When Can We Ship?? A Fun Exercise in Estimating Software Tasks
Alicia Chen17:472020
A 6.5-hour paint job turns into 14.5 hours once you ask the right questions
Product managers at Dropbox, the story goes, swapped notes on each engineer's multiplier: if John says a week, budget two; if Joe says a week, budget three. Alicia Chen, an early Dropbox engineer now Head of Engineering at CoProcure, was the one engineer a PM credited with a multiplier of one. When a missed date means angry users, sales and an exhausted team, what does she do before naming a ship date?
Learn the pre-estimate steps, buffers and pushback that make your ship dates hold
Alicia Chen, Head of Engineering at CoProcure, uses a room-painting exercise to show why software estimates are so often wrong. She argues that research, requirement confirmation and cheap prototypes must come before any date is given, that effort must be converted into elapsed time with multipliers and buffers, and that engineers need the courage to challenge unrealistic deadlines.
“You tell them, I don't know, but I'll be able to tell you in a few days.”
“And a cheap prototype can save you a lot of time.”
What you will learn
Treat research, requirement confirmation and prototyping as pre-estimate steps before committing to any date.
Use cheap prototypes to uncover hidden work, like the prep and extra coats that doubled the room-painting estimate.
Convert effort into elapsed time by accounting for meetings, vacation, outages and other interruptions with a multiplier and buffer.
Calibrate your estimating by measuring actual time on small and medium tasks and working up to larger ones.
Push back with evidence when an imposed deadline conflicts with your estimate instead of revising it downward under pressure.
Details
Level
Beginner
Format
Talk
Technical depth
Introductory technical
For
Early career, Mid-career, Senior individual contributor
Length
17:47
Published
June 2020
Questions this talk answers
Why are software estimates so often wrong?
People are over-optimistic and give a date before they understand the spec, the requirements and the hidden work. Experience helps, but software projects are rarely done exactly the same way twice.
What should you do before giving an estimate?
Research existing solutions, confirm requirements with the decision makers and build a cheap prototype. If asked for a date first, say you will have an answer in a few days and budget that time.
What is the difference between how long a task takes and when it can ship?
The first is the time strictly required to do the work; the second is elapsed time, which includes meetings, vacation, other projects and site outages. She accounts for the gap with a personal multiplier and buffer sized to the project's complexity and risk.
How can you get better at estimating?
Estimate small to medium tasks, measure the actual elapsed time, reflect on the gap and gradually work up to larger tasks. She points to Joel Spolsky's article on Evidence Based Scheduling for detail.
What should you do if someone else already set the deadline?
Do the estimating work yourself, and if your estimate differs sharply from the timeline, speak up. Do not revise your estimates downward because of pressure.
About this talk
Do you have trouble giving estimates for how long your software projects will take? Are you frustrated by people in your organization setting unreasonable timelines for delivering projects? This exercise (audience participation encouraged) will help demonstrate why estimation is hard, how you can get better at it, and how you can manage expectations related to deadlines.
Don't name a date until you've done the research (58 s)
Speak up when the deadline is wrong (88 s)
Just a quick self intro. My name is Alicia and I started off my career at Microsoft where I worked on a road desktop working on things like display drivers and networking. I then moved on to Dropbox where I was one of the earlier engineers there. I worked on all kinds of things from the diversity initiative to the intern program to my actual job, which was a software engineer on the desktop client and eventually manager.
Speaker: At some point I left Dropbox feeling like I wanted to try my hand at having some more social impact than just working at your garden variety tech company. And so I left and landed at the Chan Zuckerberg Initiative, which is Mark and Priscilla's foundation. I spent a couple of years there working on criminal justice reform, housing affordability, economic opportunity, education, lots of interesting stuff.
Speaker: Um, and then most recently I've been at CoProcure, um, which is a startup that is trying to help the government more efficiently procure goods and services, as well as help small businesses, small and diverse businesses, um, sell into government and get access to that $1.5 trillion a year that local governments are spending. Speaker: Um, I think this is not going to take up the whole 40 minutes. So, uh, if you have questions other than this talk at the end, I love talking about, uh, technology for social impact, hiring and managing software engineers, um, cats and knitting, diversity in the tech sector and low bow, low brow fantasy fiction. So just priming your Q and A.
