A Short-Story on the Factors of Why Start-Ups Fail
People who succeed to an incredible high degree by standards of our society, whether as a highly regarded scientist, a superstar in sports, a comedian, an actor or a person of exceptional public interest, they share at least one thing in common and that is failure. In basketball, one such individual is Michael Jordan and a quote of his goes like this:
I’ve missed more than 9,000 shots in my career. I’ve lost almost 300 games. Twenty six times I’ve been trusted to take the game winning shot and missed. I’ve failed over and over and over again in my life. And that is why I succeed. — Michael Jordan
Lessons are learned from failure, which prepares you to do better in the future by actively trying to avoid failure and actively trying to not make the same mistakes twice. This notion of failure is what motivated me to make this post and my curiosity drove me to ask the question: What makes a company successful? A rather passive and simple answer is: The lack of collateral mistakes. Mistakes carry different weights. Some mistakes are almost harmless, easily compensated, don’t have immediate consequences and, very importantly, they are reversible. Some other mistakes, come with collateral damage and cannot be unmade. It is the latter ones that need to be avoided at any cost, if possible. Thus, in the spirit of failure, I decided to dive into the factors that make a startup and their founders fail on a grand scheme, and as it turns out there are quite a few lessons to be learned.
In general, startups can run either profitable or non-profitable. In the long run, a non-profitable startup will close its operations once sufficient funding is lacking and future funding can’t be secured. In this post we will look at two different data analysis. The first part constitutes a word-content analysis from obituaries composed by founders, journalists and investors. The second part is an exploratory data analysis of companies that have closed their operation.
Part I: Word-content analysis
The word-content analysis contains 311 obituaries from startups which have failed in the past 5 years. Summarized final statements of why companies fail can be viewed at cbinsights with full stories in links thereof. I highly recommend that you take a few minutes and dive into this insightful treasure of gold.
A wordcloud of the most frequently used words in the obituaries is compiled in figure 1. The bigger the font the more often the word occurs. Time, product, funding and market stand out as the most often occurring terms.
Word counts of obituary statements are shown in figure 2 after common and unrelated words like “a”, “the”, “I”, etc. have been removed. Surprisingly, words with negative connotations, like problems (18 counts) and challenges (28 counts) occur considerably less often then time (78 counts), product (70 counts), funding (61 counts) and market (44 counts). In particular the term bankrupt (only 5 counts) was almost absent in the startups final statements, most likely due to its strong negative meaning. After all, it is easier to give a passive statement with a negative outcome and blame external factors, e.g. to say we ran out of funding and/or time, or that the market didn’t respond well to the product (funding, market and time are all external factors here), rather then to make an active statement and communicate that we went bankrupt because we were not profitable with our product (the product is an internal factor which the company directly shapes). An active statement requires more courage because it brings more risk to take on the blame for failure. This is further supported by the fact that the word ‘unfortunately’ occurs 34 times, an expression that communicates regret and quite often puts the blame on factors that are outside of one’s own’s reach.
As simple as it sounds, but the takeaway message here is that as long as the company has a product or service that is profitable it will have money and therefore time to keep operating. Likewise, if the market response to the product or service a startup offers is troublesome, then it can shorten the lifetime of a company and drive it into bankruptcy if there is not enough funding to keep the company alive. The grouped circular flow diagram in figure 3 demonstrates how the above word count can be reduced to a total of just four crucial elements: Funding, time, product, and market.
The total count of each of these crucial elements versus the rest of the word counts is shown in the pie chart in figure 4. Funding out weights time, product, and market by a factor of around 2–3 and all four crucial factors account for around 50 percent of the word counts. The ‘Others’ grouping includes all the word counts other then the ones shown in the above flowchart (figure 3).
We conclude that trouble with funding/cash flow is the most important factor mentioned (~22%) when it comes to a startup failing, according to the obituaries investigated. The startup runs out of money and cannot secure further funding for the future and needs to shut down its operation. For example, low-cost airline Wow Air closed in March after being unable to secure additional funding within a given time-frame.
