Slashdot Mirror
How can a Developer Estimate Times?
SubliminalVortex wonders: "Many times in the past, I have been asked on 'how long' it would take to implement a certain features/fixes in a product. What's interesting is that many times, certain 'fixes' is adjusting the wording/placement of the items in question; in other cases, users want the product to do everything they ever imagined, since it already started by following their line of thought. From there, the problem continues. From the user interface, people 'imagine' and think that 'oh, it would be easy if...' and scenarios occur, not only internally from the company using the product, but the clients themselves. Usually, several good ideas are there, but estimating times is a pain in the arse if you have a platform you're writing code for which has no documentation. How do coders estimate times to their bosses? If I know the answer outright, I'll give it, but in some cases, I don't how much time I'll take from other developers *because of the lack of documentation*. I'm going to have to bring in my D&D dice next week just to start."
The only way to tell is to design the feature, then guesstimate how long each part of the process will take (by comparing it to how long similar problems have taken). If there's a step you're unsure of, you need to be sure of it to give a reliable estimate. Then double it- something will go wrong, and its better to deliver early than late.
I still have more fans than freaks. WTF is wrong with you people?
The Estimator's Rule:
It always takes longer than estimated, even after accounting for the Estimator's Rule.
...use a random number generator? :)
More seriously, I've found from past exeperience you should always take whatever time you think the project will take and multiply that number by 2 or 3. If you are lucky (and the customer requirements don't change to much) you should make it.
Two things really... you have to break your tasks down into small tasks with known dependencies to the point you *can* estimate times to do things. Once you get that far, experience (gut feeling) will contribute. After that, you give an optimistic and pessimisitic times to your boss. Example: If all these assumptions hold and these other three things happen on time, then the time estimate is one week. If this, this, or this doesn't pan out as expected, then the time estimate is three weeks.
To take a leaf from Star Trek just do what he always did. Say it will take 3 times as long as you think it will, and you will always get half that time to complete it. And if you get it done early you will be rewarded.
Win-win.
Count how many dumbshit managers with no technical knowledge you have over you.
Double your realistic estimate for each DM.
Skivvy Niner? Email me!
HEY! Look left just ONE MORE TIME!
I've been developing as a job for almost twenty years, and I still don't know the answer to this question.
The best approach I've found is to decompose the problem into chunks that are small enough to give a reasonable estimate of, but I've hit two snags with this:
1. it may take time I don't have to do the decomposition
2. managers don't like big numbers
Good luck!
Dunx
Converting caffeine into code since 1982
Just Google "software estimation" and you get a variety of ways of doing it (the most popular formal model is COCOMO). In the real world, most people don't have the luxury of doing estimates the right way. In my experience, the stakeholders want a good guess at first (aka SWAG, OOM, VROOM, LOE). They treat that initial guess at the high watermark and then expect you to either finish early if all went well or finish on time if there were problems.
How I go about making this initial guess is by breaking the project up into sub-projects and then creating a baseline estimate by determining how long it would take me to do each piece myself. I then determine which person on my team will be working on which piece and then adjust from the baseline for that component based on the engineer's performance level compared to mine.
Once I have all those man-hours calculated, I double it. This is now my code and unit test estimate. Finally, add 10% for project management and another 25-50% more for quality assurance.
It seems kind of loose, but it works for me.
For the most part, your ability to give an estimate on a change is a matter of experience. General experience will give you a rough idea of how long it would take to implement some random feature. Specific experience with the system in question will give you a much more refined ability to make an estimate. If you can't make an estimate, or you can only make a rough estimate, be upfront about it. Tell your manager or the customer (if you're the one working directly with the customer) that your estimate is very rough, and try to add as much buffer time as possible. Also, don't get into granular estimates when you can only give a rough SWAG (Some Wild-Ass Guess). If you think it'll take you an hour, say a day. If you think it'll take you a day, say three. If you think it'll take you three days, say a week (experience will tell you how much buffer to add). If you don't have enough experience with the system to make a good estimate, ask your co-workers for help. It's always better to under-promise and over-deliver than the other way around. Besides, while you may end up over-estimating some feature, chances are you also underestimated another one and the extra buffer time you don't need for feature A will be invaluable to finishing feature B.
