MEASURE TWICE, CUT ONCE: SCALING NOVEL BUSINESS MODELS IN THE NASCENT

ONLINE FASHION INDUSTRY

Department of Management Science and Engineering

Stanford University

kme@stanford.edu

these ventures may grow quickly at first, they soon fade. Overall, our framework is reminiscent of Aesop’s

classic fable about the race between the tortoise v. hare. We also add insights about the role of data science,

optimal distinctiveness and extreme scalability, and about the process of how firms grow and more

significantly, scale. Overall, we indicate how promising ventures become successful adolescent firms.

1

In 2019, WeWork filed for an IPO at a stunning $47 billion valuation. Offering plush co-working

spaces that boasted mid-century style furniture and free cucumber water, WeWork grew rapidly in

nine years. Its attractive offering earned 80% occupancy and $1.8 billion in 2018 revenues. Yet,

the transition from promising venture to adolescent firm was difficult. WeWork’s IPO filing

revealed that in pursuing high growth, costs vastly outstripped revenue. Overnight, WeWork’s

valuation plummeted, the CEO was replaced, and the IPO was indefinitely postponed.

ultimately essential for many promising ventures to become successful adolescent firms.

Consistent with others, we define a business model as the system of interconnected activities

performed by a focal firm (and perhaps by users and partners) to create value, with part of that

requires linking external growth (e.g., revenue, customers) and profitability with internal growth

(e.g., headcount) (see also Chandler, 1990; DeSantola and Gulati, 2017).

spaces (better value than stodgy alternatives). It created this service through a combination of

activities like interior design, office management, and prime location (activity system creating

value) and charged renters (capturing some value, but clearly not enough to profit).

2

models are becoming increasingly diverse, theoretically relevant, and managerially significant.

They are central to long-term survival and growth (Massa et al., 2017). While some ventures may

persist with rapid but unprofitable growth (e.g., LinkedIn, WeWork), those without a profitable

business model will ultimately fail as standalone firms (McDonald and Eisenhardt, 2020). Beyond

their influence on performance (Casadesus-Masanell and Zhu, 2010), new business models often

open nascent markets (Gilbert, 2005) and are major sources of innovation (Casadesus-Masanell

and Zhu, 2013; Snihur and Zott, 2020). Examples like Airbnb, UpWork, and Uber, created new

markets and changed how people live, travel, and work (Demil et al., 2015).

business models. One stream identifies characteristics of high-performing business models (Zott

Rosenbloom, 2002) and Walmart (Brea‐Sols et al., 2015) with novel business models that

performed new activities in unique ways. Other work studies optimal distinctiveness – i.e.,

This and other work emphasizes processes in contexts with ambiguity, uncertainty, and rapid

change (Rindova and Kotha, 2001; Bingham and Davis, 2012; Raffaelli, 2018). These processes

3

include experimentation (e.g., Tripsas and Murray, 2004; Ozcan and Eisenhardt, 2009; Andries et

al., 2013), bricolage (Baker and Nelson, 2005), and trial-and-error (Rindova and Kotha, 2001;

Zuzul and Tripsas, 2020; Katila et al., 2008). Recent research suggests that effective entrepreneurs

use a broad repertoire of learning processes, and astutely pace their action in order to design

effective business models. For example, McDonald and Eisenhardt (2020) study five early-stage

ventures in a nascent fintech market. The effective teams engaged in trial-and-error and

experimentation, but also vicarious learning by imitating others and passive learning by pausing to

observe. Further, Ott and Eisenhardt (2020) point to the role of problem solving to organize

learning. Collectively, this research suggests that learning, bricolage and problem solving

processes are central to designing high-performing business models.

both profit and grow. Broadly, entrepreneurial research rarely considers profitability, and

examines growth primarily by studying its antecedents and outcomes (Burton and Beckman, 2007;

Eisenhardt and Schoonhoven, 1990), not the process itself. Thus, this work leaves open how firms

grow and more significantly, scale – i.e., achieve profitable growth. To address this gap, we ask:

How do entrepreneurs build scalable business models in nascent markets?

approach (Eisenhardt, 1989). The setting is the nascent online fashion industry that emerged

4

around 2010 with several novel business models. Using in-depth interview data and archival

sources, we closely track three venture-pairs with distinct business models from founding until

they reached profitability or exited. A unique feature is our spotlight on adolescent ventures – a

critical stage where promising ventures like WeWork often stumble. A second unique feature is

highlighting the compelling role of data science for high-performing business models. We also

take advantage of a turnaround to sharpen causality. Finally, while all of the ventures were initially

promising, only one in each pair achieved high growth and profitability (i.e., scaled).