“We have run out of time and have unfortunately not been able to secure funding for the company,” Chairman Skuli Mogensen said in a letter to employees.
Overall it is a mix of funding, time, product and market response that leads to a startups failure. The factors are not independent of each other. They sensitively depend on each other. For example, a negative market response to a product placement or service will reduce the cash flow and shorten the lifetime of a company. Less time available directly translates into less time for development and for adjusting the product according to customer requirements. Time and market contribute on a similar scale (around 8–9% of word counts) to a startups failure while the product contributes slightly above with ~11%.
Another example, provided by Mike Krupit, former CEO of IntroNet, whose service for professionals to make and track introductions, ultimately shut down because the service provided did not address the proper need of the customers.
“On the surface, the business didn’t succeed in the first two iterations of IntroNet for the same reason that 90% of tech startups fail: we did not find a product-market fit before the end of our cash. It’s a math equation that is pretty deterministic…”, Mike Krupit.
In my humble opinion, this might be one of the hardest parts when starting a business: ensuring that a product or service meets the customer’s demand. Predictions can be made on the success of a product or service but ultimately the customers will decide and respond whether they like or need a product at all. Several cycles of product iteration, time, money and customer feedback might be required before a product or service are placed properly within the boundaries of what customers demand and require.
Many more valuable lessons from entrepreneurs who have not succeeded can be gained from cbinsights compilation of startup failure and again, I highly recommend that you take a few minutes and dive into this insightful treasure of gold. The easiest way to prevent mistakes from happening is to make them once and learn from them, but not everyone can afford the time and money to start a company and learn all the lessons to be learned in order to succeed in the future. Carefully listening or in this case reading and extracting proper information might prevent failure or at least help to plan better for the challenges ahead. This will in particular be helpful to first-time entrepreneurs when starting their exciting endeavor of becoming an entrepreneur.
Part II: Explorative data analysis
In the second part of this post we will take a look at an exploratory data analysis of startups which have failed (negative outcome). The data is compiled and obtained from CrunchBase and contains companies from the past 20 years which have a founding date and a closing date (right half of the Venn diagram in figure 5). More technical information can be found in the methodology section at the bottom of this post. Furthermore the data is divided into companies that got closed without being acquired, which we will refer to as true negative (TN) outcome and companies that were acquired and closed (false negative (FN) outcome). The term false negative is being used since the startup did not fail. After all, the startup got acquired and compensated for the value it created. It can not be unambiguously deduced whether the acquired startup went bankrupt or whether it got absorbed by the acquiring company. Furthermore, this dataset does not contain information about companies which are still operating and successful (left half of the Venn diagram in figure 5), since we are only interested in the companies that failed. For an exploratory data analysis including successful startups you can visit Susan Li’s blog post, where she mainly focuses on successful startups.
The pie chart in figure 6 shows true negatives and false negatives of our whole dataset. Over a 20 year span only 7.2% (618 startups) out of 8586 startups have been ‘acquired & closed’ while the rest closed without compensation (‘not acquired & closed’). Again, I want to point out that this only represents all the companies that CrunchBase labeled with a starting and closing date. The overall number of companies in CrunchBase’s dataset that closed is 31686, but only 8586 have starting and closing date provided.
Next, we take a look at the number of companies that got ‘acquired & closed’ and ‘not acquired & closed’, grouped by the amount of total funding at that point of time. This can be seen in the top part of figure 7. The total funding amount is plotted on a logarithmic axis (Note: the bin width increases on a logarithmic scale with increasing funding) and a top axis is included for better readability. Also notice that the median deviates between TN and FN! For ‘not acquired & closed’ companies the median lies around $1 million while for ‘acquired & closed’ companies that value is higher by a factor of roughly 10 and it is situated at around $10 million. That is a significant difference and indicator that a certain acquisition threshold exists that lies above $1 million.
