parallel ent: insight from moderna, kwldg transfer from ML

[hyunjimoon]

Moderna mission and mindset explains 12 mindsets on urgency, parallel, risk, learning, pivot, question, push, owner, dynamic range, collective, platformize, digitize

1. we act with urgency
2. pursue options in parallel to make the best choice later
3. accept risk as the only path to impact
4. obsess over learning
5. pivot fearlessly in the face of new data
6. question convention as proven models don't always fuel the future
7. push past possible as greatness lives outside of comfort zones
8. behave like owners
9. act with dynamic range
10. remove viscosity to encourage collective action
11. prioritize platform over any single product
12. digitize everywhere possible using the power of digital information to maximize impact on patients

---

Among them, I'm most interested in parallel and pivot and they complement each other.

Parallel

This mindset allows us to uncover more possibilities - greater impact. It results in more complexity and more work, but it gives us the opportunity to choose differently later. Even when we don't exercise our options, we value them. We do more than we need to - the minimum is never enough.

Parallel interview by Christy Shaw transcript

So, yes, I think the boldness of moderna in trying many different things in parallel, has led us to where we are now in having a wide portfolio of programs in the clinic, and even more preclinically. Still, it's like.
A one of those funnels where you have tons of programs at the top and you have many in the clinic, and it's okay to fail on some of those, and we don't compromise safety or quality. It's really about taking risks and doing things in parallel. So putting a lot of resources into if you have a question that you want to address, don't just test it one way, test it multiple ways, so that you get an answer quicker, and you don't do it iteratively, if you do it in parallel. And that has allowed us to push the boundary. I think another aspect of our culture that is really special is everyone seems willing to get the job done in a way that even if they're not familiar or it's not their area of expertise, if there's a problem or if there's something that needs to be figured out, it needs to say, how do we do this? How do we move it quickly? It's really an all hands on deck. Everyone comes to the table with ideas, brainstorming, willing to go out of their you know, their training, their field of expertise, to get the job done. And while that's, of course, great for the company. I think what makes traffic to modern and statement are is that, that's great for the individual, too, because everyone can learn so much then.

Pivot

We use data to tell us when and where we can have the most impact, because impact is the only thing that matters. This mindset validates that any decision we make can be changed based on new information. We have no egos. We are regularly wrong, but we always decide for what is best Moderna long-term.

Pivot interview transcript

Thinking about the challenges and some of the tough situations that probably been at moderna. I think the easiest one is to go to the science. You know, it was pretty quick, you know, pretty rapid that we're able to identify something that we could progress forward and move forward. And, you know, a lot of excitement, you know, we were able to identify mutual space and delivery vehicles and push these things forward. And like, Okay, we're ready. We're going. And then we realized we don't exactly work the way we thought they were. And celebrate. That's, that's a challenge that you hear from the field, you know, it's all about how it is going to primate. That's the what is accepted translation into human and they didn't work the way we thought they were. And it was very deflating. Because, like, Okay, so now what we do, but then that's the modern piece, is that it was a learning opportunity. And we were aware of some potential challenges that could exist, and we were already working on that. And so it was all working on that. And so it was all you know, was able to kind of come together, but it was that, oh, maybe we're not going to enable to this modality, and how are we going to solve this problem? But we were able to swarm. That's why you hear that a lot at Moderna about how we're able to swarm when a problem is identified. That group of people get it together and go push and solve the problem.

[hyunjimoon]

Parallel vs. Sequential Search: Optimizing Value Creation Hypothesis Testing in Entrepreneurial Ventures

Research Summary:
This study investigates the efficacy of parallel versus sequential search strategies in testing value creation hypotheses for entrepreneurial ventures. We develop a Bayesian model using probabilistic programming to compare parallel "test-four-choose-one" and sequential "test-two-choose-one-twice" approaches across various contexts. Our analysis focuses on two key dimensions: technological innovation (component vs. system innovations) and customer segments (existing vs. new users). The study reveals that parallel search tends to outperform sequential search under specific conditions: (1) when the cost of each test is low relative to pivoting costs💸, (2) when uncertainty in technological innovations and customer segments is high🎲, and (3) when the true values of technological innovations and customer segments have low correlation🧩. We also examine industry-specific factors and company-level characteristics that influence the optimal search strategy, providing insights into when entrepreneurs should adopt broad, simultaneous exploration versus more focused, step-by-step approaches.