Speaker 1: So onto the actual talk at hand. Um, when I was at Dropbox, there was a product manager training and, um, I heard rumors that at that training, the PMs were swapping notes about every engineer's multiplier. Speaker 1: That is to say, if John, the engineer tells you that it's going to take a week to get this thing done, you pencil down two weeks. And if Joe tells you it's going to take a week, you better budget three weeks.
Speaker 1: Uh, and to toot my own horn a little bit. Um, this PM told me that I was the only engineer at the time who was, who was volunteered as having a multiplier of one. And could I please give a tech talk about estimating software tasks? Speaker 1: So five years later, here you all are sitting at the very first session of this talk.
Speaker 1: So if you are a software engineer or, um, have worked with software engineers, you've probably heard the question. Speaker 1: So when can we ship this thing? And, uh, almost everyone is really bad at answering this question, uh, famously bad at answering this question. Speaker 1: Now, you know, sometimes you think, well, you know, we're super agile or they'll ship when we ship. Like, why does it matter? But oftentimes there's a need for a real hard ship date.
Speaker 1: Um, sometimes you have an externally announced date, uh, comes with a press event or you're coordinating multiple teams shipping one large project. Speaker 1: And so you need to have a single date that everyone agrees on. Speaker 1: And, uh, what happens when you're wrong? Uh, typically it's that someone is over optimistic and says they can ship sooner than they can.
Speaker 1: Sometimes it goes the other way, uh, where your estimate is too far out, but that usually just results in scope creep. Speaker 1: That's a topic for another day. Um, but if you're over optimistic on how long it's going to take, uh, what happens on launch day is that your manager was really annoyed at you because you made her look pretty bad.
Speaker 1: Uh, the sales team is pretty annoyed at you because they are breaking all of their promises to their clients. Speaker 1: And your team is pretty mad because they've been working nights and weekends for a few months straight trying to meet this impossible deadline. Speaker 1: And all of your users are really mad at you because you delivered your product late or you delivered a really crappy product.
Speaker 1: Probably you did both. Speaker 1: So the next question is why is estimating so hard? Speaker 1: And to get at this question, we're going to do a little exercise that I have shamelessly stolen from an intro to development class that I took at Microsoft of how long does it take to paint a room? Speaker 1: Uh, the nice thing about this is that it has nothing to do with any specific software system.
Speaker 1: So you don't need a lot of context to do this, but I hope that you all can, uh, can join me in this exercise because it's a bit of audience participation. Speaker 1: So if you can, um, please get out a notepad digital notepads fine and take a little bit of time to think about your estimate for how long it would take to paint a room.
Speaker 1: So just do, do, do, do, do, do. Speaker 1: I'm not going to sing the whole time. Speaker 1: Yes, I see some people are getting at the problem here. Speaker 1: Yeah, so you're done with your estimate. Speaker 1: If you already have an answer that you're confident in, then you're a novice estimator because we haven't even talked about the spec yet.
Speaker 1: So, uh, let's say that the room is 12 feet by 10 feet by 10 feet. Speaker 1: So that's like an average bedroom and we're painting it off white. Speaker 1: So are we off to the races? Speaker 1: Yeah. Speaker 1: 21 year old me was like, yeah, I got everything. Speaker 1: Let's go.
Speaker 1: No, we don't have everything. Speaker 1: You must buy all of the materials. Speaker 1: So don't assume that you have everything that you need. Speaker 1: The room is fully furnished and currently being lived in. Speaker 1: It has one door, three windows, and the current color of the room is Kermit green. Speaker 1: So now we're going to do this thing.
Speaker 1: Take a minute, a little time, one minute, and really hold yourself to writing it down. Speaker 1: So let's talk a little bit about how far we got. Speaker 1: So let's suppose it looks like this. Speaker 1: Buy the stuff, prep the room, paint the room, clean up, right? Speaker 1: So two hours to get to the store and back, some prep, painting, clean up.
