In the business world (where most of my experiences in this area lie), companies are run on quarterly budgets that take into account resource allocation and potential projects. These projects usually have some sort of figure attached to them that represents a Return On Investment, or ROI. The goal then is to have ROI be as high as possible, while the resource allocation budget to produce that ROI needs to be as low as possible. For manufacturing processes, there are a number of factors that play into this – designers, prototyping, manufacturing, material resources, and so on. Physical costs, such as the base metal required to manufacture the body of a vehicle, are pretty much fixed. You can potentially swing a one-time decrease through intense negotiation, but you introduce yourself to potential shortfalls in quality to meet the new price. Not to mention that negotiating those kinds of reductions can be an extremely time-consuming and exhausting process.
The other aspect of any project is the, for lack of a better word, “flexible” cost. This is the cost for more ephemeral tasks, such as programming or architecture, where you cannot necessarily know how long the task will take before you begin. A project manager or product owner looks at their cost reduction options and sees this, thinking to themselves “Getting Joe to work a bit harder for a while will get this done faster, and be easier than reducing physical costs. Let's do that!” Without data to back up assertions that the goals imposed as a result are not achievable through reasonable effort, the programmer is left in the lurch – they have no power to change the situation, and thus often end up working long hours to meet the seemingly arbitrary goals.
The primary problem with the “just work more” approach is obvious to anyone who has done creative work for any length of time – creative work is exhausting. Performing that kind of work for any length of time can feel very much like running a marathon, conflating the mental and physical worlds into a conglomeration of perceived Herculean effort. After X number of hours, with X being different for every individual, it is simply not possible to produce at the same quality level as it was possible before X hours had elapsed.
The practical upshot of this is that quality of output for creative endeavors decreases as the time spent on that task in a session increases. This is the central point that is missed when attempting to procure more effort out of employees who derive their primary work product from the expense of mental effort – longer hours means more bugs and issues, which means longer hours to fix them. It's an infinite loop of increasing required effort, which often ends up costing more to complete than if the employee had been allowed to go at their own pace.
Note that the video game industry is horrible in this regard. Long hours (60-80 hour weeks or more) are the norm in that industry as projects move towards the completed stage. Project managers in the game industry are often constrained by hard dates – such as the holiday season – that they have no way to negotiate around, and so when you cannot manipulate your end date the only thing remaining is to attempt to manipulate the amount of effort performed each week. In many cases, this results in months of extremely long hours, where programmers and artists are pushed relentlessly to meet deadlines that are forever seeming farther away. Additionally, the cumulative exhaustion of weeks of long hours seeps in, resulting in the remaining amount of work increasing despite the longer hours as bugs introduced into the system by a tired programmer required effort to fix.
Having spent time in the game industry, I wish I could say that the above was an apocryphal example, but examples abound throughout the industry, and things rarely seem to improve. The marketing and funding realities of the industry make this a problem that is tricky to solve. Some companies have large enough cash reserves that they can simply ignore the calendar and say “It'll be done when it's done,” but most others need the infusion of cash provided by a completed project to continue to do things like pay rent, electricity, and employees (often in that order).
Analyzing the issue
The game industry has a situation that arises from external constraints, but it also is a byproduct of creative work – there is no way to know how long a task is going to take before it is started. Sure, you can have a ballpark. For example, I've written two books so far, and each took about 30 days for the final draft. Given that, it's reasonable to assume that if I were to write a third book, it would also take about 30 days for the final draft.
Except when it doesn't.
This is the most frustrating thing about programming to non-programmers. Two tasks can seem extremely similar, but one may take five minutes while another would take a week. The reasons why are numerous – fractured code bases, incompatible patterns, and so on – but ultimately there is no obvious reason that a layman can point to and say “Oh, that's why that took so long!” This problem, when repeated multiple times over the course of a project, can sour the relationship between a project manager and a programmer, taking what was originally a productive and respectful interaction process and making it confrontational. It is usually not the result of any intentional slight by the programmer or intentional misinterpretation by the PM, but instead it's an issue of trust.
Every time a programmer claims that a task will take longer than originally expected, or that they have been delayed, the PM has no choice but to assume that the delays are intentional as they have no insight whatsoever into the development process. This may not be an overt assumption, but it certainly occurs on some subconscious level and factors into decision making and interactions. This injects stress and discord into the process, which builds up over the course of a project and results in negative perceptions. A project could be on the most efficient possible track to completion, making strides that no other team could possibly make with the same material, but once the perception is soured that effort will never be seen as sufficient.