Managerial Summary:
In entrepreneurial ventures, the choice between parallel and sequential search strategies for testing value creation hypotheses depends on industry characteristics and company-specific factors. Parallel search excels in industries with fast clockspeeds, low test costs, high uncertainty, and low correlation between technological innovations and customer segments. For instance, personal computers, toys, and semiconductors benefit from parallel approaches due to their low test costs 💸, high uncertainty 🎲, and potentially low correlation 🧩. Conversely, industries like shipbuilding, petrochemicals, and commercial aircraft favor sequential strategies due to slow clockspeeds and high development costs. However, company-specific factors can override industry norms, as exemplified by Moderna's successful parallel search in the typically sequential pharmaceutical industry. Moderna's mRNA platform enables faster, cheaper development 💸, operates in a novel, uncertain space 🎲, and has potential applications across various diseases 🧩, making parallel search advantageous. This nuanced approach to search strategies enables entrepreneurs to optimize their exploration of technological innovations and customer segments based on their unique context and uncertainties.

keywords:
entrepreneurial strategy, parallel search, sequential search, value creation hypotheses, technological innovation, customer segments, industry clockspeed, uncertainty, Bayesian modeling, probabilistic programming

test2choose1 (isolate the bad ones)

ML analogy to transfer knowledge from parallel chain of posterior search to ent.search, implicit prior with bayesian simulation to ent.prior formation (strong to weak), technical debt to preventative maintenance of capability

• Leveraging Bayesian Inference for Entrepreneurial Advantage cld explores conditions favoring parallel search in entrepreneurial decision-making. We formalized three key conditions mathematically: (1) Exploration Capacity (N = B/c >> 1), indicating many affordable tests; (2) Option Independence (Σij / (σi σj) < ε for i ≠ j), showing low correlation between options; and (3) Value Uncertainty (min_i (σi / μi) > sqrt(k)), representing high variance in option values. These conditions were derived by connecting them to three axes of sequential decision-making algorithms: online vs. offline, frequentist vs. Bayesian, and independent vs. correlated. This framework provides a more precise basis for modeling when parallel search strategies are likely to outperform sequential approaches in entrepreneurial contexts.

• Modeling Parallel vs. Sequential Search Strategies for Entrepreneurs highlighted the importance of considering both industry-wide trends and company-specific factors when determining optimal search strategies for innovation. While our initial analysis suggested that slower-clockspeed industries like pharmaceuticals would generally benefit from sequential search, Moderna's case demonstrates that specific technological and organizational characteristics can enable effective parallel search even in traditionally sequential industries. Key factors enabling parallel search include low relative testing costs, high uncertainty in both technology and market applications, and low correlation between technological and market value. Moderna's focus on mRNA technology, heavy digitization, and unique funding situation create conditions conducive to parallel search despite being in the pharmaceutical industry. This underscores the need for a nuanced, context-specific approach when applying general principles of innovation strategy.

• exploring conditions for parallel ent

Parallel entrepreneurship refers to the practice of simultaneously founding and managing multiple ventures. This approach allows entrepreneurs to diversify their efforts, spread risk, and potentially increase their chances of success by pursuing multiple opportunities at once. Parallel entrepreneurs can leverage their experience, networks, and resources across different ventures, potentially creating synergies and cross-pollination of ideas. One of the primary advantages of parallel entrepreneurship is the ability to learn and adapt more quickly. By running multiple ventures simultaneously, entrepreneurs can gain insights from different markets, business models, and challenges, potentially applying these lessons across their portfolio. This can lead to faster innovation and more effective problem-solving. However, parallel entrepreneurship also comes with significant challenges, including the need to divide attention and resources, potential conflicts of interest, and the risk of spreading oneself too thin. Research suggests that parallel entrepreneurship can be particularly effective in rapidly changing industries or uncertain environments. The ability to pivot between ventures or reallocate resources based on emerging opportunities can provide a strategic advantage. Additionally, parallel entrepreneurs often develop strong skills in delegation, time management, and strategic thinking, which can benefit all their ventures. However, success in parallel entrepreneurship often depends on the entrepreneur's ability to build strong teams, establish effective systems, and maintain a clear strategic vision across multiple ventures.

Heimann93_ChallengerOrgStr.pdf is a paper that explains this with type1 and type2 error.