Speaker 1: Yeah. Speaker 1: So six and a half hours. Speaker 1: Is that right? Nope. Speaker 1: We need a prototype before we can realistically do this, especially if you're announcing a date Speaker 1: That others are depending on, you need a much higher level of confidence than just putting something out there. Speaker 1: So first you research.
Speaker 1: You're going to Google, how do you paint a room? Speaker 1: Learn about all the different types of paint. Speaker 1: Like what is primer? Speaker 1: Are you using rollers? Speaker 1: Are you using brushes? Speaker 1: Do you know what you call the things that you lay down on the ground so that the paint doesn't drip on your carpet?
Speaker 1: And then confirm the requirements. Speaker 1: Off white is not an actual paint color. Speaker 1: There are many, many different types of paints and brands. Speaker 1: And so if you're not the final decision maker, then you've got to go decide, like get everybody on the same page. Speaker 1: Also, are you painting all of the walls or just an accent wall?
Speaker 1: What color are you painting the trim, if anything? Speaker 1: And do you know what eggshell and satin refer to? Speaker 1: Those are different paint glosses. Speaker 1: There's at least six different paint glosses. Speaker 1: So you had better also confirm that. Speaker 1: You didn't even know paint glosses existed like before you Googled it.
Speaker 1: Now, the next step we're going to kind of skip over because it's not that relevant to software engineering, but obviously you would have to go buy the stuff. Speaker 1: Hopefully you made a good shopping list. Speaker 1: Otherwise, you're going back and forth to the store a few times. Speaker 1: That's also maybe part of your estimate.
Speaker 1: And then prototype. Speaker 1: So actually prepare a small area of your wall, apply the paint, let it dry. Speaker 1: And then if you're not satisfied, apply some primer. Speaker 1: Hopefully you've managed to buy a sample of that, let it dry, apply some paint over that. Speaker 1: And so after you've done your prototype, you've probably learned, oh, we've got to prime and then paint.
Speaker 1: And in fact paint a couple of coats because your wall is Kermit green and you lay on one layer of white paint over that you're going to have just slightly less Kermit green walls. Speaker 1: Even four coats probably still going to be kind of green. Speaker 1: And then the other thing you learned is that prepping to paint takes a really long time.
Speaker 1: If you've ever thought about what it takes to turn a room from this to this, there's a lot of shuffling furniture around, probably needing two people to get the bigger items out, taping and taping. Speaker 1: I mean, I don't know if you realize that like taking down the paintings off of the wall would leave holes that you're going to have to patch and that you have to like unscrew all the light switch and outlet covers.
Speaker 1: So your first estimate probably didn't account for that if you've never painted before or if you didn't do a prototype. Speaker 1: So let's go to back to software for a second. Speaker 1: Research improves your design. Speaker 1: Don't reinvent the wheel if you have a problem that needs to be solved. Speaker 1: Are there off the shelf solutions that already exist either inside your company or in the community that you can just use?
Speaker 1: How is everyone else doing it? Speaker 1: Secondly, prototype improves your estimate. Speaker 1: And a cheap prototype can save you a lot of time. Speaker 1: The number of times that I've done a project and the sequence of events went like this. Speaker 1: First, I'm like, oh, yeah, we're going to build it and we're going to build it like this.
Speaker 1: And then I do a little research and I'm like, oh, somebody else already built it. Excellent. Speaker 1: Then I prototype integrating that library and it's like, oh, somebody else built it not quite right for our purposes. Speaker 1: So we're going to have to modify it or build it ourselves after all. Speaker 1: And if your prototype was super cheap, didn't integrate with any real production systems, then you saved yourself a lot of time chasing down all of those dead ends.
Speaker 1: And so it will get you to your final product faster. Speaker 1: And of course, estimates improve with experience. Speaker 1: If you start up a website before, you know how long it takes to stand up a website. Speaker 1: So, you know. Speaker 1: Okay, so back to the real estimate. Speaker 1: Now that we've done the prototype and the research, we have an idea of how long this is actually going to take.