To help you make better estimates, you should make sure you get a solid set of requirements from your customers. Time should be spent with the customer to determine exactly what it is they want (they may not be able to articulate it clearly), and whether or not there's already a solution they could use if they're only willing to change their process. Avoid telling the customer at that point whether something is "easy" or "hard" (and definitely ignore them when they suggest something will be easy -- they don't know the inner workings of the code, so how will they know if it's hard or easy? Changing a piece of text may look trivial, but may involve a number of external factors such as localization or legal that would turn a 5 minute fix into a two month battle). Take what you've learned, distill it down into a solid set of requirements, and take that back to the customer for sign-off. Once they've signed off, they don't get to change their mind any more for this cycle (however long that may be -- days, weeks, months, or even years, though hopefully not nearly as long as that). Then you get rough estimates from the developers based on their gut feelings and knowledge of the code base, and have them investigate deeper to solidify those estimates. At that point, you have enough to build a schedule, but allow for buffer time! A developer's 8 hour day may not be 8 full hours of coding. It may be 4 hours of coding, two hours of meetings, an hour for lunch, and an hour of "filler" for the cost of task-switching depending on how the coding hours and meeting hours overlap. Based on your experience with past projects, you should have a good idea of how much time per day a dev actually gets to spend on coding and use that to build your schedule. For example, developer estimates are all in 8 hour day increments, but you know from experience that developers average 6 hours per day of productive work. That means a 5 estimated days will actually take 6.7 real days, so use that as your buffering criteria. (BTW, six hours of productive work is actually high. In most cases, it's really closer to 4 or 5.)
Finally, don't forget QA. Just because a developer has confidence that he can implement a specific feature in one hour doesn't mean it won't take QA a week to verify all of the ramifications of that change. You really need to have your QA team working in lock-step with your dev team. Your dev team should write technical specs on the features, and your QA team should estimate impact off of those. The "agile" disciplines that are en vogue lately try to incorporate testing directly into the development process, which is good. However, that means that developers will be in charge of including QA costs on their estimates, which developers are usually bad about doing
He said "double it then raise it to the next unit of time."
One hour becomes two days. A week is two months. A month is two years.
P.H.: If you are reading this I'm still in the business.
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
As others have said, you should estimate based on similar tasks, and then overestimate before giving that number to management. But there's also something to be said for being honest. Most management types I've dealt with are just fine when you say "I don't know if the application allows us to make that change quickly, so let me do some research and get back to you tomorrow with an educated guess." It helps if every so often you come back to them before the end of the day and say that it was an easy change and you've already finished it. Finally, when working on more than a few things at once, I make sure there's a prioritized list that I'm working from that management is aware of (so they understand why the latest request will take more time) Also, I make sure there's regular progress on one or more high priority items. Management and customers always sleep better when they see forward momentum even if the deadlines slip a little. Spending a week with nothing to show makes them nervous even when things are on time.
If a client insists on an accurate estimate of "how long it will take" for a non-trivial project, then you pretty much have to get them to provide you with or pay you to generate a requirements spec, a functional spec, and then an implementation spec. From the implementation spec, you can then ballpark how long each part will take to code and test. If the client balks at putting exactly what they want on paper, then you should balk at providing any kind of estimate; because there is only one truth in that situation: it will take as long as it takes.
This might help: Painless Software Schedules.
"I think it would be a good idea!"
Gandhi, about Internet Security
Don't expect anything on this list to solve your problems.
o Break the problem down into small pieces. If you have an item on your schedule that says "Do X" and the estimate is "two weeks", you do not understand what X is. Keep splitting it into X.1, then X.1a, then X.1a.ii, until the pieces are all less than three days, preferably much less.
o Clock time != development time. You will spend half your time going to meetings, preparing TPS reports, and making phone calls to find out why your test server is down. To convert development hours to schedule hours, use a factor of two.
o Keep an estimator's handbook. Ever noticed these in a technical bookstore? You can look up how long it takes to plumb a bathroom, build a wood frame wall, etc. Keep your own list of how long it takes to build a credit card form or a bulletin board. Look up the phrase "function point" in ancient literature and see if you get any useful ideas.
Time estimation is part of project managment. A good course in project management will give you the tools to effectively deal with these situations.
Short and sweet, when asked to give an estimate (any estimate) ask them if they want the quick and dirty estimate now (which may be off by 1,000% or more) or if they would like a more accurate estimate. If they want the quick and dirty then use whatever method you use now to give them the estimate. If they want an accurate estimate then tell them you'll research the current state of the project, the development effort required for functions x, y, and z, and the resources required to accomplish the effort.