Our primary contribution is an emergent theoretical framework for how entrepreneurs build high-

growth, profitable (i.e., scalable) business models. Our theoretical framework indicates that

successful entrepreneurs first 1) slowly design a simple, profitable core transaction by

how conceptualization and problem solving shape when and what is learned. Thus, while prior

research emphasizes learning per se (e.g., Andries et al., 2013; Bingham and Davis, 2012; Ott and

Eisenhardt, 2020), we indicate what is actually learned, when, and how that learning is organized.

Second, we add to the literature on optimal distinctiveness (e.g., Navis and Glynn, 2010; Zhao et

al., 2018). Specifically, we sketch how data science amplifies optimal distinctiveness, enables

extreme scaling, and underlies capabilities that become “cheaper” and “smarter” with growth.

5

feature of internal organization. Prior work focuses on attributes like team composition (Burton

and Beckman, 2007; Snihur and Zott, 2020), organizational design (Gilbert, 2005) and culture

(DeSantola and Gulati, 2017). In contrast, activity systems describe the linked processes, skills,

and assets by which work gets done. We also contribute insights into adolescent firms by noting

that they grow on the scaffolding of a simple, profitable core transaction. In contrast, without that

scaffolding, adolescent firms may grow but not for long, fail to become profitable, and fade.

METHODS

Given the limited theory and empirical evidence, we use multi-case theory-building (Eisenhardt,

1989). We use multiple cases because they typically yield emergent theory that is more robust and

generalizable than single cases (Eisenhardt and Graebner, 2007). This approach also fits with

that introduce operational complexity like inventory that make scalable business models more

difficult to achieve. Third, since online fashion emerged recently (around 2010), we can study its

pioneering ventures more easily with first-hand data while avoiding left-censoring issues.

novel business models, we mean those that add new ways to create value beyond just a new sales

channel (Zott and Amit, 2007). As we learned more, we saw other ventures with similar business

model templates and a third template. Thus, our sample became six online fashion ventures (i.e.,

6

three pairs) (Table 1). The templates were: (1) retail with personal styling (i.e., buy clothing, select

and send items to online customers who then choose to buy or not), (2) rental (i.e., acquire clothes

and rent to online customers), (3) second-hand marketplace (i.e., match second-hand clothing of

online sellers with online buyers). While each template broadly addresses the same nascent market

(online fashion), they face varied challenges: clothing recommendations (high for personal styling,

low for rental), logistics (high for rental, low for marketplaces), community engagement (high for

marketplaces, low for personal styling), and time sensitivity (high for rental, low for personal

styling). This theoretical (not random) sampling of ventures and business models allows for

replication within and across business models, and so improves reliability and generalizability.

i.e., size (about 2), age (about 30), education (mostly elite MBA), and entrepreneurship experience

Data collection

We draw on multiple data sources: 1) archival materials such as news media, company documents,

and teaching cases, 2) interviews with internal informants such as founders and executives, 3)

interviews with external informants such as board members, customers, and analysts, 4) shopping

experiences by research team members, and 5) informal emails and phone calls to clarify details.

These multiple sources enable triangulation among data sources and improve accuracy

(DeBenigno and Kellogg, 2014; Eisenhardt, 1989; Glaser & Strauss, 1967) (Table 2).

7

LexisNexis, and Factiva to search for all media articles that mentioned each venture since

inception. The first author read all media accounts for each venture. While many articles mention

the ventures, we concentrated on focal articles that are substantial (i.e., at least 750 words) and

foreground one or perhaps a few firms in detail. These focal articles appeared in major newspapers

(e.g., New York Times, Wall Street Journal), business outlets (e.g., Forbes), and technology-

oriented press (e.g., TechCrunch). Focal articles typically cover recent events like product

launches, and discuss the venture’s actions, strategy, rivals, business model, and history. We

continued this data collection during the study. We also added data from company websites

(including archival web-scrapes of early websites using Archive.org), company blogs and venture-

related social media during the study. While these latter sources were often not relevant to our

members, and managers across functions (e.g., product, community, merchandising, data science).

