Reading Bid Packages: What Five Real Ones Taught Us

I have seen a lot of opinions about how messy bid documents are, and almost no numbers. So we decided to stop guessing. We took five real public bid packages, ran every page of them through our software, and counted what came out.

They are all public documents, published by cities and towns, and anybody can download them. Nothing here comes from a customer file. That matters, because it means you can go and check any of it yourself.

What follows is what we found. Some of it confirmed what everybody in this industry already suspects about reading bid packages. A couple of things surprised me.

The Five Packages

The range is deliberate. Two of them are short invitations that any person could read over coffee. Two are full specification books. One sits in between.

I want to be straight about what that range means. The short ones are not evidence of anything.

A four page invitation does not need software. We included them because they show what a package looks like at the other extreme, and because two of them contained the single most important sentence in the whole document, which I will come to.

Finding One: The Page Count Is Not the Problem

The Idaho package was six hundred and sixty eight pages. Our software read all of them and decided that fourteen sections mattered for process equipment. Everything else was concrete, paving, electrical distribution, general conditions and permit attachments.

Fourteen out of roughly a hundred and forty sections. That is about ten percent of the book.

The New York package was a hundred and seventy nine pages, and seven sections mattered. Same story.

So the difficulty is not the volume of paper. It is that the ten percent you need is scattered, and it does not announce itself. In the Idaho book the relevant sections were spread across three different divisions, and two of them had titles that gave no hint they were relevant to the equipment being quoted. Somebody skimming a table of contents to decide what to read would have skipped them.

Finding Two: The Documents Disagree With Themselves More Than You Would Expect

In the New York package, our software found six places where the documents contradicted each other. Not ambiguities. Direct contradictions where two statements cannot both be true.

Three of them mattered enough to change what you would build:

• Deck material on the gangway. The specification, section 35 50 20 on page 102, called for one product. The drawing sheet CM-5.01, on page 117, called for grooved timber. Different material, different weight, different way of fixing it down.
• Clear width between the handrails. The specification asked for a minimum of forty eight inches. The drawing said sixty inches minimum. That is a different structure, not a detail.
• Deflection limit on the dock. One paragraph said the deflection must not exceed L over 360. Another paragraph said L over 180. Both on the same page. One of those is twice as strict as the other, and it drives the whole structure design.

That last one is the one I keep thinking about. It was not a case of the drawings disagreeing with the specification. It was one page disagreeing with itself, which means it survived the engineer writing it and everybody who reviewed it afterwards.

Finding Three: The Most Important Number Is Often Not in the Specification

In the same New York package, the length of the gangway does not appear anywhere in the specification. The section simply says to see the plans.

The number is on a drawing sheet, in the corner of a profile view. Forty six feet.

This pattern showed up again and again. The specification handles the words, the drawings hold the geometry, and the two are written by different people who assume you will read both. If you read only the specification, you have a document with a hole in it, and the hole is the most basic number in the job.

Our software goes and gets it, and marks it as coming from the drawings rather than the specification, so the estimator knows the specification never actually said it. That distinction sounds pedantic until the customer changes the drawing in an addendum.

Finding Four: One Sentence Can End the Whole Exercise

Here is where the two short packages earned their place.

The Odessa, Missouri invitation is four pages long. On page two, in the middle of the instructions to bidders, sits this sentence: substitute materials or work shall not be permitted. That single line means the manufacturer named in the specification has already won. Any other company that reads pages one and three, prices the equipment, and submits a bid has donated the time.

The Michigan bid does something similar in the other direction. It names the equipment down to the model number, "fine bar screens FPFS-M 304 stainless steel by Duperon or approved equal", and then in the same two pages it also tells you that the existing equipment must be removed, that concrete floors and walls will need partial demolition, and that a new exhaust duct is required. That is a lot of scope for two pages, and it is scope most people would miss if they were only looking for the equipment specification.

The Idaho book has the subtle version of the same thing. In the quality assurance paragraph, several pages before anything about materials, it requires the manufacturer to have at least ten installations of identically sized equipment. Not ten installations. Ten of the same size. That clause decides who can bid, and it sits nowhere near the parts of the document an estimator would naturally read first.