Then follow a good estimating process - there are many available in many books about this subject. In my case generally new projects are very similar to older projects, and are all built on the same foundations so it's easy to start with a list of tasks to get from point A to B, and list the resources and dependencies for each task. Link them all together, and then go to the boss and say, "Assuming that we have these people and these reources, the project will be completed x weeks from kick-off."
When (not if) the boss says, "Well it needs to be done by this date!" Calmy reply with, "We can meet that goal by either scaling the work back, or increasing manpower. Which is more suitable for this situation?" and so on.
The one thing you don't want to do (that all too many software engineers do on a regular basis) is over estimate or underestimate the work required. In either case the customer (generally your boss) learns not to trust your estimates. This is bad on so many levels, and is the main reason that so many software engineers work more than 40 hours per week on a regular basis - especially those that overestimate the time required trying to combat this (paradoxically).
A more accurate estimation process will include a frank discussion of risk and risk management. "This schedule is good as long as we assume the code base is well documented and follows reasonable coding practises." Then find out what the customer wants you to do if you find this assumption to be false. Give them the whole scenario - the risk may be low, but if the assumption proves wrong then perhaps the project schedule or resources just tripled. That may be too great a risk to take even if it's unlikely - so how do we mitigate that risk?
I could go on, but really this is a basic question that millions of people have asked for thousands of years. There are much better and more thorough resources to answer it. Visit your local library if your company isn't willing to send you to a simple three day project management seminar. They won't teach you much that's new, but they'll tie it all together and give you the tools you need to communicate effectively with others - most of the important stuff is being able to say, "I'll get back to you on that by thursday."
-Adam
Don't do it!
I'm serious. If we've learned anything over the past forty years, it's that trying to estimate software development time is a waste of time. Why? Well, for one things, bugs. To know how long it would take to find and fix them, you'd have to know what the problem was, which means the hard part's done. For another thing, specification flux. Most reasonable project specifications for non-trivial projects don't provide enough information to accurately predict how long it will take to code. And producing such specs, and using them, will take longer than the development would have taken in the first place. And, if that's not enough, think of Dilbert. Specifically, the one where his boss asks him "Will there be any unforeseen problems?"
I once told a client (a dot com) "I'll tell you how long this will take if you can tell me what you stock will be worth six months from now, four months after you IPO. And I'll be more accurate than you will. If you'd rather, you can tell me when the next two-day rainstorm will be." He hemmed and hawed, and basically refused to give any sort of number whatsoever. I said "Good, it looks like we've got a nice honest relationship going here. Neither of us is willing to lie to the other, or claim to know something we don't. So let's start prioritizing these features for the mock up."
Let me say it again: Don't even attempt to estimate development times.
So what should you do instead? Incremental development with frequent opportunities for feedback. Ideally, the users should be able to play with a nightly-build prototype whenever they want from the first week or so on. If they ask for a time estimate, tell them it will be done when they're happy with the result, and don't want anything more, or they're unhappy with the rate of progress, and call the whole thing off.
But don't lie to them, no matter how hard they push.
--MarkusQ
"I'm skeptical, because for the same reason 30 days of development estimted, might really be 60, 2 days of development might really be 4."
Yes, but it's better to know in two days that your estimate is wrong than wait thirty days. Also, a lot of SCRUM is breaking things down to an understandable level. I.e. down to problems like, "Design the initial schema for the main database table" rather than problems like "Define the database schema for the project." It's much easier to determine how long it will take you to write up a practical first draft for a single table than it is to do all of
1. Determine all the data your project will need.
2. Break the data into normalized chunks.
3. Select indexes to optimize query performance.
4. Denormalize to optimize query performance.
Note that the first step is completely open ended. Does your project need one table? Ten? A hundred? Waterfall development processes try to guess this ahead of time. E.g. if most projects average about thirty tables, then that's what this will need. If you only need ten, then you have too much time allocated and over examine the database schema. If you need a hundred, you allocated too little time and miss your date.
The third and fourth steps should actually be done *after* you've started working with sample data. SCRUM allows you to push out problems like this. Waterfall forces you to guess. This leads to duplicate work. Further, every step after the first is dependent on the first. Thus, if you misquote the first, you misquote every step after it. With SCRUM, you openly admit that the first step is not all you need to do and then stop worrying about the rest.
