This is the third post about Dan Vacanti’s book Actionable Agile Metrics for Predictability, An Introduction and how it relates to TameFlow. So far, the key findings are:

  • Unpredictability stems from poor flow.
  • Predictability depends not only on what you measure but also on what you do.
  • Actionable agile metrics are fundamental to TameFlow as they support the two fundamental patterns of Unity of Purpose and Community of Trust.
  • It is critical to have clarity in defining the start and end points of the process.
  • Clarity in definitions and agreement on terms is positively constructive of the essential TameFlow pattern of Unity of Purpose.
  • TameFlow works only with a pull-system, because push-systems inherently prevent the cultivation of Unity of Purpose.
  • It is of essence to focus on the Cycle Time (actual elapsed time), the time it takes Work in Progress to move through the process.
  • Cycle Time can be considered as a proxy of Operating Expense (in the context of Throughput Accounting and Financial Flow).

Now we will examine the next chapter in Dan’s book. It is all about Little’s Law.

Chapter 3 — Introduction to Little’s Law

This is a stimulating and very important chapter where Dan describes Little’s Law, presenting it in the two canonical forms (see the book to learn what they are!), and highlighting the assumptions that need to be satisfied in the two cases. Focus is entirely on understanding when Little’s Law is applicable, because understanding Little’s Law is at the foundation for using flow metrics to ensure actionable predictability.

Dan gives some background about Little’s Law; how the law was formulated on the basis of queuing theory; how it represents a relationship between averages, and not absolute values; and how that relationship interlocks the three fundamental metrics — Work in Progress (WIP), Cycle Time (CT) and Throughput (TP) — in a unique and consistent way.

In particular, Dan elaborates on the very important notion about the necessary conditions, or assumptions, for the law to be applicable. For the first canonical form the system needs to be in a steady state.

Then, the second canonical form is examined. In the second form, the law is expressed as:

TP = WIP / CT 

In simple words, one can say that the more work there is to be done, the longer it will take. Though, it is important to notice that the law actually expresses the relationship in terms of averages, and not in absolute terms. The conclusion is simple: In order to get work done faster, you need to work on less stuff! That is how Little’s Law tells what you can do to improve a process.

Dan continues to highlight how important the underlying assumptions are. The first form of Little’s Law is expressed in terms of Arrival Rate. The second form is expressed in terms of Throughput. The validity of the two expressions based on different sets of assumptions.

In TameFlow the second form is the one that is used extensively. So it is very interesting to learn what assumptions should be warranted so that Little’s Law is applicable. Referring to the second form, Dan explains that there are two different sets of assumptions, depending on weather or not the total amount of WIP is ever allowed to go to zero.

The case when WIP is allowed to go to zero is particularly significant, because then Little’s Law is exact between any two time points when WIP is zero! — provided that all work that enters the process also exits. In this case, the law is exact and applicable even if the system is not stable.

Now, from a TameFlow perspective this is very interesting.

In TameFlow we habitually promote the use of Minimum Marketable Releases (MMR). (For more information about MMRs, see the earlier posts: Virtues of Minimum Marketable Releases and Kanban Improved via Theory of Constraints). MMRs can be considered as a committed scope of work. Before a MMR is started, WIP is zero; when a MMR is finished, WIP is zero again. Therefore, Little’s Law can be applied exactly between the start and end points of an MMR. This is comforting and reassuring; it shows that TameFlow is solidly grounded on applying Little’s Law.

When using MMRs there will be several (economic) forces that will compel in making the MMRs smaller and smaller. Eventually it will be natural to consider and take the step of eliminating MMRs altogether, and move onto a situation of continuous flow — and this is the condition that is favored by Dan. However in a state of continuous flow, WIP never goes to zero, and there are additional assumptions that must be satisfied in order for Little’s Law to still be applicable. In fact there are five assumptions that must be warranted (see the book for the details!). In particular, attention must be exercised to guarantee the conservation of flow and to keep the process in a stable state. The meaning of this is expanded in later chapters of the book.

Dan makes the interesting contemplation about considering Assumptions as Process Policies. When the necessary conditions for the applicability of Little’s Law are violated, the process becomes less predictable. Therefore, those necessary conditions should give strong guidance when setting up the policies that govern the process. When process policies warrant the assumptions of Little’s Law, the entire process will become more predictable. The policies you decide to employ are what will ultimately determine the performance of your process.

Dan revisits the idea of Segmenting WIP and describes how it can be advantageous. In particular, he explains that Little’s Law will apply equally for the segments and the aggregate of WIPs. This is also relevant for TameFlow. When selecting the items to be included in a MMR, those items are often categorized according to risk profiles, sizes, or other criteria, in order to collect more finely grained flow metrics, and to fine tune forecasting. Even with segmentation and categorizations of WIP, Little’s Law still applies.

Dan then tackles a commonly held mistaken belief, namely that work items having to be of the same size. Yet that is not the case. Work items do NOT have to be of the same size in order for Little’s Law to apply.

Dan also warns that using Kanban Systems does not necessarily guarantee the assumptions underlying Little’s Law, especially when WIP limits are (mis-)used.

The chapter ends with considerations about Forecasting. Dan warns that one should not blindly use Little’s Law to project into the future. Little’s Law faithfully describes the past only. It is not intended for making deterministic predictions about the future, and therefore, it should not be used as a prediction device.

The best way to take advantage of the knowledge of Little’s Law is as a tool to understand what assumptions are needed for predictability. When the assumptions hold, the process becomes probabilistically predictable. Predictability is more about having a system that performs according to expectation, rather than making exact forecasts. So Little’s Law should be used primarily as a guidance to design your process in order to guarantee the assumptions.