Hello, dear reader! I’m so glad that you’re interested in learning how to develop with React. This process is often long and confusing. The aim of this article is to make this process as fun and straightforward as possible. As we all like to work in a lean way, this article will be a blend of links to already existing high-quality guides for you to follow.

You will learn (and re-learn) JavaScript & React development all the way from the basics to more advanced concepts. This guide is estimated to take about two-three weeks in total. When you go through this guide, take your time and try to digest the content. I suggest that you take your time and google it or ask around if something is unclear, but be aware not to overdo it. Try to conserve your resources a bit. Of course, if you find some other material that you prefer, use that instead (or as a complement). As long as you learn what is proposed in this article you’re fine. Oh, and have fun!

Let’s go down the rabbit hole!

Prerequisites

One of the foundations of any good React developer is to be comfortable in HTML/CSS/JS. I won’t provide an SCSS/SASS/Layout guide here though, learn/re-learn as you go.

But before going into React, let’s take a look at our JavaScript knowledge. Look at the description of the video below. If there are topics in the timestamp that you are unsure of, follow along in the video before you proceed. Be sure to code along to really get that muscle memory into it. For this, you can use the terminal in your web-browser. Good luck!

React developer environment

It’s important to decorate your development environment in a way that makes it easier and more fun to develop.

The most popular code editor to use is VSCode and that’s the one we’ll use. Some good extensions to VSCode:

• Auto Close Tag
• Auto Rename Tag
• Bracket Pair Colorizer
• GitLens — Git supercharged
• npm Intellisense
• ES7 React/Redux/GraphQL/React-Native Snippets Extension
• Prettier (configure this as shown in the video below)
• Babel ES6/ES7

A tip is to use the built-in terminal in VSCode. It’s nice as the path of the terminal already is in the folder you are in. Some nice quick commands:

Mac

• Open terminal: ctrl + shift + `
• Quickly find a component or file in the editor: cmd + p
• Show/hide the left side-bar: cmd + b
• Search anything in the project: cmd + shift + f

PC

• Open terminal: Ctrl + shift + ö
• Quickly find a component or file in the editor: Ctrl+ p
• Show/hide the left side-bar: Ctrl+ b
• Search anything in the project: Ctrl + shift + f

Coding time!

You will develop a few apps that will increase in complexity, finishing off with a full-stack application that you will host online.

#1: Laying a solid foundation

The second video goes over the basics of React and will provide a good foundation for you. You’ll learn the what, why and how’s of React development.

You’ll learn:

• Single Page Application (SPA)
• Single component state handling
• Project setup

Note: In the video, he declares components with the .jsx ending instead of .js which is more common. It does not matter. He also ends lines with “;”. It is common to skip that as the build automatically adds that when compiling the files.

#2: Multiple states

By this time you know how to set up a workstation and how to initialize a project. You also know single state handling, and you be able to develop a single page application. Now you’ll learn Redux. It will take some time before you’ll really internalize how Redux works but this is normal. A tip is to have a pen and paper ready to draw out the structures, it’ll make it a lot easier to internalize.

You’ll learn:

• Multiple state handling using redux

#3: Routing

Once you’ve mastered how to initialize and modify the internal state of components its time to create a multiple page application where you will be able to navigate around several different pages.

You’ll learn:

• Multiple Page Application (MPA)
• Navigation / Routing

#4: Fullstack application

Once you know state-handling and navigation it’s time for an important step which is to load these component states with data from servers.

You’ll learn:

• Backend using node + Express
• Create, Read, Update and Delete (CRUD)

#5: Hosting

Your application is not so valuable if no one has access to it. That’s why you now will host your application online. There are many different ways of hosting applications including Firebase, AWS, Heroku, GitHub-pages and more. The easiest I know of is Netlify.

You’ll learn:

• Hosting using Netlify

#6: Further learning

These are things you’ll learn along the road so don’t sweat it if it feels overwhelming.

Function components

One new feature in React (v16.8) is to use Hooks instead of stateful components. A more resource-effective way of building components is to just use pure functions.

So you use a function as a component

const TodoList = () => <div>Hello!</div>

instead of

class TodoList extends React.Component {

  render() {
     return (<div>Hello!</div>) 
  }
}

What, why, and how is described in the link below.

