The article describes a predictive project-management approach, and it's the same misguided focus on deadlines you hear from people with a predictive mindset. But an adaptive approach is so much more effective.
The adaptive approach: Products should ship as soon as they're ready for market, not according to some deadline. To ship sooner, they should be sliced into pieces that have compelling individual value. Each piece is shipped when ready. And those pieces ("valuable increments") should be sliced into even more pieces that could be shipped sooner in case new business opportunities come along. That way, when that new business opportunity comes along, you're not left with a choice between abandoning work or ignoring the opportunity in favor of finishing what you started.
There are times when deadlines matter, such as regulatory requirements, tax season, the Christmas shopping season... but people who are "setting deadlines" aren't talking about real deadlines; they're talking about artificial deadlines because they think it's the right way to manage software teams. (They don't understand that managing by deadline causes teams to compromise internal quality, and that hurts productivity, which creates stress, which forms a vicious cycle that further harms quality and productivity.)
If you do have a real deadline, the correct way to manage the risk of missing that deadline is to, once again, use an adaptive approach: identify the smallest thing that will mitigate the risk, build that first, and then incrementally expand from there in as small pieces as possible. Keep the software shippable at every step of the way and ship on the deadline with whatever you finished.
This is a good argument for continuous release and agile over waterfall.
When making estimates, I also like to differentiate between what I call happy-path projections and lazy-bayesian estimates. The former take a sort of get-there-from-here best-case-scenario approach with some buffer tossed in to make it seem like we're not completely delusional. Lazy-bayesian estimates try to identify similar projects or references as a baseline for forecasting.
Daniel Kahnemann refers to the two approaches as the inside and outside view in Thinking Fast and Slow and offers an excellent tale to illustrate the difference:
The problem with the adaptive model is in creating the culture inside the company to work with it. It might be easy if you are a software company, but for those of us where software is only a part of what the company does, it can be really difficult to get that mindset. Marketing want to know when the release is going to happen, the MD wants to get to market with this new thing that will do x for y (but won't really)....
I believe this is the main source of stress in my life! I have tried everything and I win a bit, some of the time. Despite all of this all I really gained was pushing the launch date back a bit. I have manouvered some other departments into the firing line so I don't get all of the flack. I have set the most demanding stakeholders up to argue their demands for new features in a way that makes them look bad for delaying the project. Sometimes I think we take a step towards an agile organisation, then other times we are back at square one
Also, this might work for a standalone product. But if the product integrates with other systems the client is already using, it becomes extremely difficult to ship something of which only certain parts work.
In this case, the client, usually and in my opinion, correctly, expects the product to fulfill all of the functionality planned for.
While I prefer working that way, I personally know for myself that when a deadline is coming I get more done closer to the deadline. I don’t think it’s one or the other, but a mix of both.
The adaptive approach: Products should ship as soon as they're ready for market, not according to some deadline. To ship sooner, they should be sliced into pieces that have compelling individual value. Each piece is shipped when ready. And those pieces ("valuable increments") should be sliced into even more pieces that could be shipped sooner in case new business opportunities come along.
I used to wholeheartedly agree with this view, but I've moderated my position considerably. What I've found is that, in practice, this encourages a mindset where a team (or even an entire company) delivers a lot of "half-assed" products or features. They make a minimally viable thing, ship it, and then move on to the next thing. They never come back to actually complete all the fit and finish work.
A great example of this is the "Familiar Faces" feature of Nest security cameras. Nest security cameras have a feature where you can train them to recognize people in your household. Then, when they alert, instead of sending a generic alert, they tell you which person was seen. This is great, but Nest never actually finished the feature. There is no setting which tells the camera to only notify when there's an unfamiliar face. You can get no notifications at all. Or you can get a notification every time the camera notices activity. But there's no way to tell the camera, "Hey, let me know if you see a stranger," despite the camera having the ability to distinguish strangers from friends and family.