Another advantage of SCRUM is that if you put too much into the current iteration, you can drop stuff. Then, when you do the next iteration, you can pick some or all of that back up with revised estimation. In Waterfall development, you're forced to rewrite the whole estimate every time you do that. With SCRUM, it's just part of the work you would do anyway (you already knew that you were going to plan out the next iteration).
The last thing you ever want to do when asked by a manager is give a off the cuff answer. It will almost always be wrong.
So what do I do when I get asked how long something will take? Well, to start off, if I know the code in and out, and I'm aware of the bug, then I can actually estimate what amount of time it will take. If it's not a bug, but something to be developed, and I've done the exact same thing before in my career, I already know the answer.
If however, neither of these are the case (which is about 98% of the time), I say this: "It'll take me roughly x hours to investigate this matter further and only after that will I be able to give you a timeline that is accurate."
First off, that gives you way more credibility, and way more leeway. But second, it lets you dive into the situation without having committed your life to fixing it, until you get a better grip on what's going on. If I were for example dealing with a bug on a web app (the kind I regularly work on these days), I would say something like "It'll take me 4 hours to investigate". I will most likely start with about an hour spent understanding the scope and possibly the history of the bug. "Is it reproducible?" is the most important question. If it is intermittent, I will commit to nothing at this point. If I look at the code and can see what is causing the problem, that is, if I can see a clear cause and effect chain that agrees with the test cases etc, I rely on my prior experience and make a guess at how much code needs changing. Plan for it, and plan conservatively. *DON'T* rewrite the application. Only fix the problem.
If you have time left on your initial x hours, start fixing the problem. See how it goes. Does it look like it's just going to keep on going like this until you fix it, or are you finding your being faced with odd and quirky behaviour left right and center? If you have weird behaviour, beware! If you have undocumented libraries/APIs behaving weirdly, beware! Don't get caught with your pants down. Let your manager know that you are passing some data down to the J2EE/COM/.NET/.Salsa/<NameYourCustomFramework> and that it is not acting as expected.
For actual development, the process is slightly different. Assess what needs to be done exactly in the same way as above (give an initial x hours to investigate). See what actually needs to be done. *DON'T* rewrite the STL library or .NET framework to do it. Choose the quickest cleanest path with the least amount of development time. Look at all your prior experience: have you previously used STL? Do you know it like the back of your palm? Have you only briefly used COM? And have you had issues with it the times you did use it? Keep that in mind. Try to steer the project in your domain of expertise. Put that as a coefficient in your estimation. If you know you have 100 lines or so of code to write using a library that you know in and out, estimate what you think it will take. If you will have to use a library that you're not really familiar with, pad it like crazy. I mean 2-3-4 times what you expect.
Those are the practical comments, there's also the more theoretical stuff:
Don't confuse accuracy and precision. And don't let your managers confuse the two. If I am asked to guestimate a project timeline, saying "4 months, 3 weeks, 2 days, 5 hours, 23 minutes" is more precise but radically less accurate than "4 months". Yeah, it sounds stupid, but accuracy != precision. Don't forget that. Tattoo it on your hand if you must. The longer a duration, the less precise it should be. Commiting to February 5th is ok if you are in January, it is not ok if you are in August. If you are in august, you must commit to a month coming up, and warn your managers to give it leeway. Just cause you said "it'll be done in february", doesn't mean they should plan a launch on February 18th.
The other thing is a little concept that I really like in comp sci algorithmics: divide and conqu
Go read The Personal Software Process. It's full of helpful, common sense suggestions that every programmer should know. Then go read The Team Software Process, which is a simple forward thinking way of getting programmers to work well together in teams. Then get really really depressed because no matter how hard you try at stuff like this, the software industry is so random and unpredictable that common sense is not all that common.
How we know is more important than what we know.
It is nearly impossible to predict the actual delivery time for a programming project because there are too many dependent variables. (For a good pciture of the problems caused by independent variables, read Gordratt's books, "The Goal", "It's Not Luck" and "Critical Chain". "Critical Chain" leads you to the best insights for your position, but the other books are essential prep.)
http://www.goldratt.com/
There are two keys to delivering almost always on time; Planning and Management.
Planning means thinking the project through in the best detail before you start coding. Agile programming seems to work in short, quick-and-dirty projects, but my experience is that it falls short if the project is large and complex. The less decisons you have to make in the middle of the project, and the less changes you have to make as a result, means that the work can me seen to progress along the planned path.