• https://reactjs.org/docs/hooks-intro.html

Testing react components

Testing in React with Jest and Enzyme: An Introduction

• https://djangostars.com/blog/what-and-how-to-test-with-enzyme-and-jest-full-instruction-on-react-component-testing/
• https://www.robinwieruch.de/react-testing-jest
• https://www.robinwieruch.de/react-testing-jest-enzyme

It would mean a lot if you’d show appreciation by clapping and sharing this article if you benefited from it — and be sure to follow me for more interesting articles!

The shift into AI-powered businesses today could be as important as the digitalization of companies was 20 years ago. It’s an equally fascinating topic as it is hard to explain to someone in a clear and easy way.

My goal is that after reading this you’ll be able to have a clear understanding of how to identify use cases, as well as being able to explain it to your peers.

The one minute scoop

Machine learning is the ability for computers to learn without being explicitly programmed. It is used when you want to calculate either the likelihood or the quantity of something from the past, present or future.

You can use machine learning for predictions when other methods are not powerful enough. This can be when you have 10s, 1000s or even millions of input variables to the problem.

You will need to have anything from at a bare minimum of hundreds and even up to tens of millions of data points to be able to train your machine learning algorithm sufficiently, and this is all depending on your specific problem and it’s impossible to give a general answer about the quantity of data you’ll need.

The act of training a program that can calculate probability is called machine learning. The program that you then can use for predictions is called a predictive model.

If you want to implement machine learning in your business, you should consult with a data scientist that knows the ins and outs of both the technical aspects as well as the practicalities of business implementation.

So, what is machine learning?

“[Machine Learning is the] field of study that gives computers the ability to learn without being explicitly programmed.” — Arthur Samuel (1959)

How machine learning works are analogous with how people learn. We create an internal model of the world and use that in order to be able to make the best decisions when needed.

In machine learning, the internal model of the world is called a predictive model. This model is what’s being trained and later used to answer questions.

Machine learning is the process of learning the predictive model about the world. Now, let’s look at the different parts of machine learning.

How does it all work?

Machine learning consists of five parts. After this, we’re going to look at how they function in a real-world example.

1. Data Input

This is anything that can be expressed by zeros and ones; this could be information from databases, cameras, or microphones.

2. Data (Pre)Processing

Before the data goes into the predictive model it has to be formatted into a “computer friendly” format.

3. Predictive Models

Programmes that are generated by a large amount of historical data and used to answer questions about the past, present, and future. This is done by training it to learn the answer to a specific question you have in mind.

The predictive models generate numbers called scores. These numbers are the representation of the likelihood or quantity of something.

4. Decision Rules (Rule Sets)

The predictions generated from the predictive models are of course useless on its own. The rules that govern what to do with this new information is either set by the users or the learning algorithm itself.

For example, the decision rule could be that if a customer drops its activity by 10% during a month, you’ll automatically deploy an action to prevent customer churn.

5. Response / Output

In the example, the output from the system is the action to prevent churn, which could be an email containing a good offer for the customer.

Input Data

The type of data you want to use determine which type of machine learning you’ll use. There are two types.

Labeled Data: Supervised ML

You have a set of data that contains the “right answers” that you train your internal model against.

Consider the question: “Will a 30-year old New Yorker like this movie?”. You have 1000s of examples of 30-year old New Yorkers liking and disliking different movies. The data is labeled.

Unlabeled Data: Unsupervised ML

You have a bunch of data and the process of machine learning has the given task of finding patterns and relationships in that data. This data has no labels. You may know what you are looking for but you have no right answers to train against.

For example, you have a lot of data from your customers and you want to see if there are groupings in the data such as the highest paying customers and you want to know what their common denominator is.

Which data?

If you would like to predict which of your customers that will refer the most new customers in five years, you will need to have five years of historical data over your clients' behavior.

How much?

No one can tell you directly how much data you’ll need for your specific problem, because it is unknowable. You’ll have to discover that through empirical investigation. It is depending on (1) the complexity of the problem and (2) the complexity of the learning algorithm.

To get in the ballpark, you can for example benchmark by looking at similar cases. Here are two examples of investigating customer churn: Google Scholar or Arxiv.

