The 80/20 Rule of Startup Founders


Photo by dierken

Note: Some readers disagree with both the mathematics of this post, as well as its conclusion that “many try, but few succeed.” Please keep that in mind as you read.

Our best estimates put the number of developers in the world at around 5 million.

If you appply the 80/20 rule you can estimate:

  • Of those, 20% (1 million) want to launch a startup
  • Of those, 20% (200,000) have enough motivation to start educating themselves about the process
  • Of those, 20% (40,000) will actually start building something
  • Of those, 20% (8,000) will actually finish building something
  • Of those, 20% (1,600) have prepared themselves enough to achieve some measure of success

To make it into the last group you have to make it through the four above it. What have you done today to move yourself closer to the last group?

Start Small, Get Big
Growth Secrets for Self-Funded Startups. It'll Change Your Life.
What you get for signing up:
  • A 170-page ebook collecting my best startup articles from the past 5 years
  • Previously unpublished startup-related screencasts
  • Exclusive revenue-growing techniques I don't publish on this blog
"The ideas and information Rob provides should be required reading for anyone that wants to create a successful business on the web." ~ Jeff Lewis
Startups for the Rest of Us...
If you're trying to grow your startup you've come to the right place. I'm a serial web entrepreneur here to share what I've learned in my 11 years as a self-funded startup founder. Luckily several thousand people have decided to stick around and join the conversation.

For more on why you should read this blog, go here.

17 comments ↓

#1 Roger on 02.23.11 at 9:26 am

The spirit of the post is obviously motivational, but one tongue-in-cheek comment : If the scope is “developers”, points 3 and 4 should be before point 2. Purely anecdotal, but I know many more developers willing to start/finish building something without knowing what else goes into creating a startup.

Rob Reply:

Indeed. Good point; market should come before building an app but it rarely does with developers.

Susan Jones Reply:

Roger, that is so true. I see so many questions from developers who are like “I’ve built an app so now what?” My big passion is educating people to look at the business case and evaluate the opportunity BEFORE they start building so they can invest their time and energy knowing they are building something people will want AND pay for!

#2 Karl on 02.23.11 at 9:35 am

Today, I’ve created a deployment script to automate changes to my wordpress site. This site is intended to be a platform for my projects and a way to share information with people in my niche.

It’s interesting how the 80/20 rule follows a similar pattern to a sales funnel.

#3 Tweets that mention The 80/20 Rule Applied to Technical Startup Founders | Software by Rob -- Topsy.com on 02.23.11 at 9:43 am

[…] This post was mentioned on Twitter by Aldiantoro Nugroho and ztartup, Rob Walling. Rob Walling said: The 80/20 Rule Applied to Technical Startup Founders http://su.pr/2V5R0C […]

#4 Antonin Hildebrand on 02.23.11 at 9:53 am

And you can look at it from the other side. If you have a project in progress beyond point 3 and someone tells you that he is thinking to build something similar… competition? ehm, cheap talk. The odds are he will never manage to get to the point 5.

#5 DH on 02.23.11 at 10:16 am

From personal experience, making it successfully through each group is exponentially more difficult…at least the first time through.

#6 Dave on 02.23.11 at 12:07 pm

Step 5 itself could be broken into a further five steps. One of those being: “will have built something that other people want or need” (and ‘pay’ for?).

#7 Ajnabi on 02.23.11 at 11:24 pm

So that means there are only 1600 – 8000 developers have built their own business?

Rob Reply:

Not exactly.

First, this is much more for illustration purposes than for actual numbers. If it’s 1,600 or 16,000 the end result is not much different.

Second, as I was writing this I was thinking that it’s this amount per year. That’s completely arbitrary, but I think it’s more in line with reality.

Lastly, yes…I think that the number of developers who have built a business where they can support themselves is a lot smaller than one would think.

#8 Frank Paolino on 02.24.11 at 11:49 am

It’s funny, it reminds me of how many people get a black belt in karate, and you could almost write the same rules:

5 million have considered the martial arts:

Of those, 20% (1 million) want to get a “black belt”
Of those, 20% (200,000) have enough motivation to research a nearby dojo (gym)
Of those, 20% (40,000) will actually start training at a dojo
Of those, 20% (8,000) will actually last one year
Of those, 20% (1,600) make it to black belt

I have created a software company and gotten a black belt (I run the website at: http://doshikai.net)

and the underlying rule is personal commitment to a goal and the will to do the necessary work to reach that goal.