These interviews provided rich insights such as about key events, actions, and their rationale. We

also conducted six interviews with external informants including board members, customers, and

analysts. We supplemented these interviews with 57 online interviews with internal informants.

These began within the first two years of founding for all ventures and continued until the study

end. The interviews are a rich source of real-time early information. These triangulated data from

8

multiple relevant perspectives and times provide a richer and more reliable understanding of

building business models than any single source could provide (Miller et al., 1997).

describe their role briefly. Second, we asked them to provide a chronological account of the focal

venture’s history (since founding or the prior interview). For external informants, we adjusted the

guide to fit their knowledge like industry (analysts) and products (customers). The interviews were

30-90 minutes in length, and were recorded and transcribed. We took several steps to improve data

validity. First, we gave anonymity to informants to encourage candor. Second, we also used

interview techniques such as nondirective questioning to gather open-ended narratives and

minimize response bias (Eisenhardt and Graebner, 2007). Third, we triangulated events and

actions described in interviews where possible with our archival data such as media articles,

Data analysis

Consistent with multi-case theory building (Eisenhardt and Graebner, 2007), we began analysis by

writing detailed case histories for each firm. Per above, we triangulated data from multiple

sources. The archival data (including focal articles) were particularly useful for building an initial

timeline, as well as triangulating key events and activities described by informants. The interview

data were particularly useful for fleshing out timelines and rich process details that were

unavailable in the archival data (e.g., decision making, alternatives considered but not pursued).

The five teaching cases were primarily useful for industry history, timelines, founding narratives,

15

backgrounds in hospitality and banking. Sartorially aware, they realized that many women

(including a co-founder’s sister) often bought expensive designer clothes that they wore just once

(e.g., to a special event), to avoid repeating outfits. The founders spotted an opportunity to rent

designer clothing to these fashion-conscious customers. A co-founder said,

designers using their existing websites. For a fraction of the price, customers could rent clothes

from their favorite designers, and then return the garments once worn. RentRoyale would manage

the operations, and share the revenue with the designers. A co-founder described,

The founders began by learning about supply. Over a couple months, they met with prominent

fashion designers and retail executives. They soon learned that designers were not interested in

renting clothes on their own websites. Rather, they worried about “cannibalizing” their revenue by

replacing would-be sales with rentals. A co-founder described a typical encounter,

16

prices if the venture attracted younger customers (i.e., late teens to thirties). Such customers could

be introduced to the designers’ clothing, and might later become buying customers. In turn,

wholesale prices were essential for RentRoyale’s unit economics. A co-founder explained,

economics. Particularly uncertain was whether young customers would look after expensive rental

dresses. A co-founder asked, “Once [young women] rent dresses will they destroy them? How

many times can you rent them?” Over six months (Q2-3 2009), the founders conducted three

experiments. In the first, the founders hosted a pop-up at a college campus to learn whether

customers were willing to rent at about 10% of retail. Third, the founders were surprised and

pleased that 96% of the dresses were returned on time and in perfect condition. A co-founder said,

“I figured out from this experiment that people actually really take care of things when they have

an appreciation for the aspiration of that brand.”

hire five employees, and build a basic website. The CEO described,

17

broadly about the unit economics, the team realized, for example, that storing garments at a dry

cleaner saves transport costs to/from storage. So they added a dry cleaning partner. The CEO said,

create an online account and then join a waitlist. In addition to providing time to hone the core

transaction, too little inventory would be discouraging for customers. A co-founder said, “We do

short, RentRoyale designed a simple and potentially profitable core transaction.

explained, “Our whole approach is to test things out, see what works”. Yet, they had only $1M in

revenue, 8 employees, and about 1,000 items in inventory. An industry observer critiqued their

stunted growth, “RentRoyale has only 1,000 dresses in its warehouse, and 750 of them were out on

rental on New Year’s Eve.”