Really, you should consult with your data scientist in this matter. But according to Jason Brownlee, without any knowledge of the specific problem you can get a naive guesstimate:

• you need thousands
• no less than hundreds
• ideally tens- or hundreds of thousands for average problems
• millions or tens-of-millions for “hard problems” tackled by deep learning

The Predictive Model

There are two subcategories of supervised ML that we’ll take a look at. They are classification and regression problem and these correspond to two different predictive models.

Classification: Likelihood

The classification model produces a probability score and answers the question: “What is the probability of X?”.

What is the probability that…

• …an employee will quit?
• …this customer will refer your service to a new potential customer?
• …this image is a cat, dog or mouse?
• …this customer would like to buy this product as well?

Regression: Quantity

The regression model produces a magnitude score and answers the questions: “How much?” or “How many?”.

How much/many…

• …will this customer spend in my store?
• …minutes does a phone call last on Monday mornings?

Combined Models

These two can also be combined. For example, a company wants to implement a system that automatically detects and deploys retention strategies for customers that are estimated to spend a lot of money and that are at high risk of churning. Both classification and regression models are being used.

The Algorithms

There are many different algorithms used to calculate these scores. According to Steven Finlay, the three that are the most common when calculating either probability or magnitude scores are the following:

Scorecards (Linear Models)

This is a good model that makes it easy to see the causation and correlation between different input variables. The scores are calculated in a simple and transparent way. We’ll look at an example of the scorecard model in a bit.

Decision Trees (Classification And Regression Trees/CART)

This is like the scorecard transparent, relatively easy and intuitive for non-technical people to understand. Its score is being represented by a binary tree and is easy to interpret and explain to stakeholders.

Neural Networks (NNs)

When the number of inputs is gigantic, the computational cost for the systems mentioned above is too overwhelming. How this system works is beyond the scope of this article, but check this out if you’re interested.

One thing to note is that neural networks are a bit more complex and “black box” in nature. It can be difficult to understand why it has calculated the scores and thus why a certain decision and action has been made.

Why and when use machine learning?

For one, when you have more input variables to a question than you can manage.

Let’s illustrate why eventually machine learning is needed with a probability question. The question is “will Sara like this song or not?”.

Let’s start plotting out some historical data and see where they fall in the interval with only one input variable; tempo.

We can clearly see two distinct groups here. She likes music with a high tempo and dislikes music with a relaxed tempo. This information can be used to evaluate any song that has the variable tempo available for us.

If we want to know whether she likes some song we can just see if it falls under the like or dislike interval. Easy enough.

How about two variables? Let’s add another input variable of happiness.

Now we have an area that we can use in the same way as above. She seems to like music that is happy and fast and dislikes music that is sad and relaxed.

Three variables? Let's add time period.

Same thing. We get a nice volume that we’ll use the same way. She seems to like happy, upbeat new music.

But what about 10s, 1000s or millions of input variables? Well, not quite as easy and in comes machine learning to the rescue.

More Benefits

• Fast. When implemented in an automated decision-making system, a predictive model can be used to predict the behavior of millions of individuals in seconds which would be near impossible if done manually.
• Cheap. Once a model is developed, it is usually cheaper than its human counterpart.
• More accurate than human experts. In many fields predictive models outperform human experts by 20–30%; i.e they make 20–30% fewer errors or identify 20–30% better (more important/more profitable) cases.
• Unbiased. Unlike people, predictive models do not display deliberate harmful bias based on their gender, race, ethnicity etc. (assuming that the model has been trained correctly).

Example: Reducing Heart Disease

To see how the process of machine learning works in real life lets take a look at an example using the scorecard model.

Problem Statement

Heart disease is one of the leading causes of death worldwide. We want to identify the people who are likely to develop heart disease in the next five years. Then we want to take preventive action to decrease that likelihood.

If we were to treat everyone that was in risk, it would take up too many resources. So we’ll infer a rule that says that only the top 5% of the population that is at the most risk will be taken preventive action upon.

The question is: “Will this person develop heart disease in the next five years?”

We have access to hundreds of thousands of medical records with thousands of input variables and machine learning is perfect for this task. We need probability scores and thus build a predictive model of the classification model type.

Data Input

We’we got labeled data and will thus use supervised ML. We’ll use what is called a development sample that is used to train the prediction model. This contains two parts:

• Observation data: data from the start of the five year period.
• Outcome data: data of their health over the forecast horizon. This contains the answer if they developed heart disease or not.