#9 Ryan on 02.24.11 at 2:22 pm

Interesting that you have preparation as the last bullet point, rather than the first.

Any specific reason?

Rob Reply:

The ordering implies that a lot of developers will build full software products without being prepared to launch/market them.

#10 Ali on 02.26.11 at 9:51 am

It’s hard, soul-crushing road. I’m still on the way, firmly in the ‘created something’ but not completed group.

It’s definitely an iterative process. You might find the first time you try to do it you just think about it a lot. The next time you think about it and start educating yourself. The time after you think, read and start making something. Hopefully the next around I’ll actually finish something sellable!

#11 John Haugeland on 03.04.11 at 8:02 pm

[Editor’s note: This comment was edited to remove references to a previously deleted discussion in this thread.]

This is not how the Pareto Principle works, and there is nothing called the 80-20 rule.

All you need to display the quality of this approach is to add a criterion. See how the result just changed by 400%?

That’s because this is not valid. There is no data to supporrt any of this. This is just a guy saying “here’s a list of strictures, I’m going to multiply an unsourced number by 0.2 each time and call the end result a valid result.”

One of the replies I recieved to a prior comment showed someone telling me that I should be equally open to a 70/30 rule or a 90/10 rule, to which the results are 12,150 and 50, respectively. The reason that person supported that viewpoint is that that person felt that there was no need for rules of thumb to match data.

When you consider that those two results are nearly three orders of magnitude apart, you begin to realize the actual numeric validity of this approach.

The Pareto Principle is just an observational criterion that a particular instance of the power law is frequent. Attempting to build things on top of it is a sign of a deep failure to understand mathematics and statistics in a fundamentally troubling way.

Sadly, I’ve discovered that, as a trained engineer, none of the math underlying the claims I’ve read through since is actually correct, and every time I try to point it out, instead of being thanked, I’m silenced and treated as an aggressor.

Unfortunately I cannot be honest about what this does to my trust in this advice, because then I’ll have given another excuse for this blog to be scrubbed down to where it looks like nobody has found broken math here.

Really, all you have to do is observe that some percentage of these developers like ham sandwiches. After all, no observation of the correct ratio for each criterion is given, nor is a datasource had, nor is any applicability criterion given.

In particular, RAND says that 88% of software projects are successful; one of the steps here asserts 20% for that, largely because nobody thought “well hey, making up numbers in business advice is probably not a good thing.”

The 80/20 rule *does* *not* *exist*. It’s like trying to talk to someone from reddit about language – they’ve all talked each other into believing that any common error is language development, so pointing out that every textbook, every teacher, etc say they’re wrong pretty much falls on deaf ears.

I went through the 61 developers between my day job, my IRC channel and my employees. Of them, 57 want to run a startup. Granted, 61 is not a statistically significant sample, and there is a strong selection bias, but it’s a better sample than “by applying this rule that doesn’t exist and is a failed understanding of the pareto principle.”

Of those, every single one has self-educated about the process. Every single one has started. More than half have finished at least once. More than half, in fact, have started and finished several of these – 38. Of the full set, nearly half are on their second spinup or better.

Preparing one’s self sufficiently to achieve success is of course a value judgement. If one asserts breaking even plus ten percent as a finish line, it’s 36 of my set.

So just going by a simple measurement, instead of five million down to 1600, we get 5 million down to

Of course, this result is also extremely, extremely wrong. Why?

Because it assumes that those fractions stack. This, of course, is nonsense. A simple example will suffice.

Let us posit a set of 100 flimbers. Flimbrs are people who flimb, which is my new nonsense verb. It pays well, to be a flimber.

Let us take the following observations.

Of our flimbers – and we’re actually measuring, we’re not just guessing – we want to know how many are good choices for trying to cross some local river in a publicity stunt.

So first we observe that 31 of them hate the cold. Then we observe that 27 of them can’t swim. Finally we observe that 18 of them are camera shy.

By the method described above, the ostensibly appropriate measurement would be 100 * .69 * .73 * .82, or 41.3 people. Sounds reasonable, right?