[EXHIBIT 2-5 Process flow diagrams for three product development processes]
The product development process follows a structured flow of activity and information. This allows us to draw process flow diagrams illustrating the process, as shown in Exhibit 2-5. The generic process flow diagram (a) depicts the process used to develop market-pull, technology-push, platform, process-intensive, customized, and high-risk products. Each product development phase is followed by a review to confirm that the phase is completed and to determine whether the project proceeds. Quick-build and digital products enable a spiral (or agile) product development process (b) whereby detail design, prototyping, and test activities are repeated a number of times. The process flow diagram for development of complex systems (c) shows the decomposition into parallel stages of work on the many subsystems and components.

[Heimann93_challengerOrgStr.pdf]
The paper discusses organizational reliability and structure, focusing on parallel versus serial configurations. While originally applied to NASA, these concepts are relevant to business environments, particularly in uncertain fields. Parallel structures, where multiple components operate independently, are advantageous in highly uncertain scientific or technological areas. In these contexts, the cost of missing a potential breakthrough often outweighs the cost of pursuing some unproductive paths. For example, in pharmaceutical research or rapidly evolving tech sectors, pursuing multiple development paths concurrently can increase innovation chances. The paper's k-out-of-m unit network concept is particularly applicable to portfolio management in venture capital or R&D resource allocation, allowing businesses to spread risk across multiple projects while maintaining a success threshold. This approach is especially beneficial in highly uncertain fields, though businesses must balance it with resource constraints.

[jeanbaptiste]

Sequential and Parallel Pivots: Two Exaptive Strategies for Entrepreneurial Innovation

---

Introduction: Strategic Experimentation and Pivots in Uncertain Markets

Entrepreneurs in dynamic markets face the ongoing challenge of balancing exploration with commitment. Key decisions revolve around how much to experiment and when to pivot in response to new information, which is often complicated by cognitive biases such as overconfidence. Structured experimentation programs—those that carefully manage how many experiments to run and the timing of strategic shifts—are critical for entrepreneurial success.

Pivots are strategic shifts that startups make when they discover new information or insights, usually in response to market feedback, technological advancements, or internal challenges. A pivot involves changing the business model, product offering, target market, or approach, while retaining the startup’s core vision or goals. Successful pivots allow startups to avoid stagnation and seize emerging opportunities.

Recent studies (Gans et al., 2019; Stern, 2019) emphasize that while frequent experimentation can spur innovation, excessive pivoting can disrupt learning and harm long-term performance. The Lean Startup model popularized by Eric Ries stresses iterative learning while cautioning against over-pivoting. In light of this, a well-structured experimentation framework mitigates such risks, enabling better-informed decisions.

Central to this framework is the concept of exaptation, borrowed from evolutionary biology. Exaptation describes how existing features, originally developed for one function, are repurposed for new, often unanticipated uses (Gould & Vrba, 1982). In the context of entrepreneurial innovation, two types of exaptation emerge: one where a feature previously shaped by selection is co-opted for a new use (adaptation), and another where features not initially shaped by selection are co-opted (nonaptation). These represent sequential pivots and parallel pivots, respectively. Both are crucial for understanding how startups navigate uncertainty and explore innovation.

---

Parallel and Sequential Search in Experimentation

Startups typically adopt either parallel or sequential search strategies in experimentation, each suited to different industry contexts and conducive to distinct exaptive opportunities.

---

Parallel Search: Rapid Innovation through Broad Exploration

In parallel search, startups run multiple experiments simultaneously, a strategy well-suited for industries characterized by fast-paced development and low testing costs, such as biotechnology, software, and semiconductors. This approach facilitates broad exploration, increasing the likelihood of discovering novel applications for technologies that may not have been originally designed for a particular purpose.

For instance, Moderna applied parallel search effectively during its early development of mRNA vaccines. By pursuing several mRNA programs concurrently, Moderna rapidly identified new applications for its platform. This process exemplifies nonaptation exaptation, where resources not previously optimized for a specific function are repurposed to meet emerging market needs. Such strategies expand the "adjacent possible" (Kauffman, 1995), enabling startups to discover opportunities quickly and increase entrepreneurial agility.

In high-uncertainty environments, especially for early-stage biotech startups, rapid experimentation facilitates the discovery of exaptive opportunities. This is particularly relevant in sectors where technological breakthroughs and market needs evolve rapidly, making parallel search essential for long-term survival and success.