Speaker 1: We're prepping the room, which means clearing all the walls, getting the furniture away from the walls, washing the walls, filling in any of the holes and sanding them, taping down all of the edges and trim and floors and stuff and securing drop cloth. Speaker 1: Then we got to prime the room. So mix the primer, apply the primer, wash all the rollers and brushes so you can then apply the paint.
Speaker 1: Pour and mix the paint, apply one coat, let it dry, apply second coat. Based on your prototype, you know how many coats you needed. Speaker 1: And then you got to wash the rollers and brushes. Speaker 1: Yes, I can see some of you have repainted before. Speaker 1: Once you've done this multiple times, you get really good at it.
Speaker 1: But it's very rare that you get to do a software project multiple times. Speaker 1: It's rare that you ever have to do something exactly the same twice because that's the whole point of software. Speaker 1: So that adds up to 14 hours and 30 minutes, which is a lot longer than our initial estimate of six and a half hours.
Speaker 1: In fact, looking back on this estimate, which is the estimate that I gave the first time I did this exercise at Microsoft, Speaker 1: It's actually pretty embarrassing. Speaker 1: You feel like your past self was a little bit burp burp burp burp. Speaker 1: And so this also means that the research, the requirements, like going back and forth on the scope, Speaker 1: and doing at least enough of the prototype that you're confident on what the design of the project is going to look like.
Speaker 1: These are pre estimate steps. Speaker 1: So don't even give a date until you've done these steps because you're going to look pretty dumb. Speaker 1: If your manager asks you the first time, you're like six weeks. Speaker 1: And then you turn right around and you go, never mind, just kidding, 14 weeks. Speaker 1: It's a great way to lose credibility.
Speaker 1: And yeah, doing this work takes time. Speaker 1: The research takes time. Speaker 1: The prototype takes time. Speaker 1: So if someone shows up at your desk and asks you, when can we ship this? Speaker 1: You tell them, I don't know, but I'll be able to tell you in a few days. Speaker 1: And you have to budget those few days into your work plan also.
Speaker 1: All right, so 14 and a half hours. Is that our final answer? Speaker 1: Can we get painting? Speaker 1: No, still can't get painting because the question we answered was how long will it take? Speaker 1: Strictly the time required to do the thing. Speaker 1: But the answer we're actually looking for is when can we ship?
Speaker 1: And the difference between those two is that the time elapsed includes things like meetings, vacation, other projects, dealing with site outages, checking Instagram, everything that happens during the course of a day. Speaker 1: And so how do you actually account for all of those things, some of which are pretty unpredictable? Speaker 1: There's lots of ways of dealing with this.
Speaker 1: Funny, we come back to applying a multiplier. Speaker 1: Every person has a different multiplier, so you can always try to take that into account beforehand before you give it to the PM. Speaker 1: You can also add buffer time for unexpected things like, you know, forgetting to buy a ladder, having to go back to the store.
Speaker 1: You can also measure and improve and continually get better. So start by measuring start by estimating small to medium sized tasks and projects and then measuring the actual time that elapses from the beginning to the end of doing that task. Speaker 1: And then reflect on how good your estimate was and work your way up to gradually larger tasks.
Speaker 1: There's actually a really amazing and hilarious article by Joel Spolsky on this called Evidence Based Scheduling that gets into extreme detail on how to do this, so I won't cover it. Speaker 1: If you read the article, it does seem like a lot of work. Speaker 1: It is kind of a fair amount of work, but like any skill that's worth developing, it's worth putting some time into.
Speaker 1: Now, all of this is assuming that somebody even bothered to ask you for the estimates. Speaker 1: That doesn't always happen. Speaker 1: So what if the estimates aren't yours? Speaker 1: What if your tech lead just kind of offhandedly throw out, oh, yeah, it's going to be about three weeks or your CEO tweeted to the entire world that it's going to be available by the end of the year?
Speaker 1: And yes, Jennifer makes an excellent point, which is your buffer should be relative to the overall size of the project and the relative complexity and risk. Speaker 1: Anyway, if you're on the hook for delivering this project, even though somebody else has provided the timeline for it, what can you do? Speaker 1: Well, especially if the stakes are really high, that delivering on time is really important, then first you got to do the work of estimating.