Alphabet products, in general, suffer from this mindset. Another example is the new Suggested Audio feature of Android Auto [1]. I can tap on the icon and it'll bring up a random selection of podcasts from my (now defunct) Google Podcasts subscriptions. Is there any way to customize which podcasts appear? No. Is there any way to make the icon go away, because I use a different podcast app? No. It's just there, waiting for my finger to accidentally brush against it as I try to fast forward in PocketCasts. They made a minimally viable thing, shipped it, and moved on.
I contrast this with Apple, or even Microsoft, where there seems to be a lot more focus on having products that are at least somewhat thought out from the perspective of the user, and not shipping until there is certain level of coherence. Alphabet, on the other hand, ships features as soon as they're ready, but doesn't seem to realize that on buttons are useless without their corresponding off buttons.
I agree that half-assery is a failure mode of the adaptive approach. I don't think it's an inherent problem with the approach, though, just a common mistake people make. You're supposed to split your product into pieces with compelling value. But often people misinterpret ideas such as "minimum viable product" as splitting the product into pieces with minimum value. And that's a mistake.
(Minimum viable product, of course, not being about products at all, but about product-market fit and testing your ideas as soon as possible using low-fidelity experiments.)
You can have a coherent user experience and still use an adaptive approach. I think Alphabet's failures have more to do with Alphabet's culture, particularly around rewards and promotions, and perhaps a disdain for non-engineering skills, rather than taking an adaptive approach.
Half-assed happens in both scenarios. If you have something shippable by the pre-ordained date or shortly thereafter, it ships. Whether you use an adaptive or predictive approach, the outcome is the same that further improvements may or may not happen.
The advantage of the predictive approach is better driving of your workers with a deadline. The disadvantage is low morale over time and lower quality products under the hood. Rotating workers who come in fresh can mitigate this outcome but at a cost.
Ironically you can't agile a cake, because the whole thing has to go in the oven at once for a defined time period.
Or maybe you could; if management can't or won't specify the cake in advance, say whether it should be round or square until ten minutes before the deadline, you could just bake a pile of cupcakes and squash them into the tin.
I'd say in my experience both deadlines and estimates are garbage 90% of the time.
What really matters is prioritization. Are you focusing on the right things, the impactful things? Are you also prioritizing all the 'not-so-shiny' stuff (performance, security, etc.)? Do you have actual levels of priority (instead of just making everything top priority, which is no priority at all)?
If you prioritize well, then you'll always be delivering the top priorities and some of the mid-level priorities. On a good week, you'll also knock out some of the low-priority stuff. And all of the concerns will be getting addressed. And everyone will be happy. The developers will feel good, management will be happy, and the customers will be pleased. Deadlines or estimations will be largely irrelevant.
But if you prioritize sloppily, and constantly change priorities, so you're always yanking people out of the middle of one thing to work on another that is suddenly the highest high priority among all the other tasks all of which are high priority, then you'll always get less done. Some random assortment of stuff will get done, but it doesn't look important, at least not any more important than the stuff that didn't get done. No one will be happy. Some concerns will go entirely un-addressed. And no amount of deadlines or estimation will fix that.
I'd much rather be working for a good prioritizer than the world's best psychic estimator and the world's best deadline forecaster anyday.
At my last employer (very large, not FAANG) every project started with a deadline for when it had to be done, but few details on what was being asked for, followed by detailed estimates being required that were then only used for budgeting and not how hard it was to meet the deadline. Then inevitably all the details changed daily until the deadline was missed and recast again. Rinse, repeat, until eventually the project was canceled. All of this because the company had too many execs who wanted in on expensive projects, a product team with no ability to say no, and thus the development teams worked long hours to no avail. Shipping small incremental versions was considered anathema. So one could work on a project for 18 months and ship nothing only to start another project that also didn't ship.