For the observation data, we are using the medical records of 500,000 people chosen at random that didn’t have any signs of heart disease five years ago. This data will include, amongst other things, things like age, gender, BMI, blood pressure readings, exercise amount, alcohol consumption and so on.

The outcome data is the data from the next subsequent five-year forecast horizon.

The outcome data tells us that 6% (30,000 people) went on to develop heart disease and 94% (470,000) did not. With this data, we will train the model which patients who will develop heart disease as well as the ones that will not.

Let’s see what our predictive model have come up with. We’ll just show some of the 1000s of input variables for obvious reasons.

The Scorecard Results

The starting score is 350, and we’ll add and subtract from that value depending on the variables impact. Each box represents an input variable and its predictions.

When a variable shows a positive correlation to heart disease, it adds a positive score. This means that it is more likely that the person gets the disease. The same holds for the negative score.

Green numbers: lower likelihood of heart disease. Red numbers: higher likelihood of heart disease.

For example, when looking at the scorecard we can see that if a person is below the age of 23 there is a relatively big negative score indicating that being under the age of 23 decreases the risk of getting heart disease significantly.

We can also see that being over 71 years of age is the biggest factor to getting heart disease followed by being a smoker.

But we would like to have some statistics generated also. This is provided by a score distribution table.

The Score Distribution Table

By using this score distribution table it can be shown that the top 5% of people that are in the most risk of getting heart disease will score 521 or more.

Therefore, the cut-off strategy (decision rule) to apply is that anyone scoring 521 or more should be invited for a check-up with their doctor.

Practical Business Implementation

There are different companies that offer Machine Learning SaaS, for example, AWS Machine Leaning, which I encourage you to check out. They offer both different models that you can build, train and deploy comparatively easy to building everything from scratch.

They also offer ML services that need no knowledge in the field, that you can implement straight away with no training. This includes; customer recommendations, forecasting, image and video analysis, advanced text analytics, document analysis, voice, conversational agents, translation and transcription.

Get a data scientist

And get one that both have a foot in the deep theoretical aspects and, equally important, one that knows how to draw the connection from that to the practical implementations in the business world and all the intricacies that come with that.

If a data scientist can’t explain to you what they’ve done and why then they have failed at their task.

Thank you for reading

It would mean a lot if you’d show appreciation by clapping and sharing this article if you liked it — and be sure to follow me for more interesting articles!

For further reading, I would invite you to take a look at “Artificial Intelligence and Machine Learning for Business” by Steven Finlay.

Notes and takeaways

While scrolling my Twitter feed I saw that Y Combinator promoted Reid Hoffman’s new book about something that was called Blitzscaling. I am a sucker for books with catchy titles so I immediately ordered the book.

Who should read this book and why? If you are an entrepreneur, investor or just interested in the subject — this book will give you a deep insight in the latest strategies and techniques about how to go from idea to start-up to scale-up and finally how to reach the top as a market leader.

Let’s go!

”Of all the things I’ve discussed with Reid, the most thought-provoking might be blitzscaling, […] Reid and Chris’s ideas are more practical than ever, because it is now possible to get big fast in a way that simply wasn’t feasible a few decades ago, […] the tools you’re about to gain have never been more relevant. This is an ideal moment to be reading this book. I’m glad Reid and Chris are sharing their insights.”

— Bill Gates

So tell me, what is blitzscaling?

Blitzscaling is a strategy and set of techniques for driving and managing extremely rapid growth that prioritize speed over efficiency in an environment of uncertainty.

But almost every start-up wants to scale fast, exactly how is this different?

Regular fast scaling is using the classical approach in business strategy that says that the risks taken shall be calculated ones that you can both measure and afford. This technique will inherently prioritize correctness and efficiency over speed. We’ll soon go into the details and clarify this difference further.

So why is this important?

Efficiency and certainty are very important for stable and established markets. But in the context of new markets encountered by the inventors, innovators and disruptors of the world this strategy is suboptimal. When a market is up for grabs, the real risk isn’t inefficiency, it is playing it too safe.

As you all know, first prize is a Cadillac Eldorado. Anyone wanna see second prize? Second prize is a set of steak knives. Third prize is you’re fired. Get the picture?”

