Project Delivery and Overengineering are the Same Problem
I have not been posting much lately. The proximate reason is that I spent most of October finalizing with Borners the text of the London construction costs report; there’s still another round of edits but they’re quite small. I don’t want to preview the argument in the report yet, but I do want to talk about one thing that jumped to me while looking at the cost history of London. In New York, we broke down the cost premium into three components, each stacking on the others: labor (primarily overstaffing) raised overall costs by a factor of 1.5, project delivery problems by a factor of a bit more than 2, overdesign of systems and stations combined to a factor of almost 3. Ever since then, almost four years ago, we have tried tackling procurement and poor design separately, having some successes on the latter while nothing has gotten better and much has gotten worse on the former. But what’s emerging out of what we’re seeing in London is that really, the procurement and design premiums are the same problem.
To be clear, “the same problem” should not be taken literally. Stations can be designed to be twice as large as they need to be even under a system of solid project delivery, and vice versa. However, their causes are deeply intertwined, and historically, in London (but not New York), they emerged at the same time.
New York’s cost history is unique, in the sense that high New York costs go back to the IND era of the 1930s. Already then, there was a cost premium of a factor of about 4 over London and Paris. The cost premium over Paris has further increased then, though that over London has not, London itself having a rapid increase in costs starting in the 1990s. The IND’s project delivery did not to our knowledge have any special problems, but it was severely overdesigned, with full-length mezzanines, underground flying junctions, and wide curves under streets requiring demolishing corner buildings; Sixth Avenue Line was particularly expensive as it had to be built around and under PATH and carve a new avenue through the already gentrifying Greenwich Village.
But elsewhere in the Anglosphere, even to some extent in the United States, costs exploded later. British costs were lower than Continental European costs in the era of the Victoria and Jubilee lines, and Canadian costs were similarly low. And in the UK and as far as we can tell also in Canada, the procurement and design cost premiums increased at the same time. The Jubilee line extension was already showing the procurement problems and frictions that would later come to a head on Crossrail, and also had if anything even more oversize stations than Crossrail relative to train size. Then at Crossrail, both of the problems that we thought separate were fully exhibited. On the project delivery side, there were extreme contingencies (all used up), premiums due to different parts of the project not talking to each other, and general issues with fragmentation. And on the design side, the station dig volumes were enormous: the central Crossrail stations excavated 150,000-200,000 cubic meters each, compared with 85,000 for Odenplan on Citybanan, serving trains only 10% shorter than the eventual plan for the Elizabeth line.
Some of this can be readily explained by the same causes. Crossrail, like the American projects for which we have data, has an atypically high share of systems costs relative to civils. Normally the ratio of systems to civils, in France, Italy, and Sweden, is around 1:2.5; Second Avenue Subway and Crossrail were both nearly 1:1. In London, there was an obvious problem of difficult systems integration, as different project requirements set by different organizations that didn’t talk to one another enough and didn’t prioritize value engineering forced ever more expensive solutions. The same problem of fragmentation and lack of coordination also leads to project delivery issues, as there is insufficient supervision of the contracts, which are increasingly obscured behind design-build mechanisms.
But station sizing is harder to explain except through a push and pull among different organizations that want their own requirements. London and New York both have the same topline results of stations being too large (though this is less true in New York going forward than it was for Second Avenue Subway Phase 1), but the exact set of departments and requirements varies – New York had a problem with back office space for different departments, London likes to excavate big digs for architectural purposes so while the stations are equally overpriced they at least look somewhat nice.
(You may wonder what about labor. The answer is that we never tried to seriously tackle it. It’s an extremely localized problem to the Northeastern United States, and even California, otherwise fully New Yorkized in its costs relative to urban construction difficulties, doesn’t have it.)
One cause for pessimism here is that evidently, while we did manage to pressure New York into saving some money on Second Avenue Subway Phase 2, the savings did not stick with the 125th Street subway proposal, which costs twice as much, makes the procurement problems worse by doubling down on design-build and similar privatization of project delivery (design-build being particularly important to MTA President Janno Lieber), and swaps the overly long station digs of Phase 1 for overly deep ones instead without resolving the underlying problem.
Yep – the IND 6th Avenue Subway also had to underpin the 6the Avenue Elevated Line that it would eventually replace in 1940. Also, the 6th Avenue express tracks had to be in a deep-bored tunnel under the existing PATH lines, and did not open until 1967.
Yep – the IND 6th Avenue Subway also had to underpin the 6th Avenue Elevated Line that it would eventually replace in 1940. Also, the 6th Avenue express tracks had to be in a deep-bored tunnel under the existing PATH lines, and did not open until 1967.