Speaker 1: As we just described, it's actually kind of a lot of work. Speaker 1: And then once you have the actual more correct estimates, including buffering, including multiplier, all of that stuff. Speaker 1: Then if the if your estimate is really different from the timeline that was provided, you need to speak up. Speaker 1: And this can be really hard, especially when inside the company, everyone's like, hey, Project Unicorn and Operation Rainbow are going to launch by Christmas.
Speaker 1: Everything is going to be wonderful. Speaker 1: What do you mean we can't do it in time? Speaker 1: Your negative attitude is what is holding you back from being successful in this company. Speaker 1: That's just not true. Speaker 1: What you're actually doing is trying to prevent your team or company from promising this and then in the end shipping this.
Speaker 1: So remember, what's on the line is everybody in the entire world being angry at you. Speaker 1: So don't let yourself be intimidated into science. Speaker 1: Don't revisit your estimates and revise them downwards because of pressure. Speaker 1: You know, it's easy to fool yourself into thinking, well, you know, just paint a little bit faster.
Speaker 1: It doesn't work like that. Speaker 1: It takes a lot of work to get really good at estimating. Speaker 1: It takes a fair amount of work to even just create one estimate. Speaker 1: And it takes a lot of courage to stand up and say, you're wrong. Speaker 1: We're not going to be able to ship that set of features on that timeline.
Speaker 1: But it is really important to do that because if you do it right, then everybody can be happy instead. Speaker 1: Yeah. Speaker 1: And again, Jen makes some excellent points. Speaker 1: Creating a good schedule and a very aggressive schedule with good estimates will force people to have real discussions about what's important and what's not important.
Speaker 1: This is addressed pretty well in Joel's article about about evidence based scheduling. Speaker 1: But prioritization is a whole nother set of topics that I could also chat about for a long time. Speaker 1: But anyway, that concludes this presentation. Speaker 1: Thank you.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
Why your software estimates are wrong, and how to give ones that hold
Alicia Chen uses a room-painting exercise to show the hidden work behind every ship date and why engineers must defend honest estimates.
In short
Software estimates fail mostly because people name a date before they understand the work. Alicia Chen, Head of Engineering at CoProcure, shows through a simple room-painting exercise that research, requirement checks and a cheap prototype must come first, that effort must be converted into elapsed time, and that engineers need the courage to challenge deadlines that cannot be met.
Key takeaways
Treat research, requirement confirmation and prototyping as pre-estimate steps before committing to any date.
Use cheap prototypes to uncover hidden work, like the prep and extra coats that doubled the room-painting estimate.
Convert effort into elapsed time by accounting for meetings, vacation, outages and other interruptions with a multiplier and buffer.
Calibrate your estimating by measuring actual time on small and medium tasks and working up to larger ones.
Push back with evidence when an imposed deadline conflicts with your estimate instead of revising it downward under pressure.
The multiplier nobody talks about
Chen opens with a story from her time as an early engineer at Dropbox. She heard that at a product manager training, PMs were comparing notes on how much to pad each engineer's estimates. As she puts it: "That is to say, if John, the engineer tells you that it's going to take a week to get this thing done, you pencil down two weeks." (1:39) One PM told her she was the only engineer volunteered as having a multiplier of one, and asked her to teach others how she did it.
Why does it matter? Even agile teams sometimes need a hard date: an external announcement, a press event, or several teams shipping one large project. When the estimate is over-optimistic, the fallout spreads. Managers look bad, sales breaks promises to clients, and users get a late or poor product. Chen is blunt about the human cost: "And your team is pretty mad because they've been working nights and weekends for a few months straight trying to meet this impossible deadline." (3:45)
How long does it take to paint a room?