— Blake (Salesman from the movie Glengarry Glen Ross)

First prize in the first wave of consumer social networking went to Facebook; second prize to MySpace; third prize to Friendster. You need to win first prize in order to survive in the Internet era.

This is especially true for businesses that cater to two user groups that benefit from the network effect — for example, Airbnb who both needs hosts as well as guests.

The reward might outweigh the risks

It is not to be forgotten that this strategy involves a substantially higher degree of risk. However, this risk fades in the presence of the even higher risk of being too slow.

This is not the same as getting big quick. To mitigate the increased risk, one should employ certain techniques, and this is what blitzscaling is about.

The ingredients of blitzscaling

Besides courage and skill from the entrepreneurs, blitzscaling requires two essential ingredients — human capital and financial capital, and an environment that is willing to supply this.

You can think of them as fuel and oxygen; you need both of them to propel the rocket skyward. The structure of the rocket is your organization’s infrastructure which is what you are constantly rebuilding as you rise.

The role of the leader is to provide a continuous supply of fuel and oxygen while also making necessary mechanical adjustments while it is accelerating to keep the rocket from plummeting to the ground.

So blitzscaling is a certain strategy and a set of techniques to promote massive growth, cool. But exactly when do I use it and how does this relate to my business?

We’ll get into this more in depth later on but let’s still your curiosity a bit for the time being. The only time that it makes sense to blitzscale is when speed is the critical strategy to achieve success. And also, these three following conditions must be met:

1. The product/market fit is right
2. You have a working business model
3. The market conditions are right for hypergrowth

Let’s compare blitzscaling to three other types of scaling

Classic Start-up Growth — prioritize efficiency in the face of uncertainty. Gain as much information as possible to increase certainty about what Eric Ries and Steve Blank call product/market fit while using as little resources as possible.

Classic Scale-up Growth — prioritize efficiency in an environment of certainty. This is a good strategy to maximize the return on investment in a stable market with little unknowns.

Fastscaling — prioritize speed in an environment of certainty. The risks and costs are well understood and predictable because of its stable environment. One can rather comfortably sacrifice efficiency for the sake of scale. Good for gaining market share and to achieve revenue milestones.

Blitzscaling — prioritize speed in the face of uncertainty. It is “do or die”. Achieve either death or success in a remarkably short period of time by either establishing oneself as the market leader or fail.

There are three basics of blitzscaling

1. Blitzscaling is both an offensive strategy and a defensive strategy

We’ll start with the offense. First, you can take the market by surprise and bypass heavily defended niches. An example of this would be Slacks explosive growth that blindsided competitors like Microsoft and Salesforce.

Second, you can build a long-term competitive advantage. Third, in the position of being a fast-growing market leader, securing more backing from investors that you will use to boost your growth ever more and leaving your lagging competitors behind your smoke.

On the defense — your competitors will have trouble keeping up with your pace. Often they will have to try to play catch-up with you.

2. Blitzscaling thrives on positive feedback loops

The company that grows to scale first reaps significant competitive advantages. You attract what you are, so when you soar to the top and emits success, you will attract success. Success in the form of capital and the best talent.

3. Despite (because of) its payoffs, it also comes with massive risks

The term for out of control growth in the human body is called cancer. The same is true for business. You want your organization to grow in lockstep with your revenue and customer base. Otherwise, different parts of your organization will lag behind.

The five different stages of blitzscaling

When a startup grows, it does not equal multiplying your company with an ever increasing number. What got you to one step won’t get you to the next step in terms of strategies and techniques.

There are five different stages in a company’s growth (stated below). Each and one of these steps has their own management and leadership.

We will use the different names throughout the article to reference the specific stage of a company.

The three key techniques of blitzscaling

Technique #1: Business Model Innovation

Despite the common belief of the nerds of Silicon Valley that uses their engineering skills to build amazing products and bases a company around that — the truth is that the people behind these companies are almost always business nerds too.

As the world has progressed technology-wise, the technology behind companies is not a differentiator anymore. The way that start-ups may compete against established competitors with more fuel and oxygen to their rockets is just that, business model innovation.

Dropbox founder and CEO Drew Houston state that you can not rely on better technology or outexecuting your competitors. He goes on and says that:

“If your playbook is the same as your competitor’s, you are in trouble, because chances are that they are just going to run your playbook with a lot more resources!”

