There is a cost in every bid that shows up on no line item. It was measured once, at national scale, more than twenty years ago. As far as I can tell it has never been measured on an individual job.
I wrote separately about why no two architecture firms draw a plan set the same way, and about the one buyer I could confirm has enough leverage to make anyone draw its way. What I did not get into is who actually pays for the difference, and how many of them there are.
It is not the estimator's problem, it is everyone's
Take a commercial building with steel. The structural details get read by the steel fabricator and the erector. The roofer reads a different part of the same set. So does mechanical, electrical, plumbing, fire protection, low voltage and technology, flooring, and paint.
Every one of those trades opens the same documents and goes looking for a different thing. Every one of them has to figure out how this particular firm organized it before they can find anything.
That is not one relearning cost per job. It is one per trade, per job, paid in parallel by people who never compare notes about it.
What actually varies is smaller and worse than people think
The differences that cost the most time are rarely dramatic. They are small and structural.
- The sheet index is not where you expect it.
- Rooms are called something slightly different than the last set.
- Sheet numbers are formatted one way here and another way there, and printed in a different corner.
- Some title blocks give you a sheet number and no sheet name, so the only way to learn what a sheet actually contains is to open it and look.
None of that is an error. Every one of those choices is defensible on its own. Together they mean nobody can navigate a new set on reflex, and everybody spends the first stretch of every job rebuilding a mental index that the last job already gave them for a different firm.
The file you are handed sets a ceiling on the estimate
The layout conventions are the visible half. The file itself is the other half, and it does something worse: it decides how accurate the number is allowed to be.
Sets arrive as PDF, as DWG, as a Revit model, as IFC. Which one is usually settled by the contract long before anybody downstream sees it. But the difference that actually bites is not the format name. It is whether the PDF carries real geometry or just pixels.
| What you are handed | What the estimator can do with it | What limits the number |
|---|---|---|
| PDF exported from the CAD or model file | Isolate the design layer and snap to objects | Nothing. The measurement is exact. |
| Scanned or flattened PDF | Set the scale by hand off a known dimension, then click pixels | Resolution and plotted scale |
| Vector with no searchable text layer | Measure, but not search | Every sheet has to be read by eye |
The takeoff vendors say the quiet part in their own documentation: measurements from raster files are rarely exact, and the error depends on the resolution and the scale. Their standing advice to customers is to request vector files if at all possible.
A choice made in an export dialog, by someone who will never watch anybody price the job, sets a ceiling on how precise the person pricing it can be.
Vector alone is not enough either. On one commercial set I worked through, the sheets carried no searchable text layer at all, and on a large share of the pages the drawings were rotated in ways that broke text extraction outright. You cannot search that set for a keynote or a room name. You read every sheet with your eyes and you keep the index in your head.
I do not know how common that is, and I want to be straight about it. I have not found anyone measuring what share of issued sets arrive vector, searchable, and correctly scaled. I only know what I have opened.
And none of it is an error. Every one of those exports is a defensible decision by somebody with a reason. It is just that the consequence lands two or three companies downstream from whoever made it.
The knowledge is real, and it is not written down anywhere
The people who are good at this are very good at it. Someone who has been reading sets for fifteen years knows which firms bury the finish schedule, which ones number rooms by floor, which ones you have to call because the detail callouts do not resolve.
Ask where that is written down. It is not. It is not in a procedure, it is not in the estimating software, it is not in a training document. It is in the head of whoever has been there long enough.
So when that person leaves, it leaves with them. They keep the knowledge and the relationships, and they will be just as effective at the next company. The company they left does not get to keep any of it.
A capability the business depends on, that took years to build, has no artifact anywhere. There is nothing to hand to the replacement.
The general contractor pays it twice
A GC carries this in two directions at once.
Going to market, they have to package and communicate a set whose conventions they did not choose, to a bid list of subs who each read it differently.
Then, coming back, they have to check whether the numbers they got are actually in line, which means holding every sub's read against a set that names things its own way. Anything the architecture firm does differently from the last one becomes something the GC has to absorb and translate before they can tell whether a bid is complete or a scope has a hole in it.
None of that work gets a code in the accounting system either.
