World's Biggest 'Agile' Software Project Close To Failure

00_NOP writes "'Universal Credit' — the plan to consolidate all Britain's welfare payments into one — is the world's biggest 'agile' software development project. It is now close to collapse, the British government admitted yesterday. The failure, if and when it comes, could cost billions and have dire social consequences. 'Some steps have been taken to try to rescue the project. The back end – the benefits calculation – has reportedly been shifted to a "waterfall" development process – which offers some assurances that the government at least takes its fiduciary duties seriously as it should mean no code will be deployed that has not been finished. The front end – the bit used by humans – is still meant to be “agile” – which makes some sense, but where is the testing? Agile is supposed to be about openness between developer and client and we – the taxpayers – are the clients: why can’t we see what our money is paying for?'"

Comments

Agile doesn't mean that the project won't fail

[kthreadd]

But it might make it clear that it will fail much earlier and then at a lower cost.

Re:Agile doesn't mean that the project won't fail

[Anonymous Coward]

...since there is daily accountability and you're working on smaller pieces.

So that means it's impossible to do anything big or that requires extended planning? Sometimes a developer needs to be left alone for a week to come up with something good. Regimenting the process into days and forcing a daily bullshit update is just abusive to the creative process.

Re:Agile doesn't mean that the project won't fail

[Virtucon]

I agree but the daily accountability is still something that a lot of hard core developers don't buy into. The "leave me alone" mentality still prevails in big shops. There's also a lot of gaps in Agile and IMO while Stories are great they are not a substitute for fully defined requirements analysis. Not that they can't go hand in hand but I've watched lots of Agile projects fail because of incessant changes in vision and invalidation of stories subsequent to their definition by other stakeholders. The key stakeholders either don't pay attention or louder voices who have really no relative bearing to the project somehow get suddenly important. These are often folks with something to gain by holding things up or creating confusion. That is always the problem in all projects but it seems more acute in Agile because "hey, we have a process that can allow for these changes." Ultimately the team gets into a tail chasing situation and nothing of value (to the key stakeholders) gets built and the project gets cancelled. On the successful projects that I've worked with in Agile, there's strong stakeholders, good architecture keeping the vision in place and project management that keeps things well orchestrated. Without those in the mix, it'll fail just like all software projects.

Re:Agile doesn't mean that the project won't fail

[Chris Mattern]

But it might make it clear that it will fail much earlier and then at a lower cost.

But apparently it didn't succeed in that, either. They're discovering massive failure only after three years and billions of pounds.

Re:Agile doesn't mean that the project won't fail

[tgd]

Pretty much this exactly. Also, it's tough to get programmers and managers who have never worked in an Agile environment to buy into it. My company started using it 4 years ago and we still have a few holdouts despite the obvious benefits in both productivity, cost and simply a better work environment for everyone. Hell, I think the best part about the Agile process is those one or two guys on a piece of a project that never seem to do anything and could end up causing drama simply doesn't happen in a proper Agile setup since there is daily accountability and you're working on smaller pieces.

"Waterfall" -- i.e., the "old fashioned" way of doing things -- does one thing very well that Agile loses. And that thing is something that was understood for a century of large project management planning. Waterfall ensures quality with a team of varying abilities, or large teams. Agile ensures predictable delivery, but quality is very dependent on the individual abilities of the team members.

Anyone who has done large projects would know immediately that you don't do a billion dollar project with a pure Agile methodology. You simply can't get enough people who are strong enough to deliver a quality output without a very high amount of formal planning and progress gates.

The most successful large (multi-hundred engineer) projects I've seen in the last five years tend to use waterfall for the overall project, but encourage teams to run their parts of the work in an agile manner. You get the visibility into progress that way, but the formality of process to ensure you're really building the cohesive system right.

Re:Agile doesn't mean that the project won't fail

[Anonymous Coward]

The short(er) and honest version is that projects like that are inclined to fail, spectacularly and at great expense, whilst everyone on the job gets paid either way.

People are either accountable or not. When you have teams of "project managers" that don't know shit about shit, shoveling ridiculous and disparate feature changes into the (often offshore) dev shops inbox, while their programmers simply crank out code to match, you're begging for the whole thing to go right in the toilet.

Large consulting firms like Accenture build their entire business around this kind of approach. The result is always a failure that costs millions (or in this case billions) of company/government dollars on a product that doesn't do what it was supposed to.