Technique #2: Strategy Innovation

Obviously, when blitzscaling, the goal is extreme growth. Most start-ups have the goal and the wish to achieve this but lack the understanding of how to actually implement the right strategy for this cause.

A definition of success when blitzscaling would be to achieve a competitive advantage by using the network effects to scale the business to a critical point that triggers a positive feedback loop that allows it to dominate a winner-take-all or winner-take-most market and gets a long-lasting first-scaler advantage.

An example of a strategy that prioritizes speed over efficiency is Uber’s aggressive city-to-city expansion strategy. Uber needed to get to a critical mass of users quickly to really be a valuable service for the citizens of the new city.

To achieve this extreme growth they provide heavy subsidies to both its drivers and riders. They boosted the drivers’ salaries and lowered the price substantially for its riders.

Another example is what Disneyland could have done. They could have lowered its ticket prices with 90% and predict that through network effects one million visitors would attract ten million visitors and thus in a later stage making up for the initial losses.

A reason for why silicon valley investors are so successful is their relentless strategy of extreme growth and their willingness to take big risks.

If a company does not display the wanted hockey stick growth curve they will conclude that they will either sell the company or invest more money to either reach the top of its market — or end in a big failure. A warning sign for them would be if a company displays anything less than a 40% annual growth.

Technique #3: Management Innovation

The traditional strategies that are working for a company that happily grows by 15% annually do not work in the fast-growing blitzscaling companies. And it is not comfortable to triple the number of employees each year.

This requires radically new strategies that include being comfortable with hiring good enough people, letting fires burn, ignoring angry customers and launching flawed and imperfect products.

These were the three parts that together create the high-risk high-reward situation that you will find yourself in while blitzscaling. Now let’s dive deeper into each subject.

Business Model Innovation

This is the first and most foundational technique when building a business capable of growth; to be able to build an innovative business model.

This becomes clear when looking at the dot-com era around the 2000’s when many companies became digitalized.

Why didn’t Walmart dominate e-commerce instead of Amazon? Why didn’t the existing newspaper and phone book companies dominate instead of Google? The answer is the title of this paragraph.

The startups of that era that relied on technology innovation without any real business model innovation went bust. The real value creating and what is needed for a successful startup is to combine innovative technology and innovative products or services with innovative business models.

Great business models often seem like a bad idea at the time, because inherently in its very definition it cannot point to any existing business model that has been proven.

So when designing your business model you should maximize the following four key growth factors: market size, distribution, high gross margins, network effects.

You should at the same time minimize the two key growth limiters: lack of product/market fit and operational scalability.

A brief overview of the growth factors and growth limiters

Let’s start by looking at the growth factors before moving on the growth limiters.

Growth factor #1: Market size

A big and/or growing market has both a large number of potential customers and a variety of efficient channels for reaching these customers. The last part here is important, and we’ll look a distribution later.

One of the main sources of uncertainty of blitzscaling is predicting the TAM (Total Available Market) but to get it right and invest accordingly is one of the main opportunities of high returns as in the cases Uber or Airbnb.

When raising capital from a VC-firm one should bear in mind that the investors of Silicon Valley are looking to triple their investment over the typical seven- to ten-year life of a fund to achieve an above-market of internal rate of return of 15 to 22 percent.

So if you’ve raised $50 million the market must be able to return you $150 million during this time frame.

Growth factor #2: Distribution

Building an “insanely great” product as Steve Jobs would say is not enough. If we would compare a great product with poor distribution and a good product with great distribution, the latter would win. As Drew Houston, CEO Dropbox said:

”Most of the orthodoxy in Silicon Valley is about building a good product. I think that’s because most companies in the Valley don’t survive beyond the building-the-product phase.

You have to be good at building a product, then you have to be just as good at getting users, then you have to be just as good at building a business model. If you’re missing any of the links in the chain, the whole chain is broken.”

The two main distribution techniques fall under two main categories: leveraging existing networks and virality.

Leveraging existing networks

New companies seldom have the resources to pour money into advertising campaigns so they have to be creative.

One way is to piggyback on existing networks. An example of this is how PayPal offered their payment service on eBays platform and thus tapping into its big pool of existing users.

