The checklist you have is the wrong checklist

Every bid team has a checklist. It lives in a spreadsheet, it has a tab per schedule, and it is almost entirely a compliance checklist: forms attached, signatures obtained, word limits respected, file naming convention followed, portal upload confirmed before 2pm.

That checklist is necessary and it is not the one that decides whether the response is any good. The one that decides quality is a production checklist: who drafts each section, from what material, against what standard, and who has the authority to say it is finished. That checklist usually does not exist on paper. It exists in the bid manager's head, and it gets improvised on every tender.

AI walked into that gap. Nobody decided which column it belonged in, so it ended up in both and neither. One firm hands it everything and then spends two days auditing prose against the tender pack. Another bans it for anything client-facing and writes 11 schedules by hand in nine days. Both are responses to the same missing decision.

The right axis for the split

The usual instinct is to split by how interesting the writing is. Boring sections go to the AI, strategic ones stay human. That feels right and it is the wrong axis, because plenty of boring sections carry legal weight and plenty of interesting ones are fully determined by material you already hold.

Two questions sort it properly.

The source test. Is the answer already contained in material you can put in front of the system: the tender pack, prior engagement reports, your capability statements, the transcript of the bid kickoff? If yes, drafting it is a transcoding job, and transcoding is what a source-grounded workflow is for.

The accountability test. If this sentence turns out to be wrong, who takes the call from the client, and what does it cost? A misdescribed methodology costs you an embarrassing correction. A misstated referee outcome, an overcommitted resource, or a warranty you cannot meet costs you the panel seat, or worse.

Source-determined and low-stakes goes to the AI first pass. Judgement-determined or signature-bearing stays with a named human from the beginning. Almost every argument inside a bid team about “can we use AI for this” resolves in about thirty seconds once you ask those two questions in that order.

What AI handles

1. Requirement enumeration and the compliance matrix

This is the highest-value item on the list and the one most teams do last. Before anything is written, every requirement, mandatory criterion, evaluation weighting, and schedule question needs to be pulled out of the pack and listed. On a 120-page state government ICT panel RFT that is typically 60 to 90 discrete obligations spread across the conditions of tender, the statement of requirements, and three annexures.

Extraction is exactly the job to hand over. The trap is asking for the matrix as prose. Enumerate the items first as a list, confirm the list is complete, then draft one entry per item. That ordering is the difference between a matrix that maps to the pack and one that reads well and silently drops item 47. We have written about why list sections need to be drafted one item at a time in more detail.

2. Methodology, approach, and delivery model sections

These are the long ones, they are usually 40 to 60 percent of the word count, and they are almost entirely source-determined. Your delivery approach is not invented per tender. It exists in prior proposals, in the last three engagement reports, in your internal method documentation. The tender-specific work is tailoring it to this client's scope and constraints, which is a rewrite against a structure, not original composition.

3. Company background, capability, and standing policies

Corporate profile, quality system description, WHS approach, insurance summary, environmental policy statement. Stable content that has to be reshaped into whatever structure and word limit this particular buyer asked for. If your firm has captured its voice, glossary, and company identity once as reusable workspace assets, these come out consistent instead of carrying whichever tone last year's bid writer used.

4. Reshaping prior answers into this tender's structure

Most firms have an answer library. Most answer libraries are a folder of previous submissions. The work is never copy-and-paste, because this buyer asked a differently shaped question with a 500-word cap and their own terminology. That reshaping is mechanical, high-volume, and exactly where hours disappear on a compressed timeline.

5. First-pass answers to schedule questions

Where the source material genuinely covers the question, a first pass gets the bid team out of the blank page and into editing. This is the whole argument for measuring time to first draft rather than total drafting hours. A complete rough draft on day three leaves ten days to improve it. A polished draft on day eleven leaves three.

What humans still own

1. Win themes and bid strategy

What this client actually cares about, which of your three differentiators to lead with, whether to bid the compliant option or the alternative, whether to bid at all. This comes from the client relationship and from reading the room during the industry briefing. No amount of source material substitutes for it, and every section downstream inherits it.

2. Pricing and commercial terms

Rates, margin, risk loading, payment milestones, what you will and will not accept on liability caps. A commercial decision, made by someone with the authority to make it, entered deliberately.

