A bid package is a scope boundary before it is a document

A bid package is the set of work you have decided to buy from one subcontractor, under one contract, on one basis. The drawings, specification sections, scope sheet, schedule dates and commercial terms issued with the invitation are the record of that decision. The decision itself is a line: this side is yours, that side belongs to somebody else. Everything expensive happens at the line.

Inside a package, scope is well understood — the trade has priced its own work a thousand times. At the interface between 2 packages, both bidders read the same detail and each concluded the other one had it. Nobody bought the work, and you find out during installation at a price nobody competed for.

What goes in the package, and why that is the easy part

• The invitation and bid form, stating what is being priced, in what format, by when, and what alternates and unit rates are wanted alongside the base number.
• Drawings listed by sheet number and revision, not as a folder. A package that says see the drawings has already made revision control somebody else's problem.
• The specification sections that apply, with the general and supplementary conditions that come with them.
• The scope sheet — the per-trade question set issued before bids arrive, which is what makes the returns comparable at all, argued in the scope sheet that makes bids comparable.
• Schedule dates and sequence: mobilisation, the work window, any phasing, and what the package must be finished around.
• Commercial terms, including the insurance and bond requirements the bidder has to price — and what actually proves that cover is the endorsement rather than the certificate.
• The addendum protocol: how changes reach bidders and how acknowledgement is recorded.

None of that is difficult to assemble, and a package containing all 7 can still be defective, because the contents describe the work and say nothing about its edges.

Where the line actually comes from

Three things should decide it. The market: what firms in your region bid as one job, because a package nobody self-performs gets brokered and you pay the broker. The schedule: work that must be sequenced together belongs together. And risk: what you want behind a single accountable party when it goes wrong at 7 in the morning.

There is a fourth pressure, and it is the wrong one — the way the drawing set is organised. A package split along drawing disciplines rather than along trade markets produces packages nobody can price cleanly, and the symptom is a bid spread far wider on one package than on its neighbours. How many firms to invite for the package you end up with is a separate decision, and widening the list is usually not the fix, which is the diagnosis in you invited forty subs and three of them bid.

Where scope gets bought twice, or not at all

• Supply against install. One package furnishes the item and another fixes it in place. Answered as a single question, one half of it goes unpriced.
• Backing, blocking and support. Anything one trade fixes to something another trade provides — brackets, hangers, plinths, framing behind a fixing.
• Penetrations. Who cuts, who sleeves, who seals, who fire-stops, who patches, and who inspects. That is 6 verbs and they routinely land in 3 packages.
• Final connections. Equipment delivered by one package and connected — power, water, drainage, controls, data — by another, usually 2 weeks after the delivering trade has left.
• Access, protection and clean-up. Hoisting, temporary works, protection of finished surfaces, and final clean. Every bidder assumes somebody else has these because on their last job somebody did.
• Testing, commissioning and handover documentation. Who witnesses, who records, who compiles the operating manuals, and who attends the demonstration.

Each has the same shape: 2 packages, 1 item, and a verb belonging to exactly one of them. Write the verb down.

The interface register: one row per handoff

1. List the packages, then list every pair that physically touches. Most pairs do not; on a typical building the ones that do come to somewhere between 8 and 15.
2. For each touching pair, walk the 6 item classes above and write a row for anything that exists on this job. A pair with no rows is a pair somebody has not thought about yet.
3. Answer supply and install separately on every row. Where the same package does both, write that down anyway — the row saying so is what stops the question being reopened in month 4.
4. Push each decided row into the scope sheets of both packages using identical wording. Wording that differs between 2 packages is a gap that has been formally documented.
5. Issue the open rows as explicit assumptions with the invitation, and require bidders to state theirs. Then read those statements against each other when the bids land.

Nobody loses scope in the middle of a package. It goes missing at the edge, where two bidders each read the same detail and both looked away.

What the returned bids tell you about the line you drew

The bids are a free audit of the boundary. Every bidder excluded the same item: the interface was never assigned. One number is 30 percent off the others: they read the boundary differently, and you need to know which reading is yours before deciding who is cheap. The same clarification arrives from 3 firms: the package is ambiguous, and the answer belongs in an addendum to everyone.

Requirements travel with the package too. What a bidder priced against — insurance, bonding, the compliance evidence they will have to keep current for the whole job — becomes a date-tracking problem the moment they are awarded, and the way that fails is an expired certificate found after the sub is on site.

Practically, the register is a table with 8 columns and a rule that both scope sheets read from it. Extraction can draft rows out of the specification sections — the furnished-by-others language, the responsibility matrices — leaving a person to keep or kill each one, which is the pattern we scope as AI agents and automation. It only works if the people who own the interfaces maintain it, and a register in a tool nobody opens is the preconstruction version of crews installing the app and still filling in paper. Unowned handoffs behave the same way in other operations, which is the recurring theme in AI in logistics operations. This page sits in bidding, prequalification and subcontractor management, part of our construction and contracting work.

• bidding
• scope
• preconstruction
• subcontractors