Using virality

Virality happens when one user brings in another user and so on. It can be organic or incentivized by some kind of reward.

Once again looking a PayPal, they got organic virality in the normal use of the product — one had to create its own PayPal account when receiving a payment from another PayPal user.

They also used incentivized virality. If you referred a friend you got $10 and your friend got $10. This kind of monetary incentive will decrease as you grow. PayPal decreased their incentive from $10 to $5 and eventually removed this reward.

Virality almost always requires that the product or service is either free or freemium, otherwise, the threshold for signing up would be too high.

Growth factor #3: High Gross Margins

Most of the biggest tech companies have a gross margin of 60, 70 or even 80%. This allows for a large asset in cash to use directly. An important difference is the difference between potential gross margin and realized gross margin.

Many blitzscalers maximize the potential gross margin and minimize realized gross margin to maximize market share.

For example, Xiami aims to have a realized gross margin of about 1–3 percent by using the resources on growth. As Jeff Bezos said: “Your margin is my opportunity”.

Growth factor #4: Network Effects

This last factor plays the important role in sustaining the growth long enough to build a massively valuable lasting franchise.

We’ll use the following definition of network effects: ”A product or service is subject to positive network effects when increased usage by any user increases the value of the product or service for other users.”

There are five different types of network effects as seen below.

Direct Network Effects. Increased usage increases the value of the service. (Facebook, WeChat, WhatsApp)

Indirect Network Effects. Increased usage encourage increases consumption of complementary goods. (Microsoft Windows, iOS, Android)

Two-Sided Network Effects. Increased usage from one user groups benefits the other and vice versa. (Marketplaces, Uber, Airbnb)

Local Network Effects. Instead of getting benefits from a general increase of usage one experiences benefits from the increased usage of a small subset of users. (A part of a forum that you are especially are interested in)

Compatibility and Standards. The use of a product creates an incentive to use complementary products within the same ecosystem. (Apple, iOS, Mac)

With network effect businesses you have to acclaim a critical mass of users for the service to provide its intended value for its users. This is why distribution is so incredibly important for success.

If looking at it from an economic standpoint you can say that the demand curve needs to intersect the supply curve, which is called the “tipping point”.

In Geoffrey Moore’s Crossing the Chasm, he recommends that startups focus on finding niche beachhead markets and then expanding out using the “bowling pin” strategy.

Now let’s move on the growth limiters.

Growth limiter #1: Lack of product/market fit

Marc Andreessen argues that this is the most important factors for successful startups and defines the product/market fit as:

“Product/market fit means being in a good market with a product that can satisfy that market”

A good strategy to follow here is Lean Startup made famous by Eric Ries. And a key question a new startup needs to answer is the following:

“have you discovered a nonobvious market opportunity where you have a unique advantage or approach, and one that compelling competitors won’t see until you’ve had a chance to build a healthy lead?”

And how does one find such opportunities? There are a couple of things to look for and are connected with a change in the market that incumbents aren’t willing or able to adapt to. This can be:

• Disruptive innovative technology
• Change in laws or financial regulations
• The rise of a new group of customers
• Any other major shift in a market or industry

A good article exploring these kinds of opportunities further is this one.

Proven business model patterns

• Bits instead of atoms: easier to scale and distribute.
• Platforms: can reap the benefits of compatibility and standards (network effects).
• Free/freemium: play an important role in starting network effects.
• Marketplaces: powerful because it often taps into two-sided market effects.
• Subscriptions: predictable revenue streams making it more predictable and enables more aggressive long-term investments.
• Digital goods: often nearly 100% gross margins.
• Feeds: drives ad revenue and long-term retention.

Strategy Innovation

Below are some of the strategies mentioned in the book that are easily digested. I thoroughly recommend you to read the book to the get the whole picture if you find this interesting.

There is a big new opportunity

To achieve massive success you need a big new opportunity where there is a big enough market and where there isn’t a big market leader or oligopoly.

This is often the case where there is new innovative technology that creates a new market or scrambles an existing one.

You have competition

Ask yourself: “can someone else realize this opportunity before me?”. If the answer is yes, the risk of moving fast diminishes before the risk of being too slow.

The more intense the competition, the faster you should move. If there are no advantages to moving faster, it is better to follow the traditional strategies for the time being.