The problem is, this faulty calculation would only be correct if there was zero overlap between the sets, *and* that the probabilities taken were against the remainder set. This isn’t what happened. You see, in our set of flimbers, 25 are what programmers would call “cave trolls” – shut ins – who don’t like the outside, don’t like sports and don’t like crowds. They account for all but six of our cold haters (four tropical people and two skinny people), all but two of our non-swimmers (the remaining two have phobias) and all of our camera-phobes, because they’re overweight and pasty and don’t like their physical appearance.

Indeed, in our hypothetical set, the correct number is actually 69 available staff. That error rate gets larger the narrower the subcategory is, and larger the more categories you have.

Statistically, the best way to measure the appreciable failure rate here is to start by assuming that the 80:20 rule is correct for every category, even though available data sources say it’s radically wrong.

In that situation, you actually have to account for every overlapping range set – 20% of this 20% needs to go in the small partition, and the remainder in the large partition. Oh, third criterion? Break it up by scale into those four boxes. To a statistician, “The range criterion grows as the 4:1 partitioning of all extant partitions per classification.”

This is generally the same reason why most people cannot cope with Bayes likelihoods.

So, for zero criteria we end up with 100% match, and for a single criterion, we still end up with 80/20, which I will represent as [80,20].

For two criteria, we end up with [[64,16],[16,4]], because that’s 80% of the 80 = 64 and 20% of the 80 = 16, and 80% of 20 to 16 and 20% of 20 to 4.

To clarify, consider the case that someone has three six sided dice. The pareto rate for a die is, of course, 1 in 6 (dungeons and dragons players relax, it’s not a d20, and yes I also wish it was, because d6 cheat.)

To apply a rate principle to find out how often one gets three twos, this blog’s math would suggest that you multiply the rates together.

The problem is, this is what a statistician would call a “naive bayes likelihood” – something where all the variables are disaggregated. That is appropriate for dice, and for spam checking by blind word count, and for picking car color orders as they drive under bridges.

That’s because two dice’ rolls don’t affect one another, two cars’ paint jobs aren’t chosen together, and for spam, we actually get better results by not trying to understand the text, because more spam is generated by machine than hand written, so looking for human patterns gets bad results.

Such is not the case at all with people and interests. If you take for granted that 20% of people have started a business, it is not appropriate to also take for granted that 20% will take the time to learn to do it right. The latter group will be nearly 100%. Starting a business is a ton of work, it’s terrifying, and it usually means a huge personal risk. Reading a couple books is a near no-brainer, and every single startup person I’ve ever met has either read a bunch of books or complained that they don’t know which ones to get and are afraid of bad ones, or that they can’t afford them.

Assuming these things are disaggregated is ridiculous. Someone who spends $120k and 500 hours on starting a business isn’t going to cheap out on $60 of books and 5 hours reading. The risk makes that choice make no sense. You’re going to see self training in the already-started set, in my blind guess opinion, around 90%, not 20%. Granted in my measurement it’s 100%, that probably reflects my personality in selecting who got into that IRC channel – I have a strong personal preferential bias for people who do their research.

So.

Consider for example a set of sixteen hundred programmers. Let us take the unlikely position that they precisely 50% of the time like Star Trek, Dr Who and Firefly, three flagship science fiction shows.

The naive bayes likelihood that someone likes all three is only one in eight, because half and half and half.

Now.

Sixteen hundred programmers, and they’re 50% likely to like each of (list of sci-fi shows.)

Do you really think that there are only going to be 200 out of 1600 people there who like all three shows? Fifty percent of them like science fiction!

It’s very likely that the correct number is 45% or so. One in ten, give or take, judging by my later example, dislike one out of the three. (I happen to be one of them – I don’t enjoy Firefly.) Of those one in ten, only half dislike two out of the three.

The math here suggests we should estimate 1/4 of set, when the correct rate is 9/10 of set. The more TV shows you add to the list, the wronger it gets.

You don’t see stuff like that often in math.

Do you really think that if you add another flagship show to the list, the correct reaction will be to ignore the existing preferences and assume that this preference is randomly distributed in, and that therefore someone who likes three flagship science fiction shows is only equally likely to like a fourth?

It’s /nonsense/ .

What you’ll learn is that rules of thumb are for people who either cannot source data or who do not understand the importance of sourcing data.