---

Sequential Search: Methodical Learning through Stepwise Exploration

Conversely, sequential search follows a linear, methodical approach, where each experiment informs the next step. This approach works best in industries where failure is costly or where each test requires significant investment, such as aerospace and automotive engineering. Startups adopting sequential search can more deeply analyze results from each experiment, gaining greater understanding of their assets before moving forward.

Tesla provides a notable example of sequential search. In its early stages, the company focused heavily on refining its battery technology. Each iteration in its electric powertrain design built upon the previous results, eventually expanding into related areas such as home energy storage (Powerwall). Tesla’s deliberate expansion demonstrates adaptation exaptation, where technology optimized for one use (electric vehicles) is adapted to serve broader applications like home energy systems. Sequential pivots allow startups to mitigate risk, ensuring that each strategic shift is based on solid evidence.

While sequential search fosters slower innovation, it enables more informed decisions by allowing startups to carefully evaluate the results of each experiment, reducing uncertainty and facilitating more deliberate pivots.

---

The “Test Two, Choose One” Rule and Pivot Thresholds

The "Test Two, Choose One" rule (Gans et al., 2019) offers a practical framework for balancing exploration with decision-making in uncertain environments. This rule encourages entrepreneurs to test two equally viable alternatives simultaneously and select the more promising path based on data. Such structured experimentation reduces the risk of premature commitment to a single strategy and encourages adaptability to emerging opportunities.

In parallel search, this rule supports faster pivoting, as new applications of technologies or resources can emerge from simultaneous experiments. For example, Uber initially tested both black-car services and peer-to-peer ride-sharing models in parallel. After evaluating customer demand and operational feasibility, the company chose the more scalable peer-to-peer model, which ultimately transformed Uber into a dominant player in the transportation industry.

In sequential search, the "Test Two, Choose One" rule encourages more deliberate decision-making. By limiting the number of simultaneous experiments, startups can focus their efforts on one solidified experiment before moving on to the next. Setting clear pivot thresholds—specific points in an experimental program where decisions must be made based on results—helps entrepreneurs avoid the dangers of over-pivoting and ensure that each pivot is grounded in data rather than intuition.

---

Behavioral Biases and Entrepreneurial Decision-Making

Entrepreneurs are often vulnerable to cognitive biases such as overconfidence and the sunk cost fallacy, which can lead to either excessive experimentation or reluctance to pivot. A structured experimentation program, supported by clear pivot thresholds, can mitigate these biases by enforcing data-driven decision-making processes.

Structured experimentation forces entrepreneurs to rely on empirical evidence rather than gut feeling, helping to avoid emotional attachment to failing strategies. For example, startups in sectors like healthcare or fintech often face these dilemmas, where early indicators may suggest promising avenues that turn out to be dead ends. Establishing clear pivot thresholds helps mitigate the risk of overcommitting to a strategy or technology that is unlikely to succeed.

By incorporating explicit pivot thresholds into both parallel and sequential search strategies, entrepreneurs can optimize their decision-making process, reducing the likelihood of failure due to cognitive traps and increasing the probability of long-term success.

---

Exaptation and the Adjacent Possible

Exaptation plays a critical role in expanding the adjacent possible—the set of all future innovations that become accessible based on current conditions (Kauffman, 1995). By repurposing existing resources or technologies, startups can unlock previously unforeseen opportunities and strategically pivot into new markets.

In parallel search, startups running multiple experiments simultaneously are more likely to encounter exaptive opportunities, especially for resources or technologies that were not originally designed for their current use. For instance, CRISPR Therapeutics began as a tool for bacterial immune systems but was rapidly exapted for use in human gene-editing therapies, enabling the startup to address a variety of diseases from cancer to genetic disorders. This broad exploration uncovers hidden possibilities and allows startups to pivot in response to new market data.

In sequential search, exaptive opportunities emerge more gradually, often following a deeper understanding of the technology being repurposed. Waymo, Google’s autonomous vehicle project, initially focused on self-driving technology for ride-hailing services. Over time, as the technology evolved, Waymo incrementally expanded into autonomous delivery, reflecting how sequential pivots uncover new applications over time, grounded in methodical learning.

---

Visualization of Pivot Dynamics and Exaptive Opportunities

Visualization tools can play a crucial role in tracking how exaptive processes drive entrepreneurial pivots. By visualizing the adjacent possible, entrepreneurs can identify emerging opportunities and make informed decisions about how to reconfigure their existing resources. Mansinghka (2021) and Chen et al. (2020) recommend using dynamic visualizations to track the emergence of exaptive opportunities during parallel and sequential search strategies.