Addressing the Problem
The problem, as we've already mentioned, is that the PM has no way to know when a given task will be completed. The frustrating thing is that oftentimes, neither does the programmer. Sure a programmer can make an educated guess, but there are always factors outside of their control that play into the end result. Thus, the first approach to addressing the problem is to rely solely upon the guesses of the programmer. The programmer takes a look at a task before he begins, and gives an estimate of the effort required to complete that task. As a project manager is concerned with dates and timelines, they often push for time-based estimates (“It will take four hours”), and many programmers agree to the system.
Problem Solved, Right?
Once a unit associated with time has been decided upon, everything very quickly goes to hell. If a task that should “only take four hours” suddenly takes 40, then there is a perceived deficit. A project manager looking at a project plan sees the extra 36 hours of effort and is forced to conclude that “The project is now 36 hours behind.” This amount grows or shrinks as effort continues, and the perception of the project manager changes each time. If a task takes less time than expected, then the programmer is “making good progress” and the project is “on track.” If a task takes more, then the programmer is “introducing delays” and the project is “delayed.”
Notice the attribution of effort here. This all ties back to the choice of time as a base unit. Once you are estimating in terms of hours, the natural human tendency is to start watching the clock. We've been trained to pay close attention to time periods – 8 hours of work, 1 hour for lunch, 30 minutes to commute, and so on – that we subconsciously monitor the amount of time taken. When something takes longer than stated, it is frustrating and unsettling. By tying the estimates of a programming task into this time-based system on a granular level, we introduce the same predilections and perceptions into the evaluation of progress. It removes the ability to look at a task objectively and say “we are making X amount of progress.” Instead, everything is cast in terms of time. Time elapsed, time remaining, time behind, and when looking at the time associated with a project subjective judgments are added into the project status. Choosing an estimate in terms of time has resulted in looking at the project at a micro level, removing the ability to see the forest through the trees. The subjective perceptions of “late” and “early” that have been pounded in to us from our youngest years come into play, and color our perception of the project's progress.
One of the reasons behind this is that it is far too easy to look at a project as a series of finite tasks. Once these tasks have a discrete time associated with them, it is far too easy to take each task as an element in isolation. Good project managers, though, recognize that each task is only part of a larger whole. They strive to look at a project in aggregate – that is, as a combination of all of the tasks. They see the four hour task taking 40 hours, but then realize that the project is still 25% completed with one quarter of the budgeted time elapsed.
Thus, we see the path to our solution. An individual four hour task may run long, and may result in a delay. But on average, four hour tasks will typically take four hours. Some may be shorter, some much longer, as that is the nature of creative endeavors. However, looked at over a span of time, it is possible to develop a translation. If the average of all “four hour” tasks is 4.5 hours, that gives us a reliable predictive measure. We now know that if the programmer estimates a task at 4 hours, it will take 4.5 hours on average. This allows us to adjust our estimates and determine where the project actually lies.
A Problem of Units
At its core, with this translation table we've improved upon the primary concern – developing accurate estimates of project effort. However, we haven't addressed the soft factors of the issue. By associating progress with hours of effort, we are still tapping into that primal clock-watching urge, and thus still adding in the subjective judgements that go along with it. This is the problem with any time-based estimate system – it is tied to a known and familiar system, and even though it may operate on a varying scale, there is a subconscious perception of variation that permeates the project, regardless of the intentions. Because “hours” was chosen as the unit, this is tied back to real world elapsed time, and we end up with a slightly improved version of the same problem.
As it is, we now actually have the seeds of a potential solution. We've identified that looking at effort in aggregate provides us with more accurate estimates of project completion. We've also identified that expressing estimates in terms of time introduces needless subjective judgments into the process. We finally know that despite our best efforts, we do eventually have deadlines to meet – we cannot continue to just work forever.
A Different Approach
What we are really doing when estimating a task is building a view of the entire project. A project is the sum of its parts, and having broken a project down into constituent tasks it thus follows that a project's total effort is the sum of the efforts of all of its tasks (it is more complicated than that, but usually the point holds well enough). So by adding up all of the estimates associated with a project, we can determine a total estimate for the entire project. In a time-based system, this is expressed in terms of hours and days.