Management means controlling the variables, and that's where the techniques in "Critical Chain" shine. Nothing, absolutely nothing, is allowed to delay the critical chain in the plan. The biggest dividends come from two parctices: Continuity and Resource management.
It used to be that if I had a week to do a job that I knew I could do in a day or two, I'd do whatever else needed doing and start two days before the task was due. Of course, If something came up, (essential co-worker gets sick, computer crash, virus, software needs updates, whatever) then I might be delayed during that two days and the task is delivered late. If the task was on the critical path, then the whole project becomes late. The first 10% delay on a project typically costs 25% of the profitability, the next 10% late costs 20%, so being only 20% late can cost 45% of the profitability. People tend to think that if they have some slack, they can relax. Then they wonder why they are late. They should work as if there was no slack! Relax at the END, not during.
In a large project with multiple developers, some of them may be required to cordinate along different paths, not just the critical path. This is why Godratta called it the "Critical CHain". Those resources MUST be focussed on keeping the essential workflow from bogging down anywhere that may delay the longest chain of dependencies.
This idea came out years ago, and I've been following project management ever since the 60's. This is the first big change I've seen since 1969. Everyone who told me they tried the Critical Chain method and it didn't work, didn't follow the process. Siemens practically took the whole market for 7200 rpm hard drives by developing them first using the Critical Chain method.
Good luck on your projects.
"The mind works quicker than you think!"
http://en.wikipedia.org/wiki/Hofstadter's_law
When I can't reach a 80% to 90% confidence level that I can give an accurate estimate I do something different.
I don't give an estimate how long it will take to develop said feature. Instead, I give an estimate as to how long it will take for me the give the accurate estimate.
So, for example, I will say, I need 3 days to write some test-code, do some reverse engineering, and try to get part of the feature in, and in that time I am 90% confident that I will be able to give an accurate estimate then how long it will take to fully develop the said feature.
It has always worked very well for me to upper-management, as long as you then come in with a good educated answer how long it will take you after your first estimate.
It's better than giving a blind non-educated answer the first time around.
Modesty is one of life's greatest attributes
The reason why trying to estimate times is a useless exercise is that you can only get a reasonable exercise if you can break the problem down into tasks that resemble similar tasks that have been done before.
But... ever since the days in the 1950s when the subroutine was first invented, there never should be such a thing as a "similar task." Anything that was done before should have been packaged as a subroutine. Or a reusable object. Or a programming system (entire library or language) designed to attack that class of task.
Managers that prize predictability over productivity assign their people to do trivial variations of the same work over and over again and get a predictable straight-line curve of time versus work accomplished.
Good managers allow their people to invest time into tools and techniques are new, hence risky, but pay off with a multiplier effect when they work. They get an unpredictable but exponential progress curve.
(Bad managers, of course, won't accept the notion that a new task should be estimated at the time that a similar task did in the past. They insist that the task should be estimated as if everything were going to go perfectly this time, no missteps, no problems. ("It will take about a year." "Why do you say that?" "Because, remember, the last time we developed the frammis for the foithboinder project, it took a year" "Yes, but a lot of things went wrong. We had several false starts, because the original spec we were given was wrong. And Jim left halfway through the project and it took a while for Kathy up to speed. And we lost a month when we changed over to the new source code control system." "And why do you think nothing will go wrong this time?")
"How to Do Nothing," kids activities, back in print!
first one, of course, is: "How can i make a woman tick?"
i just don't devote any more time to these 2 questions anymore...
I don't feel like it...
I once worked for an "XP" (extreme programming, not windows) shop. We had not a single line of documentation in our entire codebase. What we did have, however, was 100% test coverage (at the unit tests level, the functional test levele, and even at the customer-test level). We could estimate, as a team, the length of time it would take to implement every feature in a major product change up to six months out, and hit them all on the exact day we had said we would.
It could have been the whole XP process or whatever, but personally I believe it was the testing that allowed us to estimate so well. When something is so well tested, and when you are working from a "test-first" mentality, you get the ability to estimate what you think it will be without having to fudge-factor the results. You do this for a while and you find that your gut instinct gets better and better.
As developers, we always have a tendency to think in terms of best case scenario. And once you voice that best case once, it gets very difficult to get out of a crazy commitment.