Monika Blattmeier (2023) delves deeper into the aestheticization of business processes, emphasizing how visualizing their Gestalt can enhance collective decision-making. This visualization approach helps entrepreneurs see the holistic potential of their business models and innovation pathways. Such tools foster collective thinking in startups, helping teams make more informed decisions when navigating pivot dynamics.

---

Conclusion: Leveraging Exaptive Strategies for Startup Success

The ability to pivot effectively is crucial for startups navigating uncertain markets. Both parallel and sequential search strategies, informed by exaptation, offer valuable frameworks for experimentation and innovation. Parallel search enables rapid discovery of new opportunities, while sequential search fosters deeper, more deliberate learning, reducing the risk associated with each pivot.

By incorporating structured experimentation programs, clear pivot thresholds, and an understanding of the dynamics of exaptation, startups can enhance their ability to innovate and expand their adjacent possible. Startups that master these methods are better equipped to discover novel applications for their existing technologies and achieve sustained innovation in dynamic markets.

---

References

1. Gould, S. J., & Vrba, E. S. (1982). Adaptation and Exaptation: A Critique of Some Current Evolutionary Thought. Proceedings of the Royal Society.

2. Fine, C. H. (1998). Clockspeed: Winning Industry Control in the Age of Temporary Advantage. Harvard Business School Press.

3. Kauffman, S. (1995). At Home in the Universe: The Search for the Laws of Self-Organization and Complexity. Oxford University Press.

4. Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision Under Risk. Econometrica, 47(2), 263-291.

5. Chen, J., Wu, S., & Lee, Y. (2020). Simulation and Visualization for Entrepreneurial Pivots. Journal of Business Research.

6. Blattmeier, M. (2023). The Aestheticization of Business Processes: Visualizing Their Gestalt for Collective Thinking. Journal of Information Technology. https://journals.sagepub.com/doi/epub/10.1177/02683962231166438

7. Gans, J., Stern, S., & Wu, J. (2019). Foundations of entrepreneurial strategy. Strategic Management Journal, 40(5), 736-756. DOI: 10.1002/smj.3010  

[jeanbaptiste]

Entrepreneurs in dynamic markets face the challenge of balancing exploration and commitment, particularly when deciding when to pivot based on new information. Pivots— strategic shifts in direction - are essential for adapting to market changes, but excessive or poorly timed pivots can hinder learning and harm long-term performance. The concept of exaptation (the repurposing of existing resources for new uses), offers two key strategies for structured experimentation: parallel pivots, which involve broad exploration of simultaneously tested ideas, and sequential pivots, a methodical approach where each experiment informs the next.

However, there is uncertainty inherent in these strategic decisions. In this study, we explore how they can be mitigated by using probabilistic programming and Bayesian modeling, which offer a structured framework for managing experimentation. We compare "test-four-choose-one" parallel and "test-two-choose-one-twice" sequential approaches across varying contexts of technological innovation and customer segments. Our initial results reveal that parallel search outperforms sequential search when test costs are low relative to pivoting costs, uncertainty is high, and true values of innovations and customer segments have low correlation.

We discuss how these approaches allow startups to continuously update their beliefs and adjust probabilities based on new evidence, enabling data-driven decisions about when to pivot. They might reduce the influence of cognitive biases like overconfidence and sunk cost fallacy, ensuring that pivots are grounded in evidence rather than intuition. By applying these probabilistic tools to both parallel and sequential search strategies, entrepreneurs might optimize innovation, minimize over-pivoting, and increase their chances of long-term success. This study contributes to the understanding of how entrepreneurs can systematically manage uncertainty, leverage resources, and expand their value in dynamic market environments, offering insights into entrepreneurial innovation within complex systems.

[jeanbaptiste]

Also, my favorite paper is Heiman 93 challenger regex.

It reminded me the difference between sequential (Turing / Von Neumann architecture) computational systems vs parallel systems such as CBA ALA, RALA and similar asynchronous computers such as Ivan Sutherland research at ARC.