18

growth, and neglected simplifying the core transaction and improving its profitability. Instead,

they quickly settled on a core transaction and grew. Yet, they ended up with a more complicated

(i.e., more elements) core transaction with weaker unit economics (i.e., profitability).

opportunity with online styling similar to StarStyle. The initial idea was for remote stylists to

conduct video consults with customers. The stylists would then buy relevant clothes and mail them

to the customers who would buy what they liked, and return the rest. A co-founder said,

selling. They eschewed thinking through or broadly about the core transaction or experimenting.

Instead, they jumped in with trial-and-error learning about sales. As a cofounder noted,

months meeting with customers, styling outfits, and selling. As the CEO described,

19

The team quickly came to see excellent stylists as essential. An early employee noted,

only) in order to build the right corporate culture and customer value. As a co-founder explained,

“Folks working around the country doesn't really make sense because they're not really bought

into the kind of company and culture we want to build.

sales. In less than a year, StyleStuck’s revenue was growing ($70,000 in 5 months). The CEO saw

The team hired a sales manager and replaced remote stylists with on-site ones. In Q1 2011, they

invited customers to their office where they had added a “shopping bar” for in-person styling. This

attracted better customers (i.e., more likely to buy) and fueled more sales. A co-founder explained,

20

Later in Q1 2011, StyleStuck had spent about a year emphasizing growth. They were off to a quick

start with $5.5M in revenue, about 60 employees, and over 3,000 customers. Yet, StyleStuck’s

core transaction had costly unit economics.

logistics (like RentRoyale) with prediction and subscription pricing. The team emphasized brand

and rapid growth using online ads and social influencers to gain customers. Unlike RentRoyale,

they mostly ignored unit economics by neglecting costs (e.g., bought at retail, not wholesale).

Likewise, MarketMiss quickly settled on a complicated core transaction that combined a

marketplace with prediction and inventory. They focused on growth while their counterpart

contrast, others (StyleStuck, RentWreck, and MarketMiss) aimed for growth. They sprinted ahead

to Series A in less time, and grew faster in revenue, customers, and employees. Yet, they ended up

with a more complex and likely less profitable core transaction (Table 4).

Era 2. Cut once… Accelerating by building a complete activity system

In Era 2, the entrepreneurs fleshed out the core transaction from Era 1 by building activities. By

activities, we mean repeatable organizational processes (Ott and Eisenhardt, 2020; Porter and

Siggelkow, 2008). Underlying activities are often valuable skills (e.g. data science, sales) and

21

assets (e.g. warehouse, brand, algorithms). Effective entrepreneurs (StyleStar, RentRoyal,

MarketMagic) followed a process of: 1) focusing on profit and growth (i.e., scaling), 2) building a

core capability that was central to their business model, 3) adding superior data science skills, and

4) completing the activity system (Table 5).

transaction for a rental business model: buy clothes from fashion designers at wholesale, and rent

at 10% of retail. After Series A (Q2 2010), they turned their attention to scaling – i.e. profit and

growth. As an executive noted, “If you scale properly, it's sustainable growth… You have good

economics”. To do so, the team tried to outsource operations like supply chain logistics and dry

cleaning. As the CEO said, “I tried really hard initially to outsource”. Yet, the team came to

believe that reverse logistics – i.e., activities to receive a garment, clean, and turn it around for

The team also now saw reverse logistics as an inimitable advantage. A co-founder noted,

and helpful, this small dry cleaner was costly and unreliable. As a manager noted, “Outsourcing

was extremely expensive, unreliable.” An executive elaborated, “We could not control the timing

after drop off or the quality…. And we were missing weekends - obviously where the demand

was.” Instead, the team hired a dry cleaning star from a high-end chain. As one executive noted,

22

“We were very fortunate to find someone who had experience….But at the same time, he was not

family - like “mom and pop” dry cleaning”. Another executive elaborated,

RentRoyale also hired spotters (e.g., identify and remove stains) who were surprisingly hard to

find. An executive called them the “holy grail”. The CEO echoed, “The hardest position to recruit

has not been engineers, it’s been spotters!” The team also purchased cleaning equipment and set

up a facility. An early employee noted the significance of beginning with cleaning, summarizing,

