Hacker News Reader

Comment by pg | Hacker News Reader

"Herrmann said that the team’s research found that of the 90 percent of startups that fail, 70 percent scaled prematurely"

They must have had a strangely biased sample. Since the Bubble (when it was common) I've seen few startups do that. What kills startups is making something users don't want.

View on HN

Maybe by "scaled prematurely" they mean "started hiring and spending lots of money before they made something people want."

Edit: This wasn't intended to be snarky, I was being serious.

I assume that's what he meant, but it is comparatively rare. Nowadays most failed startups die small.

The way we've defined premature scaling, "making something users don't want" is included. For example, on the customer interaction metrics, which is what is used for "actual stage" a startup would be in the discovery or validation stages (stage 1 or 2). On the behavioral stage if they're making something people don't want, they're probably focused on streamlining their product or making it more scalable, which is a stage 3, or efficiency stage action. This would cause the startup to show up as a Behavioral Stage 3 / Actual Stage 2 and be labeled as premature scaling, just not the drastic kind, that for example WebVan exhibited — "Behavioral Stage 4 / Actual Stage 1". They had a team that was completely scaled up without even having shipped their product.

If I were you I'd just use words in the way they're ordinarily used.

If this is meant to be academic research (and I get the feeling it is), then it's unlikely they'd be able to use words as they're ordinarily used. Terms used will have to be clearly defined and stated in order to avoid confusion and misinterpretation (unless you're suggesting that such definitions already exist).

Unfortunately, this doesn't make it easy to discuss things outside the research field.

I'm not sure how else they could term it if they want a consistent label for all stages. Colloquially, it would be "they got too big for their britches", but that doesn't come across well in a study.

It would probably be better, though, to have phase-specific terms. It would probably make it easier to identify with each phase.

:) - I guess we can still improve on simplicity.

Given the mention of Webvan and the sheer number of startups in their study, it would seem their sample group does include a significant cohort from the Bubble. Trying to derive generalized rules governing startup outcomes is tough when those rules inherently keep changing.

Webvan is the first example of failed startup in Steve Blank's Four Steps to Epiphany. I think Startup Genome project was heavily influenced by Blank's book, so it might just that, a familiar example, not a data point.

That's correct. All startups in our dataset we're at least active between Feb 2010 and May 2010 when we started collecting data.

I'm thinking selection bias was at work. They surveyed startups they could find - how? VC lists? Bankruptcy filings?

How many startups failed that never got on their radar? Most?

I skimmed the report, their "methodology" link, whatever other links worked, and I did not find out how they found the startups they measured.

I regard startups that cannot get funding beyond friends/family as hobbies.

Using the term "scaling prematurely" is leaving the report open to misinterpretation: Since "scale" is one of their stated startup lifecycle stages.

In the report they use the term "consistent" which is less confusing and more descriptive of the symptoms of the "predominant" of failures. From the report....

"Consistent startups keep the customer dimension, the primary indicator of progress in a startup, in tune with product, team, ﬁnancials and business model. This means that each dimension progresses evenly compared to the others. Inconsistent startups have one or more of these dimensions far ahead or far behind the customer dimension. Premature scaling is the predominant form of inconsistency, but its much rarer opposite, dysfunctional scaling is also possible, although it’s not covered in this report."

Making (something users don't want), or not (making something users want)?

i.e. is failing to make anything at all a substantial portion of the failures, in addition to successfully making the wrong things?

They're both common. Some people build something elaborately wrong, others just don't get enough done.

one we call premature scaling, the other we call dysfunctional scaling ;) ... btw. @pg did you read any of our reports?

Reply on news.ycombinator.com