[hyunjimoon]

transcript for chat with @jeanbaptiste https://otter.ai/u/2O2FmAybt6w_cldvMbJTiqhYmfg?utm_source=copy_url

[hyunjimoon]

Given our aligned target of connecting conditions for spandrel with 'low test cost, low variance within supply and demand, low correlation between supply and demand' being the condition for parallel search, I'd highly recommend felin_kauffman_zenger21-resource_origins_search.pdf on 'Resource Origins and Search.' Their concept of 'search images' - firm-specific ways of seeing and identifying valuable resources that others might miss - offers valuable insights for our work. Rather than conducting exhaustive search or relying solely on existing endowments, they argue firms can search more effectively by starting with a functional need or problem they aim to solve. This framework could enhance our understanding of when parallel search is optimal, particularly in situations with low test costs (when the search image clearly guides what to look for), low variance (when the search image helps identify specific resource properties), and low correlation (when the search image enables recognition of novel resource combinations). Their perspective on how firms recognize dormant value in resources could help us better formalize the conditions under which parallel search strategies outperform sequential approaches, especially in evolving market environments.

[tomfid]

My eye immediately jumped to Moderna #8: behave like owners. This sounds like standard corporate BS. AFAIK Moderna is publicly held, and employees aren't primarily the owners, so the board is really just exhorting them to work harder to produce surplus that will primarily be appropriated by the shareholders. That's fine - they know what they're getting into.

However ... the presence of this baloney leaves me questioning the other 11 items. Is "try n, choose 1" formally embedded in the culture and institutions? Are there unmanaged tragedy of the commons effects around the platform? Etc.

https://www.reddit.com/r/biotech/comments/1akfex3/moderna_not_for_everyone/

[jeanbaptiste]

Images from Monika Blattmaier's article The aestheticization of business processes: Visualizing their Gestalt for collective thinking

https://journals.sagepub.com/doi/10.1177/02683962231166438

's article

[hyunjimoon]

@jeanbaptiste could you explain more on how metaphors of atoms-molecules, DNA, Cells, Organisms work in your last screenshot? I made below using this process (i found ways to share process screenshot with you 🙂) and trying to connect with the metaphor what brought us together in our first chat - gene and protein in #184

Metaphor	Meaning in Business Process Visualization
Atoms-molecules	Representing the collection and screening of basic elements or ideas
DNA	Symbolizing the selection and grouping of elements to form a coherent structure or plan
Cells	Illustrating the creation of analogies and rearrangement of elements into functional units
Organisms	Depicting the explanation and structuring of the entire process as a cohesive, functioning system

[jeanbaptiste]

I think the common denominator here is the notion of Metaphor we used when describing complex systems. Atoms/Molecules structure, DNA double-helix, cell walls and organism bodies are reductionist representation we used as human to apprehend the complexity of what happens at different scales. Things are actually more complicated than that. For example, atoms/molecules are very complex EM field forces and quantum effects, DNA protein conform in various polymeric structures, cells can be studied from various fields depending on which properties you are interested in, as well as organisms whether you are a biologist, a sociologist, etc...

In this paper by Monika Blattmaier, what I find interesting are the transitions between different scales and the notion of Gestalt that integrates them with an attention to aesthetic, meaning to the perceptive (sensory) and cultural (archetypes) dimensions of the viewer, not only the purely efficient affordances of the representation. Finding appropriate visual variables that represent not only items but also groups and more complex structures are key to the research you are doing. In this sense, Bob Horn visual languages and similar ones like Tufte or Bertin are crucial.

[hyunjimoon]

Aspect	Test-Four-Choose-One 💠 (Parallel Search)	Test-Two-Choose-One Twice 🔷-💠 (Sequential Search)
Structure
Initial Segmentation	16 small diamonds	4 big diamonds, then 4 sub-diamonds
Testing Approach	Tests 4 small diamonds simultaneously	Tests 2 big diamonds, chooses 1, then tests 2 sub-diamonds
Example	P11-M11, P12-M22, P21-M12, P22-M21	(P1-M1, P2-M2) → P1-M1 → (P11-M11, P12-M12)
Costs and Benefits
Initial Costs	Higher segmentation and testing costs	Lower initial costs
Pivoting Costs	Lower potential pivoting costs	Potentially higher pivoting costs
Learning Speed	Faster learning across segments	Slower, more focused learning
Suitability
Uncertainty Level	Better for high uncertainty	Better for low uncertainty
Segment Correlation	Effective when segments have low correlation	Effective when segments are highly correlated
Industry Clockspeed	Suitable for fast-paced industries	Suitable for slower-paced industries
Key Features
Data Quality	Avoids noisy data through non-overlapping design	May be misled by noisy data in broad segments
Commitment Level	Avoids premature commitment	Commits to a broader segment before exploring sub-segments
Opportunity Discovery	Uncovers unexpected opportunities	May miss nuanced differences between sub-segments
Modeling Considerations
Segmentation Costs	Include additional costs for detailed segmentation	Lower initial segmentation costs
Learning Between Stages	N/A	Model increased individual heterogeneity to reduce impact
Partial-Commitment Learning	Model lowered costs (e.g., using GPT)	Model lowered costs (e.g., using GPT)
Exaptation and Innovation
Spandrel Creation	Creates larger "spandrel" or free space early on	Smaller initial spandrel, grows with each stage
Exaptation Facilitation	Facilitates rapid exaptation	Allows for gradual, focused exaptation
Industry Factors
Simulation Basis	Based on empirical research on industry clock speed	Based on empirical research on industry clock speed
Example Application	Tech startups testing multiple product-market fits	Traditional auto industry focusing on one market segment before expanding