To show why estimating is hard without needing any software context, Chen borrows an exercise from an introductory development class she took at Microsoft. She asks the audience to estimate how long it takes to paint a room, and immediately warns against confidence: "If you already have an answer that you're confident in, then you're a novice estimator because we haven't even talked about the spec yet." (4:52)
The spec then unfolds. The room is 12 by 10 by 10 feet, to be painted off white. You must buy all the materials. The room is furnished and lived in, with one door, three windows and walls currently in Kermit green. A typical first plan (buy supplies, prep, paint, clean up) comes to six and a half hours.
That plan, Chen explains, is missing a prototype. Research reveals primer, rollers versus brushes and drop cloths. Confirming requirements shows that off white is not a real colour, that trim and accent walls need decisions, and that there are at least six paint glosses. A test patch on the wall shows you need primer and several coats over green, and that prep (moving furniture, patching holes, removing outlet covers, taping) takes far longer than expected. The revised estimate is 14 hours and 30 minutes.
Research improves design, prototypes improve estimates
Back in software terms, Chen draws two lessons. Research stops you reinventing the wheel by finding existing solutions inside your company or in the community. Prototyping tests whether those solutions really fit. She describes a familiar sequence: plan to build it, discover someone already built it, prototype the integration, and find it does not quite suit your needs. Her advice is simple: "And a cheap prototype can save you a lot of time." (9:20)
Experience improves estimates too, but with a catch that makes software especially hard: "But it's very rare that you get to do a software project multiple times." (11:09)
Don't name a date too early
Because research, requirements and enough prototyping to settle the design all change the answer, Chen treats them as pre-estimate steps. "So don't even give a date until you've done these steps because you're going to look pretty dumb." (11:58) Saying six weeks and then fourteen weeks is, in her words, a great way to lose credibility. When asked for a ship date on the spot, she recommends saying you will know in a few days, and budgeting those days into your plan.
From effort to a real ship date
Even 14 and a half hours is not the final answer, because the question was never only how long the work takes. "And the difference between those two is that the time elapsed includes things like meetings, vacation, other projects, dealing with site outages, checking Instagram, everything that happens during the course of a day." (12:56)
Chen offers three ways to close the gap: apply your own multiplier before handing the estimate to a PM, add buffer for the unexpected (sized, as an audience member noted, to the project's size, complexity and risk), and measure and improve. Start with small and medium tasks, record the actual elapsed time, reflect, and work up. She recommends Joel Spolsky's article on Evidence Based Scheduling for the detailed method.
When the deadline isn't yours
Often nobody asks for an estimate at all: a tech lead throws out three weeks, or a CEO announces a date publicly. If you are on the hook, Chen says, do the estimating work anyway, and if the gap is large, speak up even when everyone around you is excited about launching by Christmas. "Don't revisit your estimates and revise them downwards because of pressure." (16:12)
She acknowledges this is hard: "And it takes a lot of courage to stand up and say, you're wrong." (16:31) Done well, honest estimates force real conversations about what matters and what can be cut, so everyone ends up happier.
Watch the talk
"When Can We Ship?? A Fun Exercise in Estimating Software Tasks" by Alicia Chen, 18 minutes: https://www.womentech.net/video/alicia-chen-when-can-we-ship-fun-exercise-in-estimating-software-tasks
About the speaker
Alicia Chen is Head of Engineering at CoProcure, which she co-founded as a marketplace for local government purchasing. She was an early engineer at Dropbox, where her projects included diversity and the intern program, and later worked at the Chan Zuckerberg Initiative on economic opportunity, criminal justice reform and education.
Questions this talk answers
Why are software estimates so often wrong?
What should you do before giving an estimate?
What is the difference between how long a task takes and when it can ship?
How can you get better at estimating?
What should you do if someone else already set the deadline?
Drafted with AI from the talk transcript. Editor review pending.
Frameworks and ideas mentioned
Evidence Based Scheduling
Estimate multiplier
Resources mentioned
Evidence Based Scheduling (article by Joel Spolsky)
Cite this talk
Alicia Chen. "When Can We Ship?? A Fun Exercise in Estimating Software Tasks." WomenTech Network, 2020. https://www.womentech.net/video/alicia-chen-when-can-we-ship-fun-exercise-in-estimating-software-tasks