I would defend two aspects of IND ”overbuilding.’ First. operationally, flying junctions are ideal, as they eliminate delays at points where one late train would otherwise spread delay. Second, they are inherently safer than flat junctions. Have a look at CTA’s recent upgrade where Brown Line trains diverge from the 4 track main Red Line. As to IND stations, when I moved to NYC in early ’66 I was surprised by the 2nd-Houston Station. I learned the oddities were provisions for the still unbuilt SAS. I am reminded of newspaper stories when I was in HS in suburban DC, Pictures appeared of a ‘bridge to nowhere’ which long since was connected to segments of I 270. The bridge contract was separate–thus that builder got in, got out, got paid sooner. NYMTA was tunneling 2nd Ave 6 blocks N of the station when the bond money was absconded with to cover NYC debts. (Ford to NYC, Drop Dead)
It took decades but the ‘ extra srairways’ at B’way-Laf finally were connected to the NB 6 as planned in the ’30s
None of this is meant to defend the corrupt and needless oversized stations on the existing SAS. I am a firm believer in cut and cover, shallow tunnel subway design both for cost, but more importantly time saved between sidewalk and platform.
One of our German consultants once said: “You can only future-proof so far before you kill the project.”
While incremental upgrades are usually harder than building the same things in the first place, you also need to get off the ground.
Furthermore, the future-proofing needs to have its assumptions turn out to be correct. In Budapest, when they built the M3 metro line’s Kálvin tér station, they anticipated the future M4’s connecting station, and dug the connecting pedestrian tunnel. Since M3 was running roughly parallel to the Danube, while M4 was to cross it and this was to be the station closest, they naturally assumed that M4 would run below M3. When M4 was eventually built, it was decided it should run above M3. But since the pedestrian tunnel existed, they used it — so now the escalators between M4 and said connecting tunnel are rather longer than is actually necessary.
So per-passenger, which is the important bit really, the central Crossrail stations are similar sized or smaller.
And then you have to add on the rough doubling of cost of the additional station upgrades the Victoria Line has had after it was finished due to lack of capacity.
Except that it handles far, far larger volumes of passengers than the cheaper lines built elsewhere. Even line 14 in Paris which you like to compare to is still getting only half the number of passengers of the Victoria Line or Jubilee line extension.
Probably because it was a much more complex project than most of the others you have looked at?
Stockholm isn’t that different in the central station’s passenger load. And Marmaray/GTX/Tokyo Metro etc have same-as-Liz line passenger lines without Liz line costs. And just look at the stations volume that isn’t occupied by passengers, particularly those high ceilings and negative spaces. It looks nice, but is unnecessary expense, given that a cubic metre of space is an order of magnitude more expensive underground than above ground.
Ditto JLE, which isn’t over expensive because all of its underground stations have three escalator banks instead of the traditional 2, but because it has four deep stations with natural light reaches the platform, CW/North Greenwich have Crossrail sized station boxes for trains sets that are half the size etc. Because its stations are
PPP era upgrades are what 40 years after the Victoria line entered service? That’s about the time you’d expect large scale modernisation works. And PPP costs are all inflated compared to pre-1987 modernisation costs, plus backlogs from neglect too. And with Traditional System costs we should have had Crossrail 2.
Its not especially more complex than Marmaray in Elif’s Istanbul report. And we also have other mega regional rail elsewhere to know Crossrail has elevated costs well beyond any reasonable weight comparison, RER, Munich-S-bahn, Japanese through-running lines, GTX, and the newer express Chinese metros. Complexity/Scale can justify 1.5x Global median costs, it cannot explain 4.5-5x Global median costs.
The station quoted in the article has what 25m passengers a year, Whitechapel has 50m including interchanges compared to 13m pre-Covid when the Elizabeth Line didn’t exist. That points to the Elizabeth Line having 37m passengers a year perhaps at that stop. Farringdon is 60m including interchanges with 15m pre-Covid on Thameslink, so that points to the Elizabeth Line having 45m there, Tottenham Court Road is 70m, Bond Street is 43m, Paddington and Liverpool Street may be more in-line with the Stockholm one, but a fair number of people will be travelling through on the Elizabeth Line who would have ended their journeys at the terminus before.
Marmaray has closures at peak times, and could be built with cheap labour from Eastern Turkey.
And how many of those have the ridership of London and can’t use cheap labour from the countryside etc?