However, at this point we are working simply with numbers. We've established that due to the nature of creative work estimates are by definition non-concrete. So instead, we apply analytics to find out what “four hours” actually means in terms of calendar time. At this point, we're simply doing number analysis – averaging numbers together to produce a measure in terms of time. Since we have the same units on both sides of the equation, it follows that when these values on either side of the equation differ, then one of the choices of units is thus incorrect. It should be obvious that 4.5 hours does not equal 4 hours, so one of those figures must not really be “hours” as we know them.
Let's instead refer to the right side of that equation as points. Instead of 4.5 hours = 4 hours, we have 4.5 hours = 4 points. By removing the hours from the right hand side of that equation, we've suddenly addressed the subjective issue with our measurement system so far – we took away the time base in order to find the actual value, and as a result hit upon a number system that allows us to assign arbitrary values to tasks that develop meaning as a project progresses.
Building upon this gives us a more reliable means of estimation. Instead of assigning an hour estimate to each task, we assign a point value. The point value is indicative of the general effort required to complete the task, just as the estimate in hours was, but by removing the time value we remove subjective expectations. We assign higher numbers to challenging tasks, and lower numbers to easier tasks, knowing that on average the effort expended for each classification of task will average out to the same amount of effort, even if one specific task takes longer than another.
Tying it All Together
As we noted earlier, deadlines are inescapable. We cannot simply work forever. Thus, status reporting becomes a matter of determining how much longer the project will take to complete. We have a method of assigning difficulty to our tasks by giving them point values, and the point values are arbitrary units and thus remove all subjective judgments and perceptions. So our task at this point is to tie these points back into time periods.
This is actually what we were hoping to accomplish with our original estimation scheme – determine the amount of time it will take to complete a project. We simply hit upon the wrong units, and thus added a lot of extra cruft to our project plan. By trying to tie points back to hours, we become too granular – the points completed per hour varies far too much from task to task to give a worthwhile measurement. So the only option is to expand our averaging time window. This becomes an arbitrary choice, but I've seen the most success with week-long and two-week-long periods, known in the agile methodology as “Sprints”. You add up the points completed during each sprint, then keep a running average of the efforts completed. In doing so you tie the point values back into time values while also accounting for the aggregate variance of the data. For example – if I have a project that is has 170 units remaining, and I have been averaging 17 points completed per one-week sprint, I can say with a fair degree of confidence that there are 10 weeks remaining in the project.
This 17 number – the average point value completed during a sprint – is known as “velocity.” A project's velocity is used to relate the overall project duration back to the time required to complete it. By increasing the unit of measurement and divesting our estimates from subjective measures, we've introduced a more reliable system of estimation – simply divide points remaining by velocity. What this does is twofold. First, it gives you a quick and dirty estimate of the time remaining in a project that, over several weeks, tends to become very accurate as the amount of data to measure increases.
The second thing that it gives you is the ability to say no. Take, for example, a project that is starting to trend towards long hours. If you archive your data properly, you can point to previous examples where such an approach only made matters worse. Instead of “well, that will be pretty hard on the team,” you can instead say “We tried this before, and added 50 points of work while only achieving 47 from the increase in face time. How do you think it will be different this time around?” By looking at the point value of each task to be completed, some negotiation can take place and instead of giving subjective measurements and impressions, you can provide specific feedback with actionable data. Of course in some cases this won't do a damn bit of good, as the project deadlines may be particularly unforgiving, but in aggregate this approach should produce an environment that is collaborative and built upon a reliable history, as opposed to a relationship founded on distrust and fostered in a confrontational environment.
What does this mean for writing?
Ultimately, this method obtained from programming can be applied to any creative endeavor that can be assigned a subjective judgment of “easy” or “hard.” Simply break a project down into constituent tasks, then assign a point value to each task representing its difficulty. Of course after a certain point the tasks become repetitive – how many ways can “Write this scene!” be phrased, after all? However, one thing I have found when applying this approach to my own efforts is in support of my previous post on metrics – once you start putting your mind to measuring your progress in a more objective manner, you are able to get a better feel for how things are actually going with the project. While you may or may not have deadlines to work towards, and you may or may not have people waiting on your progress, the concrete proof of your progress is often enough to help you break through any bumps on the road, and stave off periods of despair over how long a given project is taking.