So tell me more about how blitzscaling differentiates from regular fast scaling?

The (obvious) fact is that every startup is trying to achieve the fastest possible growth.

For the two first stages, Family and Tribe (between 1 and 100 employees), it can be a challenge to move faster than your competitors. But it can be done and There are only three ways to do this:

1. You might be the only competent player in the market. (Could be hard when good markets attract competent people)
2. Be the first in your market that figures out a brilliant growth strategy (such as PayPal did)
3. Pursuing scale more aggressively. (Raising more funds, hiring more people, anticipate growth and build infrastructure that supports that growth in the future. In other words, bet all your chips on this bet.)

Management Innovation

This is the third and last of the three key techniques of blitzscaling.

There are different people in different stages

The Marines take the beach, the army takes the country and the police govern the country.

The start-up people (Marines) is the people that are used to chaos and improvisation. Once the beach is taken, the scale-up people (army soldiers) knows how to quickly take territory. Last in are the stability people (police force) that maintain rather than disrupts.

Generalists to specialists

For a long period in a startup, employees need to be able to wear many hats and be flexible. You should only hire generalists in the Family stage (1–10 employees). In the Tribe stage (10s of employees) you might on rare occasions hire a specialist in machine learning or a data scientist.

However, in the Village stage (100s of employees) it is wise to start focusing on hiring more specialists. In the City or Nation stages, almost every new hire will be a specialist.

Data-driven decisions

”If this is a decision based on opinions, then my opinion wins. However, data beats opinion. So bring data.”

— Jeff Bezos, CEO Amazon

In the stages of Family and Tribe, it is hard to make data-driven decisions when you might not have that many customers to analyze. In that case, it is best to use intuition (knowledge and experience).

However, a good practice is to start collecting data as soon as possible. So what kind of data to collect?

First of all, Selina Tobaccowala from SurveyMonkey says that she strongly believes that a company as a whole cannot get behind more than three to five metrics.

The main metrics to track include the number of users, raw engagement and churn. And according to Hacking Growth, you should pick a North Star metric that will help decision making easier to keep you aligned with your goals and to track progress. For example, at Google, their North Star was the number of hours watched each day.

Some fires will have to keep on burning

“What you say ‘no’ to is more important than what you say ‘yes’ to”
 — Joseph Ansanelli from Greylock Partners

To decide which fires you will let burn you will need to prioritize by urgency, efficiency, and dependency. First ask yourself: “which fire is going to damage or kill my business the soonest?”.

Regarding efficiency, you will need to think about which fires you have the ability to extinguish right now and which ones a better to deal with later.

And lastly dependancy. Will extinguishing Fire A make it easier to extinguish Fire B or C?

The list below is the Maslovian hierarchy of fires that start with the most important ones in the top.

1. Distribution
2. Product
3. Revenue model
4. Operations
5. Competition
6. What to do next

Throwaway work

This is a concept that feels somewhat wrong for an engineer like me. I am schooled to build things the right way in the beginning so that we only need to build it once.

But when speed is the prioritization this does no longer apply.

Do things that don’t scale and then throw away and iterate. Invest fewer resources in security, write code that doesn’t scale and wait for things to start breaking before you implement QA processes.

Raise too much money!

The only thing that is predictable in this world of uncertainty is that unforeseen events will occur.

It is then better to have the cash at hand than to have to scramble to find more funds when inevitably something happens that requires cash to solve.

To have more cash at hand outweighs the negative aspect of equity dilution. As Mariam Naficy, CEO of Minted said:

“[…] Both of us knows so many people who had good ideas and were on the right track, but just ran out of money.”

The money you raise tends to go toward to main factors: people cost and the cost of outbound customer acquisition. In Silicon Valley, it takes about six months to raise new capital.

That is why the rule of thumb is to raise enough cash for about 18–24 months. This period enables you to show enough progress to convince your future investors that they should invest in you in the next round.

The never-ending need for change should fill you with fear and hope

Fear that you never can stand still and rest. Hope, because new markets are always emerging and that everyone from Silicon Valley to Shanghai all has the ability to build something from nothing that is nothing less than a rocket ship.

Thank you for sticking it out until the end. If you enjoyed reading this be sure to follow me to receive my future articles!

A special thanks to Marcus Mazouch for proofreading.