Tokyo through running lines have multiple times the ridership of the Elizabeth Line. Just the core section of the least busy lines are busier than Elizabeth Line as a whole, and each suburban section has on the order of an Elizabeth Line by themselves as well, e.g., Tobu has about 2.4 million passengers daily mostly across the Tojo (Yurakucho/Fukutoshin), Skytree Express (Hanzomon), and Skytree Local (Hibiya) in addition to the million to millions ridership of the core section, plus sometimes another order-of-Elizabeth-Line-ridership suburban section on the other end.
GTX and the newer Chinese metro lines have much lower ridership than Elizabeth Line, but since their suburban sections are from scratch instead of over a century old, it takes more time for neighborhoods to build around them. I suspect that Korean and Chinese planners aim for ridership at least on the order of Elizabeth Line if not Tokyo.
I’m kinda assuming that the central Elizabeth line has been sensibly overbuilt to allow a second Western branch to somewhere and that the South Eastern branch is extended to Dartford – so it would support 400 million passengers or something without falling over.
My understanding is that the Tokyo metro lines top out at around 1.5 million passengers a day – so 400-550 million passengers a year depending on your definition of a day?
The Japanese certainly historically took a pretty aggressive approach to crowding people onto trains with the white gloved staff to push people onto them.
We will see how it goes for them if/when they get there!
The thing is it’s much cheaper actually to get stuff right at the start and not need to do upgrades later if you can avoid it.
And yes there are some examples elsewhere in London where its been done, specifically removing the narrow island platforms at Euston and perhaps some other places as well as the work at Bank. But there aren’t loads.
I don’t disagree for what it’s worth that Britain has a cost problem. Warwick Parkway cost £5m in 2000 whereas Thanet Parkway cost £40m for something similar.
Its also true that some newer stops are extremely big. Canary Wharf on the Elizabeth line and Nine Elms on the Battersea extension both look exceedingly voluminous for what they are. The Thamesmead extension of the Barking Gospel Oak Line was extremely expensive for something in the suburbs – and it doesn’t even go underground so you could cross the river.
The planned Bakerloo line extension out into the suburbs, the DLR crossing at Thamesmead and Crossrail 2 all really look overpriced.
And Cambridge South closer to £250 million, even if it did include about a kilometre of four-tracking…
Pretty much all of them. TBMs aren’t amenable to cheap labour like embankments or C&C. Labour will become more an issue since we cut ourselves off from European FOM.
Victoria is pretty much perfectly designed for what it was intended. They couldn’t know what happened afterwards i.e. ridership would massive increase system wide for 2 generations but because the system can’t build everything was hard. Again Paris manages to automate its lines reasonably well. We can’t. And TfL weaponises this against automation.
Yeah I put Barking Riverside in the database its expensive for an elevated line mostly in existing rights of way/publicly owned land/developer land.
TBM’s also aren’t the only cost involved in building a railway, most of the rest of the tasks can be done by cheap labour and are therefore cheaper in Turkey.
Also building frankly a cheap and cheerful line in Istanbul that already requires the station to be closed at peak times was better than the previous status quo and was probably all that could be afforded. That same rationale doesn’t make sense in London.
Even if this is the case you can’t really argue the Jubilee Line Extension was overbuilt because they built it well enough to handle 300m passengers a year from the outset. And the Tokyo metro lines that you admire (and which have similar traffic levels) also cost something in that ballpark.
I’m not sure what this has to do with this discussion?
What numbers are you comparing here? Because the 820k weekday passengers that RATP quotes seem similar to the 302/276 2019 ridership of the full Victoria/Jubilee lines.
Line 14 is designed for about 1 million daily passengers eventually.
https://en.wikipedia.org/wiki/Paris_Metro_Line_14#Statistics – 150m passengers a year. So by London standards it’s still a low traffic line.
If/when they are handling 300m passengers a year on that line without peak time closures and without any expensive additional capacity I will accept the French did it better.
This is an odd point of view. Is there anyone who thinks it is “better” to have a single line handling 300m pax pa? That M14 and other very busy Paris lines remain lower shows that the French do things better! They have pre-empted the capacity issues and built new lines or improvements to cope. M14 itself was built to relieve M13, RER-A and partly M1. M1 was retrofitted in 2011 for driverless trains and platform doors, and M4 later. I presume RER-E since opening its new 8km tunnel to La Defense last year has taken some RER-A pax and in the near future will take over a major RER-A branch. The orbital M15 will also reduce use of the other mostly radial lines, although it will likely induce quite a lot of new users from among the ≈5 million residents who live in the Petite Couronne, but certainly improve mobility for lots of people.