So *before* providing any numbers, make sure to list the assumptions that you are making, and the interferences and risks that may (and most likely will) occur:
- other projects and assignments that will steal time and resources
- spec changes and debugging time, customer interaction
Explain that you have an estimate in Effort-Days or Effort-Hours dedicated to the project, but that this is not the same as a Calendar-Day. With week-ends, holidays, admin tasks, etc, count 200-240 effort-days per year.
Explain that uncertainty remains on a number of tasks. You want to give your best-case estimate, but there are things that you don't know, but in practice best case NEVER happens. Provide a more likely estimate taking the risks into account.
Highlight the key risks and challenges being faced. Discuss ways in which those risks could be mitigated: does it make sense to create a first prototype (to check technical feasibility and/or get end-user feedback) ?
Or can hiring a consultant (that you would select) to help you with a specific technology (e.g. set-up that database back-end) help speed things up/reduce risk? (sometimes few hours here and there for a couple of days total is enough). Keep in mind that "throwing more resources" at the task often only worsen's a project's delay (training and communication overhead, etc).
Does not having a quiet and uninterrupted work environment interfere with your productivity? Point this out, make concrete suggestions for improvements.
MBA or not, your manager hopes to just get a number from you, and then bind you to it. Don't let this happen, but don't be confrontational either: you need to establish a teamwork.
It is not about you escaping any pressure from a commitment. But many interferences are out of your control, and Manager might be able to help keep those away.
It is in your mutual interest to have estimates that work. And this is also why you need to keep track of your progress, and notice deviations early on. Be dedicated, and act responsibly. Warn Manager of any delays and problems that occur, so you can rectify the estimated schedule (or the environment) together.
Hope this helps -Ivan
I wrote a tool called Mr Schedule that's based on the Painless Software Schedules technique described Joel Spolsky. I've found the technique works very well.
Cheers
Andrew
Creating new code is usually easier to estimate, because once you understand the business rules, you should be able to code it and estimate your time.
Bug fixes, on ANY code ( don't care who wrote it ) can be trickey. It can often be a situation that if you fix one thing your break something else.
There is no real set formula I use, but if I think it will take 3 weeks, I'll say 6. I always double my estimate, and often 'fish' for scope.
Lastly there is nothing wrong with saying 'I'll have to look into that to give you a more accurate estimate. You should be able to take a day or two to get into the code and look at it and figure out enough about it to get an estimate of what is involved. We do this and we call them ballparks. Our ballparks tell how much time the developer needs + qa + spec + IT time + some other things in there. It is usually an estimate.
Then we write a spec, and in the spec we uncover more. Depending on the outcome of the spec and what needs to be added or dropped the bid can change. Also in the spec we identify ALL the programs that need to be changed. These are things that should have been looked at in the ballpark to give you a better idea of what to extimate.
It sounds to me like you are a junior programmer. This has nothing to do with your coding skills and everything to do with your project skills. There are many people who can code circles around me, but they can't always give estimates that are as right on as me. As such, take a look at your code. This should be done in a day no longer. Figure out how much time you think you need to spend on it. Multiply by 3.
After you do this a few times you can start to see how good you are at estimating time. If you estimate time and you are behind on the project, then next time multiply by 4. If you end up ahead of schedule multiply by 2. After a few projects you will be able to say without a doubt exactly how long it will take. Unfortunately some projects will be a range not an exact date, like it will take 6 to 12 months, depending on scope changes.
really though it takes practice to estimate time. Knowing your own skill level, your knowledge of the language and code base, and what other things will always popup, and lastly your interpution level.
I'm sadly the goto guy at my company, so my interruption level is high.
Only 'flamers' flame!
Does slashdot hate my posts?
In my experience it's been S&M that pulls the numbers out of their asses, trying to make the sale, then management does their best to live up to it, pushing everyone under them for overtime work and scrambling to expedite delivery of materials where they're needed. I'm in the hardware end of things, but I expect things work the same way in software.
Do not confuse complexity with unknowability. A sulfuric acid manufacturing plant is complex, but every part of the process is known, and is known to be repeatable; and thus it is reasonable to ask how long it will take to build a proposed sulfuric acid manufacturing plant and how much it will cost.
Scrum is a development method for nontrivial software projects with many unknowns and little repeatability. Such projects cannot be estimated as if they were complex-but-known, no matter how strongly the project sponsors wish they could. The best that can be achieved is to control risk with short timeframe subprojects (no more than thirty days) and daily formal sessions (no more than 15 minutes) where the developers (seven plus or minus two) address only what they did yesterday, what they plan to do today, and any impediments. The best the requesters can do is to know that the developers are following a procedure that will surface problems early, tend to build the most useful parts of the system early, and provide frequent opportunities for continue / discontinue decisions during the process.
