers to find what they want. What is an attractive product in one context might be less so to another, which is one of the reasons why context collapse can hurt the matchmaking at the core of marketplaces.
    While the experience of these users is slowly degrading as the network grows, simultaneously the product teams building the network are pushing hard, continuing its growth. At its heart, this is the tension: network effects versus anti-network effects. And when the anti-network effects become strong enough to cancel out the efforts of the team, the network hits its ceiling.”
  • One way to prevent context collapse is to do what WhatsApp, Facebook, Instagram and other social media channels have done. That is to create subgroups – thus turning the Product into a network of networks. This keeps the Product alive by allowing users to group together and share content away from the entirety of the network. It is not a silver bullet solution because you want to avoid network fragmentation and the problems that come with it – discoverability, and dead ends.
  • Spam and Trolls can degrade the quality of a network over time. One of the ways to help combat that is by empowering users to fight it themselves – reporting, flagging and downvoting features. This is a great way to address it as the network size grows. Current network sizes make it impossible for people to manage without software. Providing functionality within the Product
  • Could usenet be saved from failing? the answer is prolly yes. First, networked products like Hotmail which existed in the era of usenet still exist today. At the core of saving usenet though is the principle that it was built on an open-source network, which is both a benefit and a curse. It is a benefit because it provided grounds for growth and the ability for users to experience a degree of freedom. However, as a network grows, the need to govern the network and ensure that its continued success in providing value for users becomes more pressing. A startup made of hundreds of staff who are dedicated to ensuring the success of the network
  • Moat for non-networked companies like Coca Cola or See’s Candy might be brand or a strong business model. for Netowrked-products, it is the quality of the network. It is easy to replicate the product features of Airbnb or PayPal, but it’s a whole different game to replicate their network. Moreover, in order for any new entrants to compete with them, they will need to do more than what these companies have done to gain a foothold in a market segment. Take for example Airbnb, in order for them to establish a meaningful presence in a city, they’ll need 100 listings and 300 reviews. However, a new competitor threshold for competition in the same geo will not be a 100 listings and 300 reviews. That is because when airbnb has entered the market first, they have taken the best and most easily acquirable supply and demand.
    • The anti-network effect that plague any network will be multiplied if Airbnb is already in that market
  • Moat of a networked product does not carry over automatically to other geographies or companies if the product has to prove itself again in new geos or companies. This is the problem that Uber and Slack face: Just because Uber had a strong network effect in San Francisco, that did not extend automatically to San Diego. The company had to expend resources to build the network in San Diego. Same for Slack but think companies instead.
    Airbnb’s moat, however, was international. It catered to both the demand and supply side internationally. It did not make sense for travelers or hosts to go to a different network per location.
  • When an incumbent loses part of its network due to cherry-picking -eg: Craligslist losing its short-rental or room-share network to Airbnb, it becomes very hard to recover that network for two reasons. First, as the incumbent loses density in their network, there comes a point where the network becomes illiquid – very difficult to find demand and supply on. Second, it becomes harder to attract funding especially when the losses of the incumbent are occurring while a competitor is gaining.
  • Big companies could use a startegy to overwhelm startups when entering a new network by utilizing its vast distribution, sales and marketing, and engineering resources to create a shock and awe effect against startups who operate in that specific area. However, such factors are often the reason why networked products launches by big companies would fail: they fail to create a dense network and instead create many weak networks that won’t survive on their own.
  • Networks could compete with each other by offering a better product. However, another way to compete is by stealing the hard side from the other network. Eg: Uber ran tons of promotions to get drivers to be on the Uber network and only on it. Doing so meant that Lyft and others did not have as many drivers so their networks hit surge pricing quicker. It also attracted more customers who could get a ride faster and cheaper. Uber incentivized drivers to only use Uber by offering financial incentives when a driver has completed x trips, and that incentive increased as the number of trips increased. Uber experimented with other tactics too such as guaranteeing surge pricing after x trips. Uber utilized machine learning to identify those drivers to target with such promotions.
  • Any networked product that is trying to outperform competition needs to track metrics such as market share, user engagement, or active users while they execute in the market. This way, they can track cause and effect.
  • The competition in rideshare space shows the fallacy in the bigger takes all idea. Instead you want to think of each product as networks of networks. This is why even though Uber is much bigger in aggregate, it still had a 50/50 market share in LA and SF. This model of thinking helps explain why bigger companies have a hard time beating competitors – eg: FB and Snapchat, Zoom and competitors
  • Large companies that have big networks cannot assume automatic success when it comes to cross-sell and upsell strategies that accompany their launches of new product. While their network is huge, and while they can get users on the network to try their new products, this isn’t a guaranteed recipe to overcome the Cold Start problem. That is because in order for the new Product to stick in the existing network, it needs to be great and provides something superior to existing similar products. A couple of examples: Internet Explorer bundling with WIN 95. Earlier version of IE did not take off because they were not as good.
  • A success example of capitalizing on a large and existing network is that of Facebook and Instagram. You can sign up on Instagram using Facebook. Further, you can see comments and interactions with content on both networks. Instagram capitalized on the powerful data set that Facebook has made of the interactions that friends had with other friends and used that to create an engaged Instagram community. This went against the prevailing wisdom at the time which stated that in order to drive engagement and adoption of Instagram, you had to get celebrities and influencers to join the network. While regular users followed influencers, influencers did not follow new users. The interaction that friends have with each other in Instagram kept them coming back.
  • One of the ways Microsoft succeeded in beating competition is by bringing their entire ecosystem: developers, customers and PC manufacturers to compete on multiple levels, not just building more features.
    A case in point is creating tools for developers within Microsoft OS to create applications. This made it really easy for small businesses to build applications using Visual Basic or hire small consultancies to do so. And those applications were made compatible with future releases of Windows, even though that ended up costing Micrsofot a lot. MSFT took on that burden to save developers from having to take on the burden of updating!
  • Bundling works but it doesn’t come without its drawbacks. MSFT rellied on bundling to dominate the Internet Browser battle – at least in the late 90s to the early 2000s. When MSFT first saw Netscape and realized its potential, it cobbled together IE 1.0 which was subpar in quality. It did not achieve user stickiness. MSFT partnered up with AOL – creator of Netscape Navigator, to create a browser that was running IE in the background but had the AOL brand. This made web developers test their websites for compatibility with IE and was the linchpin for MSFT to ultimately gain dominance in the space.
    MSFT’s decision to bundle up products came at the cost of less elegant software, since the focus was always on developers.
    MSFT tried to do the same thing in the mobile space, but it failed: Google gave away Android for free and monetized the ecosystem while MSFT charged hardware developers a licensing fee.

Leave a comment