You're gonna get more people aligned if you focus on things that other teams happen to be working on at the same time. If you decide to work on something else when there is a huge infrastrucuture design project going on, just because of some precommitted sprint plan, you're gonna play catch up later.
If you keep dropping a task for another because of arbitrary sprint plan, you're gonna be jerking yourself and the team around and never getting people get aligned with what you're building for long enough and it'll ge harder to get feedback from them.
Momentum matters, especially for multi product teams, such as in data science and data engineering.
I agree, deadlines and estimates are used to cover up for product actually not having a solid use case and design in the first place.
Which is also not just blame the people that work as product managers, but also a critique of Agile swerving too far away from waterfall and now having no method whatsoever for requirement analysis and leaving a lot of really important work to fall through the cracks.
We never really had deadlines, especially not firm ones. Things went out when they were done, we maintained very minimal technical debt and a very high quality app. Things got love when they needed it, not on a timeline. We had a very loyal customer base.
We got bought out. We got firm deadlines for features we didn't want from on high. Quality dipped, technical debt soared, and customer retention dropped. Our overall sales went down in the name of pushing out low quality features fast.
Products are like creative works, they should be built with depth, deep dives and aim to be a fun experience and friend of the user.
Bottom up processes can create great products with polish and love, they are value creation processes.
Top down processes and deadlines create shallow features and products, they are value extraction processes.
Games for instance especially should focus on the fun and gameplay mechanic. Arbitrary timelines, milestones and push can kill both of those very important things.
This doesn't apply to just games though, products should be fun, bring joy and be almost a friend/companion of the user.
All that is lost when you move to today's finance/marketing/business led value extraction style over the creative/design/development approach of value creation.
Agile was supposed to give more time to the value creators, but instead it became a micromanagement tool of the value extractors and the depth has been made shallow. When you slow the "velocity" to add in some love/fun, you are now the problem not the solution, which is completely backwards.
The best value creation process is just an iterative/incremental cycle, with lots of power in the court of the value creators over the value extractors.
I'm on a project that started with a deadline before we really even knew what we were building (months away from that deadline and we still don't). Despite seemly unlimited budget to throw at it, technical debt is out of control and we're already moving slow.
I get pulled into 8 hour meetings every other week where 30+ people try to "plan" our way into building faster.
There are countless challenges with this project, but the forced timeline is by far the worst of it.
> I get pulled into 8 hour meetings every other week where 30+ people try to "plan" our way into building faster.
Somehow we've thus far managed to keep our "as few meetings as possible" culture. Our original founder (whose since been let go, if that's any indication) hated meetings and had rules that were basically:
1. No meetings that could be an email
2. No one has to be in a meeting who doesn't want to be there
3. No PowerPoints
Other departments have bi-weekly day long planning meetings. We have hour long "goals for the quarter" meetings, quarterly. Once in a while a retro when PMs from higher up force them on us.
Yeah, this doesn't sound like a modern approach to managing timelines.
What's described here is "take the estimate and multiply it by some factor, adding a buffer, and then commit to it" approach, which is an ancient rule of thumb.
As other commenters here pointed out, a more realistic approach is to either have a fixed deadline with flexible scope (time-based delivery) or fixed scope but flexible on when it's going to be finished (feature-based delivery).
Problem with feature-based delivery is that nothing else can depend time-wise on it. If you have a big product launch and need to prepare marketing, buy ads, or execute on some contracts, "it'll be done when it's done" is not going to fly. This is an example of a real, hard deadline, that needs to be dealt with.
The management problem I see in many organizations is managers conflating "estimates" (internally produced timelines that may or may not be achieved) with "deadlines" (externally produced constraints). These two are often handled in the same way. In a team where the managers/product owners/clients do understand and keep the difference in mind, engineers feel free to estimate without fear of putting a noose around their neck, and in turn understand that some things may be externally forced at a specific date.
The article sums up the approach I first read by the c++ spec committee.