The Ile de France system (Metro + RER) has 800km of lines compared to London’s ≈400km and they transport 3.1bn and 1.7bn pa respectively. Better. The RER system was built in the 60s-70s to relieve the pressure placed on the Metro from all those suburbanites debarking at the main rail terminals. Similar to why the Crossrail-1/Elizabeth Line was built … but 45 years after RER-A. Crossrail-2 was “to alleviate severe overcrowding that would otherwise occur on commuter rail routes into Central London” and specifically on Victoria Line. Who knows when it will be built but, just like CR1+2 and HS1+2 there is the peculiarly Anglosphere notion that to alleviate costs it is a good idea to stretch out the timetable for building such infrastructure: “To ease the funding issues TfL recommended spreading the funding over a longer period and completing the project by the 2040s, ten years after the initial projection.[30]” Can the Vic line achieve 400m?
Likewise Tokyo is under-built
On the other hand of course the Paris system has much higher running costs per rider than the London system.
So there’s your trade off. And over the years those running costs add up to a lot more than a few billion extra capital spending on the Jubilee line extension and Crossrail.
Especially when anything outside zone 1 should be able to be constructed to a lower standard anyway.
As usual, you are missing the point. Your claim that Victoria Line’s super-crowded conditions (and the fares being about twice Paris), make London “better” is almost Trumpian.
You are very confident about those cost comparisons. I remain sceptical. But even if true, perhaps it is due to the scale difference, ie. 800km v 400km of track and presumably almost double the trains and a lot more stations (560 v 272).
Your last sentence is opaque.
@Michael there’s a pretty persuasive argument that for Crossrail 2 you split it into 2 lines built to French standards assuming French ridership though – as because London is longer east-west than north-south two new lines through the middle would make sense.
@Michael, I don’t think ultra crowded lines are better, although they clearly justify stronger service, but if they need more infrastructure and that costs more that seems pretty reasonable?
Especially as outside the very centre you shouldn’t need much if any extra infrastructure so your overall costs for a whole line should be more reasonable.
As it happens the final segment of the Sydney M1 Metro line opened last Friday (30 October 2026). It is an RER or Crossrail type line, deep underground in the city sections and with huge stations. A sense of the stations can be seen in the picture of Victoria Cross station in the Wiki. They seem even more voluminous than any RER station I remember. The whole project is estimated to have cost A$30bn but it is ambitious with the depth, the fully-auto trains, platform doors and having a cross-harbour tunnel. At 66km it is claimed to be the second longest fully automated driverless metro after Istanbul M11.
The 13km section that just opened looks the most expensive at A$10bn given that it is above ground and takes over an existing train line (T3) and no monster stations etc. Still, there hasn’t been much fuss about it probably because Sydney was so overdue for improved train service (first new line since 1979 I believe!) and its immediate success. The opening at 4.30 am on Friday was swamped with enthusiasts. In 2025 it carried 72m and with the new section it should pass 100m or more, which is stellar for an Australian city.
I have a feeling this somehow relates to complexity and loss of a holistic approach. I’ve come across things like a strong beliefs that every discipline/domain expert optimizing their speciality will somehow lead to an overall optimum solution and that the end users are the absolute authority that the project team should always bow to. Both are wrong of course. The first leads to overengineering and second leads to excessive requirements.
In my case the first was related to not understanding what project management actually does. After all, it doesn’t produce plans or other tangible deliverables so how can it be useful? All of these also tend to involve managers far removed from the substance of the work. And MBAs.
I’m also reminded of the time we decided to try and get around some software limitations and licensing costs for electronic displays by procuring the software as open source and the displays separately. It turned out that systems integration is quite hard. Who integrates all the technical domains into a coherent system?
Project management does produce a project plan and other tangible deliverables. Or at least it’s supposed to. In aerospace world the most important work product out of project management is called the Integrated Master Schedule, with tasks laid out and linked to each other by logic, then resource- and budget-loaded. With this you can tie to critical program milestones and then periodically perform analyses, including but not limited to critical path, variance (to cost or schedule), and performance (by reporting period or at complete). Subcontracted tasks and projects are supposed to be replicated in whole or represented in a simplified manner in the top-level project plan. The most common visual representation of this is a Gantt chart.
A group of planners, managers, and engineers who are adept at both understanding the work they’re doing and how to measure it such that performance can be reported accurately against spending is absolutely critical for success.
Project management is also supposed to produce process documentation for the project to follow, such as risk and opportunity management, issue tracking and resolution, environmental health and safety, or how this particular project will implement and/or tailor the organization’s processes to this project. These are supposed to be laid out in the contract as deliverables (if this aspect is contracted out) or produced in-house and placed in the program documents repository for the assigned personnel to understand and follow.
