This essay was part of my final project for our 'Introduction to Cybernetics and the Foundations of Systems Design' class with Paul Pangaro.
Build. Measure. Learn. This is the Lean Startup mantra, a core methodological tenet to steer a new business to growth under conditions of great uncertainty. It's also a feedback loop, in which data ('measure') on a market's reaction to a product or service is used to make adjustments ('learn') to the product or service in short development cycles ('build'). The cyclical feedback-driven nature of the methodology makes it highly suitable to study in cybernetic terms.
Based on the Lean Manufacturing philosophy developed by Toyota in the 1980s, the Lean Startup movement was created and popularized by Eric Ries. He drew heavily on the Japanese car manufacturer's ideas around waste reduction, with waste broadly constituting any practice that didn't target creating value for the end customer. Applied to startups, the goal is to ensure what is being developed is creating customer value without wasting time, effort and money. It is something of an antidote to startups riskily developing a product in secret, at great cost, only to deliver it to an uninterested market (creating significant waste).
Embedded in the build-measure-learn process is the concept of 'validated learning', in which metrics and key performance indicators are used to steer product development. A startup following the methodology should seek to cycle through the process as fast as possible, wasting as little resource as possible while gathering valuable customer feedback. This frequently means developing a 'Minimum Viable Product' such as a prototype.
In cybernetic terms the Lean Startup methodology may be seen as a second-order loop. The inner loop sees the product or service trials which follow the build-measure-learn process. The outer loop controls the inner loop's goal and may adjust it based on consolidated data and learning.
Lean Startup Model [Generic] - click to expand
As much as the process seeks to be rational, any new company must have a business-hypothesis based on assumptions. Such assumptions directly inform the system's goals.
For example, a startup may have a business-hypothesis (assumption) that millions of people will want an app that lets them save movie and book recommendations from friends, and that they will be able to grow this business significantly with suitable funding. They will build a simple prototype and test it with a limited set of beta users.
Lean Startup Model [Example] - click to expand
￼The goal of this example inner loop (product trial) is to get at least 500 users, 25% of whom use the app at least once a week. Delivering the prototype is the actuator, and they may make adjustments to the prototype on a regular basis as part of the trial. Indeed, the Lean Startup methodology encourages 'continuous deployment' and multivariate testing as ways to optimize the product. The prototype is placed in the marketplace, specifically to a set of early adopting beta testers. This is the system's environment, in cybernetic terms. Usage is sensed in the form of data and analytics, and may be compared to the goal. The comparison is quantitative, but the learnings and how they may affect the actuator (next release) are subjective.
The outer loop may observe and sense the performance of the product trial, as measured by consolidated data and analytics, and compare this to its larger goal. This may be broadly to grow the business or become profitable, or it may be a more tangible next step towards this such as securing $2 million in funding to progress the business. The business has an assumption that to meet this goal, the product trial must be successful (i.e the inner loop goal is met). Eric Ries advocates a critical meeting every month or two to decide whether to 'persevere or pivot'. Much rides on this for the startup. There is no exact science dictating when persevering or pivoting is the better strategy, but the methodology encourages validated learning to inform the decision.
In cybernetic terms persevering or pivoting has a specific meaning for our model. Either the goal of the inner loop is changed, or it remains the same and the inner loop continues attempting to reach it.
We may consider the Requisite Variety for our model, and specifically the inner loop. Disturbances may include a lack of interest from the beta users, or a competitor or substitute product entering the marketplace and creating a threat. The system should have a variety of responses including user research, customer research, multivariate testing and marketing drives. Whether such responses represent enough variety to combat disturbances and reach the goal will be specific to each individual business and team.
Lean Startup Model [RV] - click to expand
While not an explicit part of the Lean Startup philosophy, I would argue that extrapolation is a key underlying principle. If X works for Y, then X will also work for many more Ys (scaling). This has two problems. First, if the environment is not representative of a larger environment, that scaling may not be possible. This is the classic early adopter problem, in that hyper-engaged 'ahead of the curve' users are not indicative of a wider market. We may suffer false positives. The second problem relates to Metcalfe's Law of network effect (specifically stating that the value of a network is proportional to the square of the number of connected users). If our environment is too small, with too few users connected, the value of the product may not be apparent. We may suffer false negatives.
￼ Such extrapolation errors are not intrinsic to the model. They may arise when the range is not set correctly, meaning the environment (limited marketplace) accessed will not reap effective data and valid learning. The range must be small enough not to create waste, but large enough to enable meaningful extrapolation.
While the range is not explicitly set by the outer loop, it is set contextually by the goal. If the goal is to trial a product with 10 users, a range of 100,000 is excessive. Similarly, if the goal is to test with 5,000 users, a range of 50 is prohibitively limited. Setting the correct goal, and proximately range, for the inner loop is a critical function of the outer loop.
Goal-setting within the model is the most subjective component. These are the assumptions, the business-hypotheses. Adapting the inner loop goal will primarily be the result of 'persevere or pivot' meetings. Ries details ten types of pivot: Zoom-in Pivot,
Zoom-out Pivot, Customer Segment Pivot, Customer Need Pivot,
Platform Pivot, Business Architecture Pivot, Value Capture Pivot, Engine of Growth Pivot, Channel Pivot, and Technology Pivot. These need not be discussed here, but note that only two of these relate specifically to changing the environment in our model. Customer Segment Pivot proposes focusing on different customers with the same problem the product or service is attempting to solve, and Customer Need Pivot proposes identifying a different need for the same customers.
In adapting the inner goal, a startup may benefit from a more nuanced range adjustment. This is less dramatic than the pivots above, but changes the 'persevere' focus slightly from optimizing the product to rethinking the environment, the range being the system's scope for accessing this environment.
Aligning goal and range optimally should lead to greater validated learning, and more robust extrapolation to guide the (hopefully) growing business forward through uncertainty.