"pick a date, but not features, or pick features but not a date"
I now often have arbitrary release dates for projects, but when I'm asked if x will be included or not I reply I don't know. We'll all find out together on that day.
After that date the project gets regular updates (often weekly) with new stuff. Updates are on the "when it's ready" basis.
This approach has worked well, when asked I explain the logic, and it seems to satisfy those that need a date (marketing et al) and also those needing features. The former can plan events around the date, the latter know they'll get their pet feature as soon as its ready. Both understand that I have constraints that make predicting unreliable.
+12 years PM here. Good teams know how to estimate. Estimations are that, estimations and not solid predictions. Deadlines should be blurry enough to have the idea when business can have something, and never fixed dates.
The reason is simple: people get sick, people find new corner cases, problems, technical difficulties and day-to-day operations will sabotage (indirectly) the excution to deliver that new project.
- A MVP that verifies a hypothesis and taking care the whole customer journey
- iteration and improvement of mvp
- part(s) of customer journey
- feature(s)
Deadline means different thing to different team. As other comment point out, there are real deadlines. When not real, deadline is about a rthym, about a pacing, about expectation at different stage, rather than a specific expectation that by an arbitrary date everything will be "done done". And who's doing the estimation is as important as what the estimation is.
In music, the time signature is to give expectation what pacing a piece is in. It is futile to pack more notes than the member can play. The job of the composer is to know what is within the possible realm of playability and to assemble a team that can handle it. You don't "push" a member, you understand the member. Same for software product building
Readit News
The adaptive approach: Products should ship as soon as they're ready for market, not according to some deadline. To ship sooner, they should be sliced into pieces that have compelling individual value. Each piece is shipped when ready. And those pieces ("valuable increments") should be sliced into even more pieces that could be shipped sooner in case new business opportunities come along. That way, when that new business opportunity comes along, you're not left with a choice between abandoning work or ignoring the opportunity in favor of finishing what you started.
There are times when deadlines matter, such as regulatory requirements, tax season, the Christmas shopping season... but people who are "setting deadlines" aren't talking about real deadlines; they're talking about artificial deadlines because they think it's the right way to manage software teams. (They don't understand that managing by deadline causes teams to compromise internal quality, and that hurts productivity, which creates stress, which forms a vicious cycle that further harms quality and productivity.)
If you do have a real deadline, the correct way to manage the risk of missing that deadline is to, once again, use an adaptive approach: identify the smallest thing that will mitigate the risk, build that first, and then incrementally expand from there in as small pieces as possible. Keep the software shippable at every step of the way and ship on the deadline with whatever you finished.
When making estimates, I also like to differentiate between what I call happy-path projections and lazy-bayesian estimates. The former take a sort of get-there-from-here best-case-scenario approach with some buffer tossed in to make it seem like we're not completely delusional. Lazy-bayesian estimates try to identify similar projects or references as a baseline for forecasting.
Daniel Kahnemann refers to the two approaches as the inside and outside view in Thinking Fast and Slow and offers an excellent tale to illustrate the difference:
http://txti.es/kahneman-outside-view
I wish I could get every project manager I work with to read this. (I've tried.)
I believe this is the main source of stress in my life! I have tried everything and I win a bit, some of the time. Despite all of this all I really gained was pushing the launch date back a bit. I have manouvered some other departments into the firing line so I don't get all of the flack. I have set the most demanding stakeholders up to argue their demands for new features in a way that makes them look bad for delaying the project. Sometimes I think we take a step towards an agile organisation, then other times we are back at square one
In this case, the client, usually and in my opinion, correctly, expects the product to fulfill all of the functionality planned for.
I used to wholeheartedly agree with this view, but I've moderated my position considerably. What I've found is that, in practice, this encourages a mindset where a team (or even an entire company) delivers a lot of "half-assed" products or features. They make a minimally viable thing, ship it, and then move on to the next thing. They never come back to actually complete all the fit and finish work.