For small changes you can generally guesstimate from exerience with that system. If you know the code well there's a better chance of being accurate. If you have to wade through someone else's code you don't know then be very generous (generally double your guess).
With all but the most trivial changes you need something in writing to confirm what is wanted - even just an email to restate what you talked about and put your estimate in writing. Get in the habit of this. You can talk about it then look at the code for 30minutes and make a far better estimate than being pressuried into giving an answer straight away.
On much larger projects the time spent on actual coding gets less and less and the time spend on specifications, prototypes, testing, documenting, reviewing etc gets much bigger. I recently worked on a $250m sw project where coding was about 20-25% of the time. It depends on what the code is for. Normally coding is about 40% of a project but the more critical a system the less time spent on code and the more spent on design and testing. I've worked on some safety critical systems (ie where someone dies if your code is wrong) and coding was about 10% of the project. This might take a lot of joy out of being a programmer but it pays to get it right!
pithy comment
I can really recommend you the book "Software Estimation: Demystifying the Black Art" by Steve McConnell (the same guy from Code Complete).
He recommends that you create two or three estimates: the best case, the worst case and the expected case. Using three estimates will give you several advantages:
- you express the uncertainty in your estimate to your managers
- Using some simple statistics you can calculate a much more reliable overall estimate (the errors in individual estimates average out if you combine them into a larger estimate).
But most importantly, he claims that the single point estimate you and I have been giving is in fact the best-case estimate. If you read the book you'll see that it is guaranteed to be impossible to hit the targets we set based on those single-point best-case estimates.
The Scrum process requires meaningful-to-stakeholder progress every 30 days, not every 24 hours. The 30-day period is called a "sprint." What the stakeholders consider meaningful is completely up to them, so differences in the perceived value of fixing an awkward API in a library do not arise. Sounds like magic, but it's just process. The stakeholders and the development team agree before each sprint on what the increment of value (as measured by stakeholders) will be. For example, "we will achieve end to end throughput beginning with a 50 line dummy data input file and ending with three ugly but functional developer-specified reports. We will also meet with these end-users and prepare the first draft of their report requirements."
At the daily scrum, "why the hell are you reporting something to me that I don't understand" is out of order. The manager may ask only "What did you do, what are you going to do, and are there any impediments." These questions are to be addressed in concrete terms.
Dramatic simulation
Developer 1: "I reduced the argument count in the core library API and conformed the argument order and checked it into the mainline."
Manager: "From the looks on their faces, I believe other developers may have questions about that, so please have a work meeting after this scrum. Also, yesterday you said you were going to review the API, not change it. Please use these meetings to give advanced warning. Also, after the work meeting, will you please tell me what an API is?
Developer 2: "I created a warpo input file with known-bad items, including obsoletes, wrong currencies, randomly-scattered minus signs, a few 75,000 character fields and everything else my demented brain could come up with in an hour. By the way, the system now responds to negative zipcodes by rejecting them rather than freezing."
Manager: "Could you put that last sentence in correct form? We're trying to make the process visible, not just list benefits."
Developer: "Whoops, old habits die hard. 'I fed the warpo file to the system and observed the consequences, then changed the hyphen-parser logic in the zipcode field.' Today I will investigate why the system stopped compiling in my workspace after I integrated the latest changes from the mainline, starting with Mr. Speedy's core library changes. My impediment is that the security patch applied to the database server yesterday afternoon invalidated all the test passwords, and the system administrator is swamped with update requests."
Manager: "I will contact him and try to get you bumped to the head of the queue."
Finally, the only people permitted to speak at the daily meeting are the people who have committed to delivering. Others may attend, but must remain silent. The nontechnical stakeholders can satisfy themselves that their technical people are on speaking terms and possibly assist in removing impediments.
What you can report every day is your actions and your intended actions. "We drove 325 miles yesterday, 150 in the wrong direction, 150 back, then 25 ahead" is an action. Reporting that as "We progressed 25 miles yesterday," though less embarrassing, hides far too much. Checklist management - "what did you accomplish" - is appropriate for "defined control processes" such as taking inventory in a warehouse or assembling several floors fu
Hahahaha! Good one!