Re:Agile doesn't mean that the project won't fail

[Anonymous Coward]

Agreed. Agile is a WHOLE lot easier to get wrong than it is to get right. The benefits it promises come only from the synthesis of ALL its components. Project managers who think they understand it, but don't quite get it, wind up tweaking it just a bit (in a way that seems to make perfect sense) and that tweak completely undoes the approach. The effect often isn't instantaneous...it is felt over the life of the project.

Be that as it may, for some types of projects Agile is flat out wrong.

Agile is awesome when you can't afford to frontload your development costs, don't really know which features will get a lot of use and which ones will just collect dust, and when your project manager actually understands agile. But there is a very severe drawback when you are building a huge feature-rich juggernaut of a system that needs a very sophisticated and precise back-end: refactoring costs go through the roof.

Agile proponents often get angry when one says anything bad about it, and yell about how writing maintainable code is one of those elements of agile that you can't sacrifice. And I agree. And the refactoring cost gets more evenly spread that way, but remains astronomically high for complicated systems. Here is exactly why:

Every feature you add to a back end system, no matter how well coded, increases the complexity of the system.
The more complicated the system, the more expensive it is to add a new feature to it.

Waterfall can give you some savings in this specific department because the initial system design already includes the final (or near-final) level of complexity, so you don't pay the extra cost of adding a new feature to an already existing system over and over again. Waterfall still has costs in that some of those features may never be used, and hence are wasted effort, and also misestimation and requirement changes are more expensive and put a project past deadline. But for huge systems, these costs are preferable to the project-destroying costs of agile refactoring.

There isn't one approach that is right for every problem, and agile is not the right approach for this one.

Re:Agile doesn't mean that the project won't fail

[LifesABeach]

This should be modded Score 5, Funny. Your thesis is that everyone knows what thery're doing, except the programmer that can't make it work, and is honest enough to tell you so.

Repeat after me, "The King has no clothes." It just doesn't get old.

Re:Agile doesn't mean that the project won't fail

[gutnor]

Actually Agile would probably handle mediocrity quite well, at least making it apparent.

What agile really sucks at is handling the political aspect of the project, because simply agile requires complete honesty and honesty does not scale very well above a small team. In a very large project involving loads of teams, management is a lot closer to a poker game and you don't win at poker by showing your cards to all the players.

Re:Agile doesn't mean that the project won't fail

[sphealey]

- - - - - A developer's job, in a project like that, isn't to come up with something good. Their job is to implement a specific piece of functionality in the specific way defined by the people whose job is to have the broader view of the project. - - - - -

The unstated assumption in that theory is that there is a group of human beings who are unto demigods, capable of defining detailed specifications that cover all eventualities, see far into the future, and can be implement mechanistically by simple code grinders. I have run into two or three such people in my entire career, out of the thousands of system architects and business analysts I have worked with in some capacity. And even those three people (a) often got caught by surprise by unintended side-effects (b) were often part of a process that took so long to come to fruition the system so designed was obsolete by the time it was deployed.

The only industry I am aware of where strict, spec-controlled waterfall is probably appropriate and generally works is aerospace controls systems. Of course that industry is notorious for huge schedule and cost overruns (the A400 software fiasco is particularly instructive in this regard), and even their software is not free from bugs and unintended consequences. Ref the numerous changes Boeing has made to the electrical system control software on the B-787 since it went into service - why didn't they just "get the spec right the first time"? Enrico Fermi's comment on how to manage the unknown is very instructive.

sPh

Re:Agile doesn't mean that the project won't fail

[naroom]

If all "Agile" really means is "Do the project in such a way that it will succeed", then "Agile" is a useless word. Agile has a bad case of "No true Scotsman".

Re:Agile doesn't mean that the project won't fail

[ShieldW0lf]

I hate to break it to you but "fully defined requirements analysis" is a pipe dream filled with rainbows and unicorns. I have never, not once, seen a requirements document that accurately captures exactly what the system will do.

Well, I've written them, and I've never had a project fail.

One example involved interviewing

i) the owners of the company

ii) an executive from each department

iii) a "regular joe" representative from each department

This became a 40+ page project specification, which was signed off by all stakeholders and became the contract.
 
Then this document was fed into a series of code generation engines, which created hundreds of thousands of lines of code. This was all done with an eye towards allowing various professionals to go away and do what they do best without getting held up waiting on each other or tripping over each other, filling in the missing functionality in the generated code.
 
That system is still in operation close to a decade later, organizing the working lives of thousands and serving the needs of millions.

Now I work in Agile. I hate it. I'm always having to check with other people constantly to move forward, I never get in the zone, there's a lack of clarity and vision, and I feel like I'm getting stupider each day and I'm not producing my best work.