At least this is how it’s supposed to work in my corner of the world, and my sincere, perhaps naive, hope is that engineering is enough of a universal language that this is broadly true across orgs and practices. It certainly seems basic and nonnegotiable to me.
None of this is new or difficult to learn, but it takes quite a while to master for long and complex projects. There’s a big universe of software tools out there available for this purpose. And big orgs do struggle with it for a number of reasons. But hopefully this gives you an idea of the tangible work products of a project manager.
I think I’ll need to argue two sides to this 🙂
On one side: yes. These are not really appreciated or even well known on the civils planning side in Finland. Project management at civil engineering consultancies is surprisingly underdeveloped. Most planning projects simply aren’t that big. And our managers in the public sector often come from this background. On the construction side the most common method is simply to place large financial incentives on site managers at let their greed work it out one way or another.
On the other side I feel there can be a lot of unnecessary overhead in running things by the IPMA book. It very much depends on the complexity and safety-criticality of things whether you really need all that stuff. It also tends to be rather rigid. I’ve been involved in delivering projects worth hundreds of millions of euros on time and under budget while ignoring a lot of stuff from PMBOK. Actually not even hearing about a lot of it. Setting a common goal, low hierarchies, colocation of designers and builders and maintaining momentum can take you surprisingly far.
I’ve been away from project delivery for four years now, so my knowledge is probably dated. At the time I moved away I was very interested in getting further into lean construction which my last project experimented with. Methods like the last planner system claim to provide a systematic approach while maintaining agility.
Sure, for small projects you don’t need much of this. I was able to knock out a labor-only resource-loaded schedule for a relatively small project of mine in a few hours by myself, and it took $500M project where we had dedicated, full-time planners and project managers, and just managing the schedule and performance takes a significant chunk of everyone’s time.
You’ll get no argument from me about clearly defined goals, flat hierarchies and small, agile teams. You can do away with this top-heavy project management structure in such an environment without sacrificing performance and quality. The caveat is that the culture of such teams has to be strong and the composition has to be of top performers, and it may not be suitable for a project above a certain size. Lastly, for government clients, it may not even be permitted; a well-meaning emphasis on accountability of taxpayer funds may require the preparation of these complex products and implementation of these processes. And large rentier organizations are the most well-equipped to siphon value from the taxpayer by gaming such a system.
Okay, comment mishap: the first paragraph should read “Contrast this with a >$500M” instead of “and it took”. I’ll take this opportunity to add that management staffing scales with project scope, and if it’s worse than, say, linear, then management practices need to be examined for efficiency.
No argument from me.
And of course our project to the tune of 500 M€ did involve a lot of project management and managers. Although a lot of it distributed. This was possible because relatively clean partitioning was possible, eg. depot separately and construction divided into five areas with manageable interfaces. And I don’t even know exactly how the contractors managed the works. I do think the master schedule was mostly a compilation of the area and rail systems schedules though, not an independent product per se.
I believe one problem is that very few organizations really have experience delivering megaprojects. Or very complex projects. And they lack the methods too. What worked for us probably would not work for a project in the billions.
Your note about top performers is relevant as well. Our successes involved quite a bit of heroics. Not good for people and not scalable. I’ve never gotten as close to burnout as I did on the Jokeri project.
Do you have an rough idea of when the London report will come out?
If you’ve dug a hole, filling it with air cost approximately nothing. Slapping some tile on the concrete that would be there anyway doesn’t cost much more. Air weighs a lot less than the dirt you would have to cart back in, there may be less concrete.
I’m not in the mood to find all the posts whining about flat junctions, deinterlining and why didn’t they ask you for your omniscience. Make up your mind are they good or are they bad.
Buildings are the epitome of not-in-the-street. Why is it okay to tear down buildings so yet more luxury one bedroom apartments can be near the train station but not okay to tear down buildings to make the trains faster and then build stuff over it? Stuff another million people into New York City something somewhere is going to be torn down. And they will need more trains.
They carved a new street through what is now the East Village and Soho for Layfayette Street. They carved a connection between Seventh Ave and Varick Street for the reconfiguring the subway for the Seventh Ave lines.
Greenwich Village was always gentrified. It was the far suburbs in the 18th Century and the vicinity of 6th and Waverly was always filled with rich people. Most of the carving was south of Houston. Which is in what has become SoHo. Which isn’t Greenwich Village. Neither is Tribeca.
Building it under the Sixth Avenue El raised costs. Beyond the boundaries of Greenwich Village and past the aspirations of the Hudson and Manhattan.