Management also often has information you may not haveHoo-haa! You're on a roll!
he/she isn't likely to be around much longerStop, please, you're killing me!
Proud neuron in the Slashdot hivemind since 2002.
I have been in the industry over 20 years and this is one of the more difficult questions to answer. For me, having worked in both development and management capacities, the most accurate answer that can be provided is "it depends".
Primarily, it depends on what you know. There are a number of techniques beyond the W.A.G. of developer's experience with the software at hand (a google search for "estimation techniques" turns up over 900,000 matches), and depending upon the circumstances one technique may be better suited than another. IMHO they all boil down to one simple fact: the more you know the more accurate an estimate can be. As a manager, you develop a feel for which of your developers pad their estimates and by how much. When the numbers seem high, it is usually because they don't have enough information to make a more reasonable "guess". As a developer you don't want to get caught short when your "rough estimate" turns into a schedule deadline, so you take into account all the things that are likely to come up.
For me, the best way to stay out of trouble is to offer a range in your estimates. The less you know the wider the range and vice versa. You can always qualify the range estimate based upon what you know. Other posts have suggested taking time to "get back to you with that estimate" and that is a very valid technique to "know more", thus reducing the range of your estimate. If you have reasonable management and business users, they will understand that an estimate is a "best guess given the information available" rather than a due date to be slapped on a schedule.
How to come up with accurate estimates, isn't an easy question to answer. If it were, there would be one or two simple ways to do it accurately, and everyone would be following those procedures to come up with estimates.
FWIW...
I've done the math, I know the odds, but I'm still disappointed when I don't win the lottery.
I read a bit about Scrum and realized I misunderstood the role of stakeholders at daily meetings, specifically that the meetings are optional for stakeholders and that stakeholders attend primarily to gauge the team's morale, check for dysfunctionality, and discover ways to help. (I see now that you mentioned that, but it slipped by me.) That makes much more sense to me.
I'm still leery of the distorting effect of reporting, though. If a manager's credibility or ability to act is negatively affected by his developers' daily reports, you can be sure there will be pressure to stop hindering his work. As a developer, I appreciate whatever gives me the widest scope to tackle a problem directly and honestly, and having day-to-day results disseminated beyond technical personnel has always had the opposite effect.
One advantage of traditional project management is that it gives techies a layer of insulation from management politics; they can be completely honest with their manager and trust him to suppress any information that could be used to sabotage the project. Many developers fail to demonstrate the most basic level of political competence even when they spend all day thinking about it. I include myself in that bunch. If it comes up, I tell my managers that any plan requiring me to be politically sensitive is a bad plan.
Perhaps Scrum is just poorly suited to my particular strengths and weaknesses, but I consider myself a pretty typical developer. I value my career over any company's success, I prefer complete honesty because I have never succeeded with any other strategy, and when management turns on bully mode, I just do what they want and call it their loss. That isn't to say I don't value my integrity, merely that I protect my integrity by choosing my working environments very carefully, not by martyring myself if I end up in a bad situation. Putting guys like me out there in front of assorted stakeholders with many people's credibility dependent on our statements is liable to bring all kinds of pressure on us to account for political consequences when we do technical work, and we will not stand up to it very well. I imagine Scrum being of value only in businesses that can depend on esprit de corps to trump politics.
My old software manager came up with an outstanding way of trying to estimate project completion for trying to tell upper management when something would be done.
He asked each of the software engineers how long it would take, based on their "gut" instincts.
Then he multiplied the figure by two and moved it to the next higher order of time. I.E. weeks become months, months become years. One time he asked us to estimate when we could get something accomplished and we told him two years. He cried for a little bit and went to upper management and said "Never, we can't get that done".
Surprisingly this turned out to be a very effective way for him to gague when things would actually get accomplished, and more often than not he was dead on correct. He also didn't let me in on his secret until just before he quit, but it has also helped me to make my own estimates for customers when I'm doing contract work now. I do the same thing where I make a preliminary estimate based on known issues and then apply the above rule to my own guess. I have (fortunately) been fairly accurate with my guesses at that point, even if it makes the customer angry with them insisting that it must be done in a shorter period of time.
30 seconds means 15 minutes.
10 minutes means 5 hours.
12 hours means 6 days.
14 days means 7 weeks.
8 weeks means 4 months.
6 months means 3 years.
* First Rule of Project Management: Divide estimate by two, and round up to next higher unit.