Re:Agile doesn't mean that the project won't fail

[Aighearach]

The benefits it promises come only from the synthesis of ALL its components.

Yes, yes, we've heard it a thousand times, if an agile project fails, it must not have been truly agile. Probably isn't a true Scotsman, either.

Re:Agile doesn't mean that the project won't fail

[Anonymous Coward]

...since there is daily accountability and you're working on smaller pieces.

So that means it's impossible to do anything big or that requires extended planning? Sometimes a developer needs to be left alone for a week to come up with something good. Regimenting the process into days and forcing a daily bullshit update is just abusive to the creative process.

You've already failed at project of that size if you're letting a developer be "alone for a week to come up with something good". A developer's job, in a project like that, isn't to come up with something good. Their job is to implement a specific piece of functionality in the specific way defined by the people whose job is to have the broader view of the project.

Parent poster here. I see your point, however, my experience has been that separating the designer/architect role from the developer role is fraught with pitfalls. The people writing the code should be the ones designing it, otherwise you end up with a skyscraper put together with superglue instead of bolts and welds. The response to that is to have precise specs, but the time spent writing specs that are precise enough to be flawless is more productively spent doing the actual coding.

The bottom line to all management fads and whether a development process is effective and ineffective are fundamental human limitations:
1. Quadratically growing communication overhead as teams get larger
2. Finite human memory capacity, limiting how much information about a project each developer can be aware of (and each developer must be aware of the whole project, to avoid causing havoc or reinventing the wheel, or they must be sequestered in their own subcomponent)

A good team size appears to be five, and an organization should not have more than five teams that need to intercommunicate. The challenge is to break up a piece of software to fit this kind of arrangement. Anything else will lead to company-wide split-brain, NIH, and communication bottlenecks. This is true regardless of the development process selected. Fundamental human limits are like the speed of light, you can get closer to them, but you cannot exceed them. So assemble good teams, architect your project to fit those teams (or vice-versa), and remember that everyone is human.

What's that saying about agile?

[NotSoHeavyD3]

Agile assumes you have smart, talented, dedicated individuals doing the work. Then again if you have that pretty much anything works.

Re:What's that saying about agile?

[jacobsm]

Agile is just the latest management fad. In a year or two something else will come along, and the lemmings will follow.

Re:What's that saying about agile?

[ph1ll]

"DWP IT chief and government chief information officer Joe Harley said in May 2011 that agile would ensure his department delivered Universal Credit on time in October 2013."

So, a two year iteration and a guaranteed delivery date? Yeah, that really sounds like Agile.

The article goes on: "Attention must turn to Accenture and IBM, who are on track to earn £1bn between them as lead developers of the system. They may have played the most significant part in agile's failure at DWP, or DWP's failure at agile. Accenture and IBM may have found agile commercially inconvenient. Neither has yet been able to speak about it."

Ass-Center and IBM? Yes, two companies who are well known for their love of Agile [rolls eyes].

BTW should I mention my pet name for Agile

[NotSoHeavyD3]

I refer to it as "Monkey's Paw Development" (And before anybody asks agile to me ends up being "Hey let's ask developers what they'd wish for in an awesome development environment. Then give them that but give it to them in such a way that they regret ever asking for it.")

world's biggest?

[hackula]

"World's biggest" and "agile" don't really go together. One of the core tenants to agile is to break things down into small chunks. Multi year contracts for a predetermined end product are waterfall by definition. Either way, I have seen waterfall work just fine and I have seen True Agile[tm] fail hilariously miserably (to which most Agilistas respond with some form of the "No True Agile" fallacy). The most important thing is tight iterations. If a 2 week sprint fails, then it is not that big of a deal. If a 2 year death march fails? Someone's getting fired, since its the equivalent in agile-land of failing 52 sprints straight.

Re:world's biggest?

[Kjella]

The most important thing is tight iterations. If a 2 week sprint fails, then it is not that big of a deal. If a 2 year death march fails? Someone's getting fired, since its the equivalent in agile-land of failing 52 sprints straight.

But is it two weeks sprint down a dead end? For a project this size, agile is like trying to build a skyscraper first as a one story building, then two story building, then three story building and so on. Apparently you're making great progress the first sprint and you have a shack up, that's 1/100 floors done already. Except it doesn't work like that, so sometime around the 20th floor you've got people all over the first 19 trying to build in extra support columns and stronger walls and propping up the foundation. Things grind to a halt and you're not making any real progress. Then the orders come to get moving and you start going upwards again more and more rickety until eventually you find the straw that broke the mule's back and it all comes crumbling down.