For this final example, I will use a dataset you guys can go check yourselves. It’s unfortunate that I have to talk about 61 people only I know; if someone can think of a public datasource for that argument, I’d really like to open it up to inspection.

If you look at, for example, the dataset from the Netflix Prize, you get a dataset of sparse ratings over some sampling of 17,770 people over 480,000-odd films. It’s very sparse, but there are more than a hundred million ratings, so it’s a pretty damned good place to start. The only data you get is {film, arbitrary user id, 1-5 rating as integer}. The arbitrary user IDs are random but uniform – if something is a rating by user 346, you’ll never know anything else about user 346, but every rating attributed to user 346 is by that person, and that person only has ratings in that dataset under that ID.

So take for example the eleven star trek films. Using ratings across seventeen thousand people, I can make the following observations:

1) The chance that you will like one of these films at four out of five stars or better is approx. 71%, which is much higher than I had expected, because let’s be honest, several of those movies are garbage.

2) According to this math, the rate of people who would like all of these films at four out of five stars or better is 0.71^11, or 2.31%; the observed rate is 67% against whole, or if you prefer, 94% of criterion (that is, 67% liked all, whereas 71% like any one, and 67/71=.94,) so the estimate in this method is wrong by 96.5% of a possible 100%. That is, it’s within two standard deviations of perfectly wrong.

3) Two people who feel that they are informed by this math are willing to take bases at exponents which slide from 0.2 to either 0.1 or 0.3; they espoused that viewpoint in the hour before the comment was deleted. I believe this to be common to the viewership.

I bring this up not as a criticism or as an insult, but to point out how important seemingly small changes like that sound. An estimate of descending fifths from 5 million in five steps – that is, this blog’s estimate – yields 1600. That is, people are expecting a 1600 success rate from 5 million – one thirtieth of one percent. If you use tenths or 30% like those two posters suggested, you end up with 12150 or 50 respectively – a fifth of a percent or one *one* *thousandth* of one percent. Therefore, the rule of thumb is offering expectation within a range of 0.2% to 0.001%. Phrased that way, it no longer seems like a small difference. (Funny thing: SBA statistics are actually available on the topic of startup success, and they suggest that the correct number isn’t 0.00032 like the blog suggests, but rather 8.6%, or a failure rate of about 2500 times. This is equivalent to suggesting that the distance from the Earth to the Moon is about 95.6 miles. Do the math – I didn’t make that number up. The commenter’s willingness to accept 90/10 makes that number more like 2.77 miles – that is, the distance to the Moon is shorter than the typical CEO’s morning jog.)

The problem here is that the people applying the math by name don’t actually know what the math does. This is roughly equivalent to someone digging through a list of things that a builder has, and saying “okay I’m going to use these to make a wall.” They’re going to end up with something that looks kind of house-ish, and they’ll be very proud of what they made.

The problem is, the second someone comes along and says “dude, this floor is going to collapse and kill someone, it’s made out of drywall, it’ll be lucky to hold the weight of a chair,” they’re going to delete your comment because it’s insulting.

And I mean, then you end up with a blog about how awesome these drywall houses are, because their floor is so smooth and easy to vacuum, and you never have to listen to wood creak at night, and yadda yadda yadda. And you get ten thousand Twitter followers saying “wow check out these drywall houses!”

The Pareto Principle is a neat observation of the shape of number sets. It is not a guiding mechanic, it is not a law, and it may not be used to decide things. Anyone doing so is, generously, completely unaware of what the math says. This is a common mechanism for people running Ponzi schemes to make financial data look like it says something very different than what it actually says.

So I mean, I’m sorry about your drywall floor, Rob.

Let’s see if Rob proves me wrong.

#12 Marakas on 03.10.11 at 9:04 pm

I agree with your theory, but where did the magic numbers 80 and 20 originally come from? ( could it be 75 / 25 or is that just crazy talk ? 🙂

I am forever stuck in the grey-zone between number 3 and 4, I always get as far as starting work on my ideas and then just fizzle out before the code is finished.

My next project (already started coding!) is going to be so small that I have no excuse not to finish it. However, I am sure the profit will be on an equally small size as well..

Rob Reply:

>>could it be 75 / 25 or is that just crazy talk ?

Definitely; the exact math is not critical. The important piece to take away is that “many start but few finish.”

Good luck with on next project…each one you do gets a little easier.