“The biggest lift obviously is going to be cleaning.”

science. This Chief Algorithms Officer began to intertwine physical operations with optimization

availability of clothes. The CEO described their value, “An operational hub is incredibly

important to solve problems. It also allows us to have a last minute business in those cities.”

that combined efficient physical operations with ever-smarter algorithms. This created faster

turnaround of garments, and so increased revenue (i.e., more items available for rent) for the same

inventory. Similarly, better cleaning increased the # of wears per garment (i.e., revenue per item)

23

that was critical for profitability. In short, better reverse logistics activities led to more inventory

turns, more quickly, and so fueled more profitable growth. The CEO described,

(e.g., garment lifetime) and better predict what inventory to buy. As a co-founder noted,

RentRoyale also shared these valuable, yet rare data with their designers. As one designer

enthused, “You learn a lot from that and from reading the comments the clients make online. It’s

great market research!” These data became a valuable, rare resource that deepened ties with

designers and lowered inventory costs via better deals with designers. The CEO noted,

early on, but impaired profitable growth by underpricing some garments and overpricing others.

The data scientists made a significant move by drawing on an analogy to the dynamic pricing AI

algorithms of airlines that become smarter with more transactions. An executive explained,

24

As more customers could find a garment at the right price, growth surged. An executive extolled,

“This sophisticated pricing engine… altered the trajectory of user growth”. Simultaneously, the

algorithms became smarter and their unit costs decreased (for this scale-free resource).

and other capabilities. Underlying these activity bundles were valuable, often rare and inimitable

skills (e.g., cleaning, data science) and assets (e.g., warehouse, rental data). They distinguished this

activity system for scaling (i.e., profitable growth) from short-term actions to boost growth

temporarily (i.e., “growth hacks”) like paid marketing. An executive explained,

Another executive echoed the value of RentRoyal’s activity system,

system around a core capability: reverse logistics. The CEO summarized, “RentRoyale is a

proprietary reverse logistics platform. It's an operations company.”

transaction for a retail business model: Customers complete a style survey and pay $20. In-house

31

activities like boutique supplier relationships to an integrated activity system that provided

product-market fit for a small market. A co-founder contrasted StyleStar and StyleStuck,

acquired by a large, “brick-and-mortar” retailer that wrote down the investment about a year later.

StyleStuck never became profitable. Instead, they were closed. An executive lamented, “We didn't

figure out how to take down the bigger opportunities, the StyleStar opportunity.”

rental business model template. Yet, they complicated the core transaction: Customers complete a

style survey and pay a monthly subscription (~$50). RentWreck would predict 5 items that the

media influencers and Facebook ads. A VC described, “They were doing grassroots marketing…

nothing that I would say is total rocket science”. They also invested in brand which they saw as a

valuable asset: a strong brand makes adding customers easier. An executive said,

Yet unlike StyleStar, they did not hire top talent or blend algorithms with humans. Instead, since

32

their data scientists built an algorithm that served 80% of customers without humans, the team

fired most of the stylists. Yet, while the algorithm modestly improved styling (e.g., faster), the

team had assumed (but not tested) whether renters wanted styling. Yet after building the algorithm,

the team realized that renters actually prefer to pick their own clothes. As a co-founder noted, “If

we really wanted to be a long-term part of someone's life, we had to give them more control.”

After wasting 9 months of time and investment, the team eliminated styling.

cleaning. Yet unlike RentRoyale that hired top talent (e.g., dry cleaning experts, top data

scientists) and spent two years honing reverse logistics, RentWreck lost focus in under a year.

With incomplete activities (e.g., one warehouse), they switched back to marketing. For example,

they formed partnerships to rent celebrities’ clothing lines. An investor said, “They figured out

better”. In 2017, RentWreck had $75M in revenue, but was far from profitable. The founders

were replaced. As a board member said, “The founders were in over their heads.” RentWreck

merged with an offline retailer, was never profitable, and the merged firm ultimately failed. An

investor summarized, noting the incomplete activity system,

33

reason is their simple core transaction from Era 1. It is easier to build a complete activity system

on a simple transaction. Complexity often adds costs and more ways to fail. Second, they anchored

their activity systems on a core capability that was valuable, rare, and difficult to imitate. This core

capability was central to the business model, and was a source of competitive advantage. Third,

they emphasized data science (e.g., hiring top data scientists, placing them in c-suite roles, and

synergistically blending algorithms and humans) to improve revenue and/or cut costs. A key point

is that AI algorithms support scaling – i.e., become more profitable with growth because they

become more accurate (i.e., learn) and are scale-free (i.e., per unit cost declines).