It is paid per bid, not per job
Here is the multiplier nobody applies to this. The relearning does not happen once per project. It happens once per bid. An estimator works out how a firm organizes its set in order to price a job they will probably not win.
| Type of work | Industry benchmark bid to hit ratio | Times the set gets figured out per job won |
|---|---|---|
| Negotiated | About 4 to 1 | 4 |
| Private bid | About 5 to 1 | 5 |
| Public | 7 to 1 out to 11 to 1 | 7 to 11 |
Read that against the conventions problem. On public work, the set gets figured out somewhere between seven and eleven times for every job actually won. Exactly one of those has a project to absorb the cost. The rest come out of overhead, on jobs that went to somebody else.
The same benchmark advice says a ratio above roughly eleven to one is not sustainable, and the reason given is estimating expense. That is the industry stating plainly that the cost of preparing bids is a binding constraint on how much work a company can chase.
The relearning sits inside that constraint. It is one of the things making bids expensive to produce, and it is not itemized anywhere in the number the industry is watching.
Then the last part. In a survey of more than two thousand contractors, fewer than ten percent said they track their bid to hit ratio at all. So the multiplier that turns this from a per job cost into a per bid cost is, for most of the companies paying it, not a number anybody in the building knows.
Asking is not a reliable way out either
The implicit escape hatch in all of this is that you can stop and ask. That deserves a number too.
The 2013 Navigant Construction Forum analysis of requests for information looked at 1,362 projects and more than a million RFIs. It put the average cost to review and respond to a single RFI at $1,080, and the median reply at 9.7 days.
The figure that should stop you is a different one. 21.9 percent of those RFIs received no response at all.
More than one in five times, the formal mechanism the industry built for resolving an unclear drawing produced nothing. Not a slow answer. No answer.
Two caveats I am not going to bury. That study is from 2013, and its data skews heavily toward projects in Australia and New Zealand. And it measures RFIs raised during construction, not questions asked during bidding, which run through a different channel on a much tighter clock.
So it is not a measurement of anyone's bid window. It is evidence that the formal channel is expensive, slow, and unreliable, which is the reason people price around ambiguity instead of resolving it.
Why the software keeps choosing the closed door
This is also why so much construction software ends up asking you to work inside it rather than reading what you already have.
If you cannot control the format coming in, you cannot standardize on it. A homebuilder using a dozen outside architecture firms has a dozen conventions arriving, and no leverage over any of them.
Higharc's answer to that is coherent and worth understanding. Builders design in the platform, and the platform produces the permit ready construction documents and the plan based estimating data. Standardize the drawing by owning the drawing.
That genuinely solves the problem, and the trade is right there in the design. Everything has to be drawn their way, inside their tool. For a builder with the volume and the appetite to move, that may be a good deal. For everyone working from sets they did not commission and cannot dictate, it is not an option at all.
I do not think that approach is wrong. I think it is the honest consequence of an industry that never agreed on a format, and it tells you how hard the problem is that a credible answer to it is to replace the drawing rather than read it.
The only forcing function in the whole chain is the permit office
There is exactly one point where any of this converges, and it is not the design team or the GC. It is the authority having jurisdiction. The permit office requires a defined set of sheets to be submitted, so every project has to produce at least that much, in a form somebody will accept.
But notice what that actually standardizes. It sets a floor on what has to exist. It does not say where the sheet index goes, how rooms get numbered, or whether the title block tells you what the sheet is. And plenty of teams submit more than the minimum because the job needs the detail.
The one place the industry has real continuity gives everyone downstream almost nothing to navigate by.
What I could not confirm
Somebody did put a number on the aggregate. NIST commissioned a study pricing inadequate interoperability across the US capital facilities industry, and named a lack of standardization among its causes. That was measured in 2002.
I am not going to quote a figure that old as though it describes this year, and if anyone puts it in a sales deck you should discount it. The reason it is worth raising is the opposite of a statistic. The aggregate has been on the record, from a federal standards body, since before a lot of people currently estimating were in the trade, and it did not change what happens on any single job.
I could not find anything that prices this at the job level. Not per trade, not per GC, not per relearning. I looked.
A cost nobody has to own
What still does not exist is that cost appearing anywhere near where it occurs.
A change order gets coded to design error. A request for information gets logged under whatever it asked about. The hours somebody spent learning to read one firm's conventions before they could price anything get coded to estimating taking a while.
Nobody has to be hiding anything for this to happen. There is simply no field for it.
That is the thing I keep coming back to. A cost with a national estimate and no local line item is one everybody agrees is real and nobody has to answer for. Twenty years of knowing the number in aggregate has not put it on a single job's books, and it is very hard to fund a fix for something that never shows up where it actually happens.
It is also why I would rather build software that reads a set on its own terms than software that asks an industry to change how it draws. One of those has been tried for twenty years.