A great example of this is the "Familiar Faces" feature of Nest security cameras. Nest security cameras have a feature where you can train them to recognize people in your household. Then, when they alert, instead of sending a generic alert, they tell you which person was seen. This is great, but Nest never actually finished the feature. There is no setting which tells the camera to only notify when there's an unfamiliar face. You can get no notifications at all. Or you can get a notification every time the camera notices activity. But there's no way to tell the camera, "Hey, let me know if you see a stranger," despite the camera having the ability to distinguish strangers from friends and family.
Alphabet products, in general, suffer from this mindset. Another example is the new Suggested Audio feature of Android Auto [1]. I can tap on the icon and it'll bring up a random selection of podcasts from my (now defunct) Google Podcasts subscriptions. Is there any way to customize which podcasts appear? No. Is there any way to make the icon go away, because I use a different podcast app? No. It's just there, waiting for my finger to accidentally brush against it as I try to fast forward in PocketCasts. They made a minimally viable thing, shipped it, and moved on.
I contrast this with Apple, or even Microsoft, where there seems to be a lot more focus on having products that are at least somewhat thought out from the perspective of the user, and not shipping until there is certain level of coherence. Alphabet, on the other hand, ships features as soon as they're ready, but doesn't seem to realize that on buttons are useless without their corresponding off buttons.
[1]: https://www.androidpolice.com/2021/08/12/android-auto-starts...
(Minimum viable product, of course, not being about products at all, but about product-market fit and testing your ideas as soon as possible using low-fidelity experiments.)
You can have a coherent user experience and still use an adaptive approach. I think Alphabet's failures have more to do with Alphabet's culture, particularly around rewards and promotions, and perhaps a disdain for non-engineering skills, rather than taking an adaptive approach.
The advantage of the predictive approach is better driving of your workers with a deadline. The disadvantage is low morale over time and lower quality products under the hood. Rotating workers who come in fresh can mitigate this outcome but at a cost.
Or maybe you could; if management can't or won't specify the cake in advance, say whether it should be round or square until ten minutes before the deadline, you could just bake a pile of cupcakes and squash them into the tin.
Dead Comment
What really matters is prioritization. Are you focusing on the right things, the impactful things? Are you also prioritizing all the 'not-so-shiny' stuff (performance, security, etc.)? Do you have actual levels of priority (instead of just making everything top priority, which is no priority at all)?
If you prioritize well, then you'll always be delivering the top priorities and some of the mid-level priorities. On a good week, you'll also knock out some of the low-priority stuff. And all of the concerns will be getting addressed. And everyone will be happy. The developers will feel good, management will be happy, and the customers will be pleased. Deadlines or estimations will be largely irrelevant.
But if you prioritize sloppily, and constantly change priorities, so you're always yanking people out of the middle of one thing to work on another that is suddenly the highest high priority among all the other tasks all of which are high priority, then you'll always get less done. Some random assortment of stuff will get done, but it doesn't look important, at least not any more important than the stuff that didn't get done. No one will be happy. Some concerns will go entirely un-addressed. And no amount of deadlines or estimation will fix that.
I'd much rather be working for a good prioritizer than the world's best psychic estimator and the world's best deadline forecaster anyday.
You're gonna get more people aligned if you focus on things that other teams happen to be working on at the same time. If you decide to work on something else when there is a huge infrastrucuture design project going on, just because of some precommitted sprint plan, you're gonna play catch up later.
If you keep dropping a task for another because of arbitrary sprint plan, you're gonna be jerking yourself and the team around and never getting people get aligned with what you're building for long enough and it'll ge harder to get feedback from them.
Momentum matters, especially for multi product teams, such as in data science and data engineering.
Which is also not just blame the people that work as product managers, but also a critique of Agile swerving too far away from waterfall and now having no method whatsoever for requirement analysis and leaving a lot of really important work to fall through the cracks.