market fit and/or growth hacks. While some (StyleStuck, RentWreck) built core capabilities, these

RentRoyale), MarketMagic slowly designed a simple and likely profitable core transaction. As the

CEO said, “I love the power of simplicity - how simplifying and progressively simplifying leads to

scale!” Drawing on an analogy to StubHub, MarketMagic charged sellers a simple 20% fee for

any sale on the platform. They also simplified shipping by enabling sellers to mail items directly to

buyers at a flat fee. Finally, the team delayed growth through their invite-only approach.

34

capability: community engagement. The team believed that a highly engaged community was

central to its marketplace business model. A co-founder said, “Ultimately having the largest seller

community allows you to have the largest buyer community, which allows you to have the largest

platform over time.”

entice new ones. As an experiment, the team hosted a small event in Las Vegas, and invited

several potential customers and local fashion bloggers. While only 15 people attended the event, it

unexpectedly generated significant media coverage. The CEO said,

growing. July was an exponential month!” MarketMagic spent the next year pursuing growth.

They engaged in online ads, PR campaigns, and marketing promotions. As social media became

very effective for mobile marketing, growth exploded. As the CEO summarized,

35

lacked a complete activity system (e.g., robust platform, secure payments, reliable shipping). As

profitability tanked, the CEO noted, “The unit economics were deteriorating… The engagement of

the users was really high, but just the whole system was not yet there.” The platform began failing

with too many parties in late 2013. While these parties engaged the community, their high activity

caused frequent crashes. The CEO lamented, “I’m fighting the infrastructure. And parties - the

concurrency levels are insanely high… It's real time buying and selling… And so the system is

under incredible stress.” Another noted, “Payments was breaking”. A third echoed, “The

infrastructure pieces… We should have really invested much sooner.”

36

second-hand clothing sellers to buyers. Yet, MarketMiss focused on growth in Era 1, and designed

a complex core transaction: A seller receives a box with pre-paid postage to fill with unwanted

clothes and provides details like sizes and styles. Once MarketMiss finds a matching buyer, the

seller mails the box to the buyer who pays a $13 fee and supplies a new box. So it blended

community engagement (MarketMagic), logistics (RentRoyale) and prediction matching

(StyleStar). It was complex like multiple revenue models, multiple steps for buyers and sellers,

and many different items per box. Nonetheless, MarketMiss initially grew quickly.

on growth by adding products like toys (Q4 2010) and books (Q2 2011), and partnering with

charities (Q3 2010). They added a daily blog (Q1 2011) that the CEO justified, “We think about

Similarly, while MarketMagic made listing items easy (i.e., post a smartphone photo on the

website), MarketMiss lacked simplicity for sellers and so the marketplace stalled with too little

supply. The CEO complained, “We couldn't get enough velocity. If you were a seller putting the

stuff on the platform, there was just enough friction. They were like, this isn't really worth it.”

Instead, they pursued growth by growth hacks and searching for product-market fit with a complex

core transaction. They peaked at several hundred thousand users and were never profitable. As an

37

executive admitted, “The first thing we need to do is kind of own up to the fact that we have to

sunset our peer-to-peer marketplace”. In 2012, the team shut down MarketMiss. The CEO

summarized, “Peer-to-peer wasn't the right business”.

DISCUSSION

We began by observing that while promising ventures like WeWork may initially grow, without a

scalable business model they may ultimately struggle in the transition to a profitable, growing

adolescent firm. By studying six promising ventures across three business models in online

fashion, we develop a theoretical framework for how entrepreneurs build scalable (i.e., profitable

growth) business models (Table 6). We contribute to the learning literature by indicating when and

what is actually learned. To the business model literature, we add insights into how entrepreneurs

achieve high-performing business models, including how data science can create optimal

Emergent theoretical framework

models (Table 6). Our theoretical framework indicates that successful entrepreneurs first 1) slowly

design a simple and profitable core transaction by actively delaying growth, and then 2) accelerate

by sequentially building a core capability followed by a complete activity system. Thus, they

follow a phased and exponential-growth process.