Agile is nice if you're close enough you can start delivering actual features that would belong in the end product at the end. In practice it often means you build the first iteration with string and duct tape planning to replace it with something more solid on the back end in time, but I think everyone knows how that goes - the string and duct tape has a tendency to stay because that part is "done". Of course hindsight is always much easier but agile I feel lacks foresight, we do this now to meet our sprint goals and then if we need to change something to meet our next sprint goals, we'll deal with that then. In practice, there's not time to go back and rework things every time you figure out this should have been done differently.

Re:world's biggest?

[darkstar949]

You can build a perfectly suitable "one room shack" by slapping some canvas & boards as temporary walls around the first story of a steel-beam skyscraper. Sure, it'll be one of the most over-engineered one room shacks in history, but that shack is architected for future expansion, which is why those giant steel beams are being used as support members, rather than scrap 2-by-4's.

If you find that you've reached the 20th floor only to realize that all your steel beams are only able to carry 20% of their expected load, then that is a failure of your architects, who failed to provide a design for a building which will be 100 stories tall, but which must get built just a couple stories at a time.

I think what the grandparent was trying to say is that a lot of Agile projects tail to give people proper time to actually do some initial architecture of the project before people jump into the sprints. Realistically, you need to combine waterfall with Agile methodologies since large projects need to have a solid foundation to work off of that might take a couple months to put in place.

To go back to the building analogy, if you are building a skyscraper you are digging a basement and putting footings and everything else in before you even build the first floor.

Yeah...

[Greyfox]

Just because you're agile, doesn't mean you crap daisies and unicorns. I often see inept upper managers latch onto agile as the latest magic bullet which will solve all their problems with no other changes on their part. Except they keep all the micromanagement bits, discard all the engineer empowerment bits and hand their scrum team a year's worth of priority 1 stories to implement in the next sprint. Good managers will likely be successful no mater what methodology they use, bad ones will likely fail no matter what methodology they use and the ones in between will have mixed results no matter what methodology they use.

Agile is (usually) BS

[Anonymous Coward]

All Agile methodologies really are are different ways of implementing the Spiral Model of development. If used correctly, any of them can work fine. Unfortunately, that's only in theory. In practice, Agile generally becomes an excuse to use Code and Fix, which is the worst methodology and the most prone to failure. Beware anyone who claims that Agile is the solution to anything.

Re:Because

[PNutts]

You wouldn't understand it.
You would exploit it.
You wouldn't do anything to make it better.
You would waste time complaining about everything.

Burma Shave!

Re:And who were the contractors?

[abirdman]

Notice they've got Oracle in that list. This vendor list is a nasty bunch of international billionaires-- individuals and corporations. These are the kind of companies who want to "partner" with you if you use their products-- one doesn't "buy" Oracle (or IBM or BT) products, one carries them like an STD. Note the three local contractors and sub-contractors who sell to the government, and then sub out to a bunch of bloated global corporations who have no (non-monetary) interest whatsoever in the project working, and probably won't repatriate the profits. This does keep the salaries in the field high. And the government has no choice but to bid out another contract for a plum software project right soon. There's a lot more partnering to do.

Brogrammers

[Anonymous Coward]

A lot of the 'agile' models reward talkers and people who take immediate action and can rattle off buzzwords, at the expense of more introverted engineers who like to investigate and plan before they act. In a continuously moving environment such as a social networking site, that reward system might be appropriate. In a financial back end doing mission critical work, that sounds like a disaster. So, no surprises here. One size does not fit all companies.

Re:Where's the testing?

[Anonymous Coward]

That question would indicate to me that they're doing Agile wrong.

Agile, unlike everything else in the world, is the perfect silver bullet. There can't be anything wrong with Agile. The seminar even said so.

That is the worst article I ever red

[devent]

This should be suppose an article about "agile" and the Universal Credit. After reading the article there is no information what-so-ever, except that the Universal Credit project has been admitted to be failing.

So why is Universal Credit an "agile" project?
Why it is failing?
What is Universal Credit anyway?

Maybe that is why Twitter is so successful, the whole article is just a Twitter message: "Universal Credit, suppose to be biggest Agile Software Project, is failing".

Here is some more information:
http://www.guardian.co.uk/society/2013/apr/29/universal-credit-pilot-scheme
http://www.guardian.co.uk/commentisfree/2013/apr/30/universal-credit-iain-duncan-smith

Is it called "agile" because it's a "step-by-step approach" ?