3. Named personnel, availability, and CVs

Naming a principal consultant on a bid is a commitment to their calendar in four months. The person who knows whether that commitment is real is the resourcing lead, not a document workflow. Scrubbing CVs to match evaluation criteria is a human judgement about emphasis, and it is the section evaluators most reliably check against reality.

4. Past performance and referee claims

Every specific claim about a previous engagement (the value, the outcome, the percentage improvement, the client's name) is a verifiable statement of fact about a third party. These need a human who was there, or a document from the engagement, standing behind them. This is the single most common place where a plausible-sounding draft becomes a real problem, which is why the difference between source-grounded and hallucinated output matters more here than anywhere else in the response.

5. Anything you sign

Declarations, conflict-of-interest statements, warranties, departures from the conditions of contract, the tender response form itself. A signature is the accountability test answering itself.

6. The final compliance read

One named person, reading the submitted response against the tender pack and every addendum, with the matrix open. Not the person who wrote most of it. This is a two-hour job that people try to compress into twenty minutes on submission morning, and it is where responses get disqualified on technicalities that had nothing to do with the quality of the work.

The four handover points where bids go wrong

The split above is not the hard part. The hard part is the seams between the two columns.

Addenda issued mid-bid. Buyer issues addendum 3 on day eight, changing a mandatory criterion and extending one schedule. If your drafting material is a folder someone downloaded on day one, half the response is now answering the old question. Whatever tool you use, the source set has to be updatable mid-flight and the sections drafted before the change have to be re-checked, not assumed.

The answer copied from a losing bid. An answer library has no memory of outcomes. Reused content propagates whatever was weak about the submission it came from. Someone has to decide what goes in the library, and that decision is a judgement call about quality, not a file operation.

The question that got answered silently. When something genuinely is not in the source material (the client never stated a transition timeframe, the pack is ambiguous about who provides the test environment), the useful behaviour is to surface the question, not to write a confident paragraph around the gap. A clarifications step that puts the open question in front of the bid manager, and feeds the answer forward into every later section, is the mechanism that keeps gaps visible instead of buried in prose.

The executive summary written first. Because it is the first thing in the document, it gets drafted before the response knows what it says, and then never properly rewritten. It should be the last section produced, against the finished response and the win themes, by whoever owns the relationship.

The checklist, in order

Run it in four stages. The label in brackets is who owns the first pass.

Stage 1: intake, before anything is drafted.

Bid/no-bid decision recorded, with the reason [human]
Full requirement list extracted from every part of the pack, including annexures [AI, human confirms completeness]
Evaluation criteria and weightings listed against the sections that answer them [AI first pass]
Win themes agreed and written down in one paragraph [human]
Source set assembled: pack, addenda to date, prior relevant engagement material, kickoff transcript [human]
Named owner for pricing, personnel, and the final read [human]

Stage 2: drafting.

Compliance matrix drafted one requirement at a time [AI]
Methodology, approach, and delivery sections drafted against the template, section by section with a review gate on each [AI first pass, human reviews before the next section runs]
Corporate, capability, and policy sections [AI]
Open questions surfaced and answered before they propagate [human answers]
Pricing schedule [human, in parallel]
Personnel, CVs, availability confirmed [human, in parallel]
Past performance claims verified against real documents [human]

Stage 3: review.

Every claim about a third party traced to a source [human]
Word and page limits checked per schedule [either]
One voice across all schedules, not four [AI assists, human decides]
Executive summary written now, not earlier [human]
Red team read by someone not on the bid [human]

Stage 4: submission.

Full compliance read against the pack plus every addendum, by one named person [human]
Signatures, declarations, forms [human]
File formats, naming, portal upload, confirmation receipt saved [human]
Template and reusable assets updated with anything learned [human, 30 minutes, almost always skipped]

How to check your own split

Take your last submission. List every section. Mark each one S if the answer was determined by material you already held, or J if it required a judgement someone had to stand behind. Then write the hours each one took.

In most bid teams the S column is 60 to 70 percent of the hours and close to none of the differentiation. That is the column to move. The J column is where the bid is actually won, and it is usually the column that got squeezed because the S work ran long.

SkyDraft is built for the S column specifically: a template that holds your response structure and the standard each section has to meet, your source material read as context, section-by-section generation with a review gate you cannot skip, and a clarifications step that asks rather than invents. The J column stays exactly where it is. See the how-it-works walkthrough for the mechanics, or the tender and RFP use cases for where firms are pointing it first.