We wrote more about that pattern in Or Approved Equal: When to Bid and When to Walk Away.

Finding Five: Some Contradictions Are Deliberate

This was the finding that changed how we built the software.

In the New York package, the floating dock decking appeared as composite in one place and treated pine in another. On the face of it that is another contradiction, and an earlier version of our software reported it as one.

It was not. The bid form carried composite and timber as separate priced alternates, which means the owner wanted a number for each and intended to choose. The specification was doing exactly what it was supposed to do.

Meanwhile the gangway decking contradiction, in the same package, had no alternate behind it anywhere. That one was real.

Two differences that look identical in a list. One is the owner making a decision on purpose. The other is a mistake that needs a question asked before anybody prices it. Telling them apart means reading the bid form and the technical sections together, which is not how most people read these documents, because the bid form is usually treated as paperwork rather than as information.

Our software now checks for this and says so in plain words on the screen. It is the single most useful line it produces.

Finding Six: The Requirements Are Not Where You Would Guess

We track a few dozen values that decide the price of a job. Here is roughly where they turned up.

• Materials and standards: almost always in the technical section, as expected
• Dimensions: split between the technical section and the drawings, with the drawings winning any argument in practice
• Loads and performance: technical section, but frequently in a table that is easy to skim past
• Testing and factory acceptance: deep in the technical section, near the end, where attention is lowest
• Manufacturer field service days: near the very end of the section, and it is real money
• Bonds, damages, delivery dates: commercial documents, which the technical reader usually never opens
• Whether you can bid at all: quality assurance paragraph or instructions to bidders, both far away from everything else

Look at that list again. The three things most likely to hurt you, which are the qualification clause, the field service obligation and the liquidated damages, all sit in places an engineer reading for scope would not naturally go.

In the Idaho book, the field service requirement was a factory-trained representative on site for two consecutive eight hour periods during installation, plus two four hour training sessions during startup. That is several days of somebody's time, in a paragraph on page six hundred and sixty six.

What Reading Bid Packages Actually Involves

If I had to compress five packages into one sentence, it would be this.

The work is not reading a lot of pages. The work is finding perhaps forty statements that are deliberately spread across a document written by four different people who never met.

That is why the job takes an experienced person most of a day, and why handing it to a junior does not work. It is not about reading speed. It is about knowing that the qualification clause exists, that the bid form changes the meaning of a contradiction, and that the specification will hand you off to the drawings without warning.

It is also why we think reading bid packages is a good problem for software, and a bad problem for a chat window that only sees the pages you paste into it. We wrote about that in Using ChatGPT to Read Bid Packages.

The Honest Limits

A few things this exercise does not prove.

Five packages is not a study. It is five packages. The patterns were consistent, but I am not going to tell you that ninety three percent of bid documents contradict themselves, because I do not know that and neither does anybody else.

Four of the five were water and marine work. Different trades will have different habits. I would expect the same underlying pattern in any engineered-to-order world where a third party writes the documents, but I have not measured it yet.

Our software gets things wrong too. On an earlier run of one package it reported eighteen contradictions, and most of them were not contradictions at all. They were long lists of requirements, like every component in a machine having its own material, which the software was comparing against each other.

We fixed it and the count came down to two, which was the truth. Any tool in this space, ours included, should be judged on how it behaves when it is wrong, not only on what it finds when it is right.

What Mavlon Does

Mavlon is an AI quoting platform for engineer-to-order manufacturers. It reads the package your customer sends, in whatever form it arrives, and produces a record where every requirement carries the page it came from. Then it connects that to your own past quotes, in whatever condition they are in, so the estimate is anchored to jobs you have actually built.

If you would rather watch than read, there is a four minute walkthrough of the New York package here: Spec Says 48 Inches, Drawing Says 60. It reads all one hundred and seventy nine pages in forty eight seconds.

Also worth reading: AI Quoting for Gangways and Docks, which walks through how the reading step works in more detail.

Want the same run on a package from your world? Send me a public bid and I will do it. Book a 30-minute demo.

Book a 30-Minute Demo →