We got bought out. We got firm deadlines for features we didn't want from on high. Quality dipped, technical debt soared, and customer retention dropped. Our overall sales went down in the name of pushing out low quality features fast.
It was nice while it lasted.
Bottom up processes can create great products with polish and love, they are value creation processes.
Top down processes and deadlines create shallow features and products, they are value extraction processes.
Games for instance especially should focus on the fun and gameplay mechanic. Arbitrary timelines, milestones and push can kill both of those very important things.
This doesn't apply to just games though, products should be fun, bring joy and be almost a friend/companion of the user.
All that is lost when you move to today's finance/marketing/business led value extraction style over the creative/design/development approach of value creation.
Agile was supposed to give more time to the value creators, but instead it became a micromanagement tool of the value extractors and the depth has been made shallow. When you slow the "velocity" to add in some love/fun, you are now the problem not the solution, which is completely backwards.
The best value creation process is just an iterative/incremental cycle, with lots of power in the court of the value creators over the value extractors.
I get pulled into 8 hour meetings every other week where 30+ people try to "plan" our way into building faster.
There are countless challenges with this project, but the forced timeline is by far the worst of it.
Somehow we've thus far managed to keep our "as few meetings as possible" culture. Our original founder (whose since been let go, if that's any indication) hated meetings and had rules that were basically:
Other departments have bi-weekly day long planning meetings. We have hour long "goals for the quarter" meetings, quarterly. Once in a while a retro when PMs from higher up force them on us.The beatings will continue until morale improves.
A few key people left, the team is somewhat overwhelmed. And now we have to ship no matter what.
Quality is dropping already.
What's described here is "take the estimate and multiply it by some factor, adding a buffer, and then commit to it" approach, which is an ancient rule of thumb.
As other commenters here pointed out, a more realistic approach is to either have a fixed deadline with flexible scope (time-based delivery) or fixed scope but flexible on when it's going to be finished (feature-based delivery).
Problem with feature-based delivery is that nothing else can depend time-wise on it. If you have a big product launch and need to prepare marketing, buy ads, or execute on some contracts, "it'll be done when it's done" is not going to fly. This is an example of a real, hard deadline, that needs to be dealt with.
The management problem I see in many organizations is managers conflating "estimates" (internally produced timelines that may or may not be achieved) with "deadlines" (externally produced constraints). These two are often handled in the same way. In a team where the managers/product owners/clients do understand and keep the difference in mind, engineers feel free to estimate without fear of putting a noose around their neck, and in turn understand that some things may be externally forced at a specific date.
I now often have arbitrary release dates for projects, but when I'm asked if x will be included or not I reply I don't know. We'll all find out together on that day.
After that date the project gets regular updates (often weekly) with new stuff. Updates are on the "when it's ready" basis.
This approach has worked well, when asked I explain the logic, and it seems to satisfy those that need a date (marketing et al) and also those needing features. The former can plan events around the date, the latter know they'll get their pet feature as soon as its ready. Both understand that I have constraints that make predicting unreliable.
The reason is simple: people get sick, people find new corner cases, problems, technical difficulties and day-to-day operations will sabotage (indirectly) the excution to deliver that new project.
- an one off project
- an imaginary product solving imaginary problem
- A MVP that verifies a hypothesis and taking care the whole customer journey
- iteration and improvement of mvp
- part(s) of customer journey
- feature(s)
Deadline means different thing to different team. As other comment point out, there are real deadlines. When not real, deadline is about a rthym, about a pacing, about expectation at different stage, rather than a specific expectation that by an arbitrary date everything will be "done done". And who's doing the estimation is as important as what the estimation is.
In music, the time signature is to give expectation what pacing a piece is in. It is futile to pack more notes than the member can play. The job of the composer is to know what is within the possible realm of playability and to assemble a team that can handle it. You don't "push" a member, you understand the member. Same for software product building