These entrepreneurs use a variety of learning approaches consistent with prior literature, such as

experimentation, trial and error, and analogical reasoning (e.g., Murray and Tripsas, 2004; Gavetti

38

et al., 2005; Ott et al., 2017). Yet, their common outcome is a simple and profitable core

transaction. A simple transaction (i.e., fewer elements) clarifies the unit economics, and contains

fewer parts that must work together well. Such a simple “semi-structure” is also robust in highly

uncertain markets like nascent ones (Davis et al., 2009). Further, while growth may solve many

problems, it does not solve unprofitable unit economics. Thus, establishing a simple, profitable

core transaction provides the scaffolding for scaling. The cost of this deliberate and broad learning,

however, is slow pace and low growth. Indeed, successful entrepreneurs actively delayed growth

to provide time to conceptualize and learn broadly how to be “smartly simple”. In contrast, less

successful entrepreneurs race to add revenue. Yet, by learning and conceptualizing narrowly, they

resolve few uncertainties and often end up with a complex and likely unprofitable core transaction.

adaptation. In contrast, less successful entrepreneurs proceed with “growth hacks” like ads and

price-cutting promotions that may spur revenue that may not last. While these entrepreneurs may

build a core capability, they often do not complete their activity systems and integrate their

activities too soon or not at all.

Learning literature: Equifinality, simplicity and profitability

We also contribute to the learning literature by highlighting what is learned. Prior research

indicates relevant processes like experimentation (Murray and Tripsas, 2004; Camuffo et al., 2019;

39

Bremner and Eisenhardt, 2020), trial-and-error (Rindova and Kotha, 2001), bricolage (Baker and

Nelson, 2005), and analogical reasoning (Gavetti et al., 2005). Extending this work, we find that

successful entrepreneurs use a variety of processes. For example, RentRoyale emphasized the

experimentation approach of lean startup while StarStyle engaged in more conceptualization and

trial-and-error as well as experimentation. MarketMagic often relied on analogical reasoning to

similar firms like StubHub and to their own prior venture. Yet they all reached the same outcome

of a simple and profitable core transaction. Thus, while many processes can be useful, the common

practice is a) learning deliberately and broadly to identify a simple and profitable core transaction,

and b) organizing problem solving to focus on a core capability and then activities.

money on every transaction even as they grow. Without simplicity, transactions are prone to

difficulties in making the transaction actually work, high costs due to more elements, less clarity

about unit economics, and unresolved uncertainties.

models. By using data science, firms can amplify optimal distinctiveness. That is, they take

familiar offline capabilities like merchandising and personal styling, and transform them by

synergistically integrating algorithms with human action. The resulting activities are familiar in

40

terms of their functionality but also different (and often cheaper and more effective) from what

they were. Data science also enables extreme scaling. With the scale free economics of algorithms

(Levinthal and Wu, 2010), they become cheaper per unit with growth. Thus, when algorithms take

advantage of what humans do well and blend those strengths with the computational power (and

scale-free economics) of algorithms, the resulting activities are highly scalable. Finally, algorithms

built on machine learning and other forms of AI become smarter with growth, further enhancing

scaling. Indeed, StyleStar’s machine learning widened the gap as the venture grew relative to

StyleStuck that used traditional software that may scale, but does not become smarter.

system as a key feature of internal organization. While activity systems are recognized in strategy

outcomes (e.g., Eggers, 2012), but overlooks the activities that support them. By emphasizing core

capabilities (i.e., VRIN resources), this work underplays the activity system as a whole. We

contribute by unpacking how activity systems are built and their relationship to core capabilities.

work emphasizes antecedents of growth like team features (e.g., Burton and Beckman, 2007;

Eisenhardt and Schoonhoven, 1990) and outcomes like routinizing (DeSantola and Gulati, 2017),

but leaves open the process of growth. Recent work on scaling adds by offering that internal

Community

Engagment

(High)

TABLE 6. Emergent theoretical framework: Scaling business models.