Another way of communicating this insight is by plotting the acquisition percentage of closed companies (ratio of FN/TN), shown in the bottom part of figure 7. The higher the total funding amount of a startup the higher the chances of being acquired (and at some point in future of closing). The chance of being acquired peaks around $100 million with a ~27% chance of being acquired. Above $300 million no acquisition were made and most probably that is due data bias in the data obtained from CrunchBase. CrunchBase does not have all the data available and as a result the acquisition percentage tail after $300 million is cut off. There are plenty of acquisition made above the $1 billion range. Wikipedia offers an astounding list of largest mergers and acquisitions as well as deals which have failed to complete worth several tenths of billions of dollars, even in the hundredths billion dollar range. Now, the lack of data in that range in CrunchBase’s dataset might be due to CrunchBase not categorizing a merger as either an acquisition or a closure since a new company is formed during a merger, thus resulting in bias in the data. It is also possible that above such a high funding amount the companies don’t close their operation after they are acquired because they are successful which is the more probable explanation.
Figure 8 shows the data of companies grouped by categories in which they operate in. While software and internet services have the overall highest number of startups in terms of being ‘acquired and closing’ and of being ‘ not acquired and closing’, they do not exceed in respective acquisition percentages. Surprisingly, privacy and security lead the acquisition percentage of ‘acquired and closed’ with 13.6%, while biotechnology has one of the lowest percentages with 3.9%. This is an indicator that once a biotechnology startup is acquired that it is less likely going to be closed.
The averaged acquisition percentage of closed companies over all categories is 7.3%, close enough to the value from figure 6. The amount of category counts is 26563 (all counts in figure 8 summed up), which is higher then total amount of companies in our dataset because some companies operate in multiple categories. For instance, Netflix would be considered within the Internet Service and Media and Entertainment categories, if not more.
Last but not least, we will look at the lifetime of a startup before it closes or gets acquired, shown in figure 9. Most startups close operations within 2–3 years, or get ‘acquired & close’ within 3–4 years, which is similar to Susan Li’s analysis. The bottom part of figure 9 answers the question: What are the chances being ‘acquired and having to close’, given that startup has already survived e.g. for 15 years? The answer is around 22%. The chances of being ‘acquired & closed’ go steadily up and hit a first maximum around 17 years of lifetime and slowly decay with an outlier around 22 years.
Figure 10 shows the accumulated percentages (accumulated starting from 0) and answers a slightly different questions: Given that a company has just been founded, what are the chances of being ‘acquired & having to close’ within a certain number of years? That value converges to around 7.3% on the long run, connecting back to where we started this data analysis in figure 6.
In part I, we have performed a word-sentiment analysis from companies' obituaries and deduced 4 main factors which are responsible for a companies failure. They are: funding, time, product and market. We have visualized and discussed how those factors sensitively depend on each other and assigned percentage weights to each one of them. Funding is twice to thrice as important for failing as compared to any of the other factors.
Part II contains an exploratory data analysis on 8586 startups which got closed in the past 20 years. 618 were closed but acquired prior to being closed. Data is presented for closed companies grouped by the total funding amount and categories in which they operate in. The median of a closed companies total funding lies around $1 million and categories are dominated by closed companies in the software and internet services. The largest number of companies closes within year two to three, while chances of getting acquired (and having to close thereafter) are increasing.
Data for part I was obtained and processed from cbsinsights. Data for part II was obtained from CrunchBase on a free one week trial account, since the free version does not allow for sufficient companies to be displayed. Upon contacting that data support CrunchBase allows to directly use their API but restricts data access to the year 2013. In their license agreement Crunchbase allows to use the data for non-profit purposes as was done in this study. Initially ,a selenium webscrapper was set up and used to obtain data from CrunchBase but Crunchbase prohibits and shuts down the use of an automated webscrapping tool on their webpage fairly quickly. Finally, the data was obtained by filtering according to a year by year basis of companies that have failed/were shut down in that respective year. Up to 50 companies were displayed per page (before clicking to the next page) and the information was copied to a textfile by hand. A total of over 8500 data points was collected. Texfiles with the data points were further cleaned and processed within a Jupyter Notebook environment. These and other script generating plots can be viewed and used for non-commercial purposes on GitHub.