Been there, still there

[Loki_666]

Over my career, i've worked in the UK Benefits Agency processing claims, i've worked in their IT departments, i've worked for the outsourced departments later supporting them, and i've worked for a software company which loves agile (but will do waterfall if pushed).

The problem here isn't waterfall/agile. The problem here isn't .Net/Linux.

The problem here is the parties involved. On one side you have a government agency where people obtain seniority largely through age, not skills, and the main skill that is relevant is passing the buck when things go wrong and taking the credit when things go right (really, this is government agencies through and through - not to mention, most people with real skills/brains get out as soon as they can). On the other side you have the dinosaurs of development (not necessarily age, but sizewise).

Somebody earlier in the thread stated this whole project could have been delivered with much lower cost, with just a few devs, in a much shorter time. I'm 100% in agreement with them. The only real complexity of most government systems is the labyrinthine workflows, but they are documents and strictly followed in their paper variants, its just a matter of getting an understanding of this and turning it into software.

My recommended development approach for this project would have been as follows:

1) Hire some decent devs. They don't need to be hotshots, what is being developed is fairly simple from a technical standpoint. Mainly guys who can follow a spec.

2) Take a bunch of people who actually do the work for real, the paper pushers. Take them down the pub and get them rat-arsed. Listen to them bitch and whine about all the idiotic things they have to do in day-to-day operations.

3) Take notes of their bitching! It may help if you are drunk!

4) Any requirements given to you through the official channels are probably worthless. Dump them. They will simply mislead you from what is actually required.

5) Build the system based on what you learned from the drunk employees.

6) Demo it to the stakeholders and hand over.

7) Contract fulfilled.

Oh, and of course.... 4) Profit... erm...

Re:It is based on Linux....

[rtfa-troll]

There are many possible reasons for failure.

• you are stupid
• you base your product on .NET
• you fail to start a testing program
• you are the British government
• bad luck.

Just because the failure of one project is caused by .NET does not mean that a project not based on .NET is going to succeed. In fact, traditionally 80% of software projects fail.

This project is clearly failing for the second from last reason. It is also failing because it is not an "agile project". An "agile" project cannot fail and cost Billions because it must always deliver runnable software with a maximum of a few weeks delay if you use some "semi agile" process like scrum or immediately any point if you use some true agile process.

Once you deviate from "Working software" for more than a couple of sprints (everybody can make a mistake) then you are no longer doing agile. I have seen so many "agile" projects which seem to define "Working" as meaning something like "a prototype which would never work at full scale" and so they have never addressed the major problems of their class of system.

If they are "billions" of pounds down whilst doing agile, then they should have already delivered plenty of working systems and have hundreds of happy users. In this case they are a "success" even if they were a bit slower and more expensive than some other projects. If, however, they really haven't delivered anything then what they were doing was an unplanned disaster using "agile" as an excuse for not having a proper plan.

Whilst I know that the "waterfall" method of development is famed for it's failures. Whilst I know that those failures are spectacular and huge. I really don't see how you deliver, for example, 5% of a working mobile phone network. You just have to have a big interlocked plan with a working phone, transmitter, backend, management and interconnection all planned together. I don't believe that such a thing can be done in a true "agile" way and pretending that you are doing it in an agile way is a dangerous fantasy. Only once you have a working network can you start to improve it in Agile increments.

Mandatory requirements and Agile fallacies

[Anonymous Brave Guy]

An "agile" project cannot fail and cost Billions because it must always deliver runnable software with a maximum of a few weeks delay if you use some "semi agile" process like scrum or immediately any point if you use some true agile process.

The trouble is, you can have software that runs and passes some tests, yet still does not meet all of the mandatory requirements for the project and therefore may have no value at all in the real world. You don't get any credit for meeting 90% of the mandatory requirements on a job like this. The idea that having software that runs and maybe passes some tests has some sort of inherent value might be the biggest fallacy perpetuated by the whole Agile movement. It's just not true, and therefore neither is the claim that agile projects can't fail as a result.

Re:Mandatory requirements and Agile fallacies

[khchung]

Sorry, but you are completely missing my point. My comment wasn't really about tests, it was about the fact that these huge projects are practically all-or-nothing in their success.

Exactly. You cannot use Agile to build a 100-mile canal, as the whole thing would be useless even if you completed 99 miles.

If the system cannot be useful until a large set of functions are in place and working, then it is not suitable for Agile, period.