When a major infrastructure project overruns, the news covers the cranes and the concrete. The real mistake was often made months earlier, in a room full of documents, before anyone broke ground.
I spent eleven years as the group Chief Risk Officer of one of the world’s largest infrastructure firms, reviewing our biggest and most exposed projects across the globe. Before that I worked on the client side, ran an engineering group, and led bids on projects worth billions. Today I sit as a judge in the French commercial courts, where the disputes between contractors and their suppliers land on my desk. From every one of those seats, the pattern is the same. The most expensive errors in our industry are made at the bid stage, and they are almost always errors or misinterpretation of requirements.
Where does infrastructure risk actually come from?
A large tender is not one document. It is hundreds, sometimes thousands of pages: general standards & regulations, project-specific specifications, requirements carried over from other projects, technical appendices. Those documents are often produced by different teams and much of it is assembled under time pressure, and some of it is copied from earlier tenders without being checked for consistency. That is where the trouble starts. You end up with a requirement in one document that directly contradicts a requirement in another, and no team reading in parallel can hold the entire set in their heads at once.
What about the bid you never win?
The first risk on a mega-bid is not losing, it is the risk of the project not getting awarded at all. In such a case, the cost of bidding itself, on the largest design, build, finance and operate tenders runs into the millions. You commit that money up front, trusting the documents make sense. They do not always:
“The largest risk you run is that the client gets something wrong in the process, and you end up with a requirement in document 12 which is completely the opposite of the requirement in document 37. The client does not know how to manage it and cancels the project, while you have spent ten million dollars bidding for the job.”
There is also the quieter trap of requirements that state what to achieve but not how it will be measured. On one water project, the industry-normal practice was to demonstrate performance over a forty-eight hour window, which we could meet. At the last minute the client decided to measure over seven days instead. Same requirement, opposite outcome, because the measurement method was never agreed upfront.
How can a single requirement cost half a billion dollars?
Here is a real pattern, with no names. On a set of power projects, a civil works contract and a turbine manufacturing contract, bid separately were merged into one, with the civil contractor having to accept the turbine manufacturer as a subcontractor and their two efficiency requirements never matched. The consequences were severe:
“The requirement for the overall contract was 98 percent efficiency, and the requirement in the turbine subcontract was only 97 percent. On one project the companies went bankrupt, and on the other, the civil contractor had to issue a profit warning and post two huge quarterly losses, since it was listed. I think they lost 500 million dollars on those two projects, and the share price of that company dropped pretty significantly.”
Half a billion dollars, traced to a single percentage point that did not reconcile across two documents.
Why do these errors keep happening?
Partly because the documents themselves invite it. The general specifications are often assembled from older projects and never fully checked against the project at hand:
“That is when you start having problems with conflicting requirements, because often these documents are cut and paste from other projects, and people do not necessarily check the full consistency between the requirements specific to the project and the general requirements included in those cut and paste documents.”
And partly because the tools have not matched the scale. A modern bid is fifty people reading thousands of pages against a fixed clock, often across a consortium of up to ten firms who may never have worked together, each forming its own reading of the same tender. Writers like Ethan Mollick have argued that the limits of AI are rarely about raw intelligence and far more about context. The same is true here. When the context is fragmented across thousands of pages and ten organisations, errors are not bad luck. They are the predictable result.
What actually prevents them?
A single, reliable version of the requirements, read by an independent layer, held consistently from the first day of the bid through to delivery and handover. Every partner works from the same baseline. Contradictions surface before they are signed, not after they detonate. The commitments made to the client carry through to the supply chain, to commissioning, to the warranties that outlive the build. This is not a productivity gadget. It is risk infrastructure.
Why this matters beyond the construction site
For anyone looking at this industry as an investor, the lesson is direct. The failure mode I have described is the one that erodes margin and destroys equity value, and it is invisible on a balance sheet until it is not. A company that removes it is not selling speed. It is selling the prevention of the single most expensive class of mistake in one of the world’s largest industries. I have spent a career on the wrong end of this problem:
“For eleven of those fourteen years I was the group Chief Risk Officer, looking at our largest and most risky projects around the world. And I can tell you, I wish we had Aitenders at that time. That would have avoided a lot of problems.”
In over forty years I have watched a great deal of money disappear into problems that were created before a shovel hit the ground. The cranes were never the risk. The documents were.
Author: Regis Damour, Director