Can anyone translate this into language I can understand?
> Experimentation is contrasted with another mode of learning that I term Analysis. Classical competitive strategy emphasizes learning through analysis (Porter, 1980) whereby, instead of generating real options, entrepreneurs generate option sets and arrive at a decision through optimization (see Delmar and Shane, 2003, for an example). In my study, analysis includes gathering market, product, and regulatory data or developing business plans. Gans et al. (2019) distinguish analysis from experimentation in two important ways. Analysis is “commitment-free” whereas in experiments some irreversible investment in the ideas being tested is inescapable. In turn, the commitment-free nature of analysis implies that it can only yield noisy information, whereas experiments produce high-fidelity signals. I add a third difference, which is that analysis, in and of itself, conforms to standard and routinized practices,⁶ whereas experiments are, by definition, innovative and require counterfactual thinking (Camuffo et al., 2020). This delineation between the two modes of learning described above is also present in the practitioner literature (Bhidé, 2000; Blank, 2013; Ries, 2011).
I've been in the software business for a long time. I've been in startups and big companies, and worked with angels and VCs. But I have no idea what any of this means!
What is the market for a paper like this? And why can't it be written in plain English?
>Can anyone translate this into language I can understand?
(1) "classical competitive analysis" (Porter) would be basing business decisions on market trends.
(2) "experimentation" is what many of today's startups call "MVP minimum viable product" + "iteration", or "lean startup" popularized by Eric Ries[1].
So (1) would be more theoretical work in spreadsheets to model market size, risks, projected revenue, etc -- vs -- (2) would be start building something small in Javascript/PHP/C++ on AWS and keep iterating on features to see what "sticks"
The (2) experimentation is emphasized lately because cloud infrastructure like AWS makes software experiments very cheap and fast to test business ideas.
But (1) classical analysis is often the only choice especially for building physical products requiring upfront heavy capital investment. You can't "cheaply iterate" to build airplanes or launch a constellation of commercial communications satellites. You have to commit billions to build them first based on theoretical market projections and hope customers will buy it. E.g. both Airbus A380 and Boeing 787 were designed by working backwards from market data analysis. Neither had the money to iterate by having the factory build 10 different airplanes.
>What is the market for a paper like this? And why can't it be written in plain English?
But (1) classical analysis is often the only choice especially for building physical products requiring upfront heavy capital investment. You can't "cheaply iterate" to build airplanes or launch a constellation of commercial communications satellites. You have to commit billions to build them first based on theoretical market projections and hope customers will buy it. E.g. both Airbus A380 and Boeing 787 were designed by working backwards from market data analysis. Neither had the money to iterate by having the factory build 10 different airplanes.
But this is exactly what SpaceX is doing? Ditto with Starlink.
>But this is exactly what SpaceX is doing? Ditto with Starlink.
No, not the same.
For SpaceX, Elon spent almost all of of his $160 million PayPal wealth on 3 failed rockets before the 4th finally worked. And that 4th successful rocket test only demonstrated a proof-of-concept and didn't have a real payload from paying customers. That's an example of R&D in physical materials not iterating cheaply like the software world.
For Starlink, some google search finds this:
"Despite the heavy investment needed to build Starlink, the company's leadership estimates Starlink will cost about $10 billion or more to build"
$10 billion is not cheap to build/iterate.
Compare to the software world... Kevin Systrom building an iOS app for location check in (Burbn) and then pivoting to a different idea of photo filters (Instagram) is way cheaper than spending millions/billions on destroying rockets and launching satellites.
>, but fundamentally what SpaceX do is iterative development.
Yes, SpaceX and many other companies (aircraft companies, car companies, etc) also "iterate" ... but you're losing sight of the original question[1] that was ask by gp (Stratoscope) and the context of "iterative development" for this thread.
The author of the paper was trying to explain the difference between "classical analysis" vs "experimentation" to drive new business strategy.
Yes the companies forced to do "classical analysis" _also_ do product "experimentation/iteration". However, when the author breaks out "experimentation/iteration" as a separate type of "business learning", he's talking about Eric Ries' version of "fast & cheap iteration" and not SpaceX's slower more expensive iteration.
>It is ten billion dollars to build the old system, [...] Yes, they are all more expensive than software,
And these are the real-world financial constraints that bias "classical analysis" over cheap iteration of the author's two learning styles.
You kinda can “iterate cheaply” for SOME constellations. If your satellite can scale down to cubesat or smaller, it’s becoming feasible to iterate relatively cheaply.
I’m thinking of the space IoT startup Swarm (of radio space piracy infamy).
1 - Analysis
He's defining as using data from existing companies, economic trends, etc. Much of this data is relatively low cost and can be fine quickly.
Analysis tends to provide insight into existing supply and demand. E.g. How many companies are selling electric cars? How much did the sales increase/decrease this year?
Based on analysis, a startup might want to create a new charging station.
2 - Experiments
This is where someone puts an offer into the market and sees if there are buyers.
Usually these are for unique features, or possibly new products.
Example: clubhouse was an experiment
He's saying experiments usually cost more, because in theory you are at least prototyping + spending money on marketing/sales + support of early buyers.
____
Analysis looks to identify useful patterns in what is already happening, so factual.
Experiments try to find gaps that look for needs that aren't being met, so no behavior yet... So "counter factual".
____
Possibly he's also saying that deeper innovation happens based on experimentation.
And he's further asking: Do angel investors support experimentation more than other types of investors.
If it helps, skip to section 6 for the conclusions (spoiler: Angels are). The section you've quoted is academic signaling (in 'entrepreneurship research') to context and differentiate this paper from other research and entrepreneurship books/papers. And yes, I strongly agree with you: perhaps someday academia will embrace plain English and real hyperlinks.
And, in the fields which make heavy use of mathematical abstractions, “open-source” proofs written in a language amenable to automated proof assistants.
It is written like an undergraduate essay, which is what it seems to be. Judging by the responses it seems like most people haven't actually read it beyond the title.
My understanding: how will your company learn and innovate? Hire McKinsey to create a plan for you (analysis) or give a small team resources and autonomy to try something new (experimentation)?
> Experimentation is contrasted with another mode of learning that I term Analysis. Classical competitive strategy emphasizes learning through analysis (Porter, 1980) whereby, instead of generating real options, entrepreneurs generate option sets and arrive at a decision through optimization (see Delmar and Shane, 2003, for an example). In my study, analysis includes gathering market, product, and regulatory data or developing business plans. Gans et al. (2019) distinguish analysis from experimentation in two important ways. Analysis is “commitment-free” whereas in experiments some irreversible investment in the ideas being tested is inescapable. In turn, the commitment-free nature of analysis implies that it can only yield noisy information, whereas experiments produce high-fidelity signals. I add a third difference, which is that analysis, in and of itself, conforms to standard and routinized practices,⁶ whereas experiments are, by definition, innovative and require counterfactual thinking (Camuffo et al., 2020). This delineation between the two modes of learning described above is also present in the practitioner literature (Bhidé, 2000; Blank, 2013; Ries, 2011).
I've been in the software business for a long time. I've been in startups and big companies, and worked with angels and VCs. But I have no idea what any of this means!
What is the market for a paper like this? And why can't it be written in plain English?