[hyunjimoon]

based on

1. angie's prep for abstract submission Navigating Exploration and Commitment in Uncertain Markets cld

2. summary of yichen's feedback
   • 🔄 Comparability of test-two-choose-one-twice vs. test-four-choose-one: Angie suggests lowered cost of partial-commitment learning and faster testing clockspeed (e.g., GPT) may make the two scenarios more comparable
   • 🧠 Learning and development between stages in sequential search: Angie proposes increasing individual heterogeneity to reduce the impact of learning between stages
   • 💰 Need for additional costs in parallel search (segmentation costs): Angie agrees to add segmentation costs to the model for parallel search
   • 🌀 Circularity in the argument about exaptation and parallel search: Angie explains that parallel search creates a larger "spandrel" or free space early on, facilitating exaptation
   • 🏭 Clarity needed on industry-specific factors and company-level characteristics: Angie confirms this part is based on simulation using empirical research on industry clock speed

[tomfid]

Whether tests are organized as sequential or parallel seems relatively unimportant to me. Yes, there are some theoretical considerations of option value, whether the first test informs the second, etc. that are interesting. But I would expect that a real entrepreneurial case will usually be dominated by situation-specific factors: how many ideas do you actually have, how many people do you have to construct and supervise the tests, etc.? I would expect that you only have the luxury of multiple tests for things that are relatively unimportant (web site blue vs red theme and serif vs sans serif text); for the important strategic questions you'll be lucky to have 2 options.

I'm sure there are exceptions, as in pharma where there's a tradeoff between # compounds tested and sample size or depth of each test.

[hyunjimoon]

Just so you know, I may not have best label yet, but the idea of parallel has some relevance with "start big" modeling spirit for hierarchical model (our very first mail exchange) - and charlie approved of this start big that can prevent locked in from noisiness.

now, let me understand your comment more carefully

[hyunjimoon]

you only have the luxury of multiple tests for things that are relatively unimportant

I disagree. this should be one's choice.

seasoned startup lawyer john chory argues founders should speak with multiple investors although this may be against their instinct to move fast in #257 (comment) . relevant to your comment🗣️ when i introduced resource rationality tradeoff around action speed - degree of decision stupidity VS fixed cost.

underlying trade off there that isn't quite captured by our current model, that's interesting. So one end of it is, if you spend money really, really fast, you're probably making stupid decisions (not managing that money well), your burn rate is literally burning money, not using money intelligently. But there's got to be some, if you go the other direction and try to extend the runway for years and years and years, then fixed costs kill you.

GershmanHorvitzTenenbaum15.pdf

---

to summarize,

• There's a balance between making fast decisions (which may be suboptimal due to lack of information) and slower, more deliberate decisions (which may incur higher costs due to delay).
• It's important to consider multiple options, even for seemingly important decisions

To contextualize computation rationality theory in business processes

General Concept	Business Process Context
Value of result in cost-free world (ideal outcome if time and resources were not constrained)	Ideal business outcome if time and resources were unlimited (e.g., perfect product-market fit, unlimited funding)
Net value of action (actual value gained from a decision, considering both benefits and costs)	Actual business value gained from a decision, considering both potential returns and costs of implementation
Cost of delay (losses incurred by taking time to make a decision)	Losses due to delayed decision-making (e.g., missed market opportunities, ongoing fixed costs, competitor actions)