The solicitation is out, the pink team read the technical approach, and the finding is the same one that shows up on data and AI pursuits every year: the section that describes how the system actually gets built is thin, and the evaluator will see it. There are thirty days. The instinct is to write around the gap. The better move, if the gap is real, is to add a specialist subcontractor who can fill it with named engineers, a defensible estimate and text that fits your template. Thirty days is enough. It is enough only if the first three days are spent correctly, so this is the plan day by day, and the short list of things you have to hand over on day one.
We build AI systems, data platforms, cloud infrastructure and full-stack applications and deliver them into production inside federal agencies. We join pursuits at this stage regularly, because this is when a capture team finally knows what it is missing. What follows is written for the capture director making the call, not for the specialist.
First decide whether the gap is real
Adding a sub late costs you contracts time, a redistribution of workshare, a subcontracting plan update in some cases, and a new set of representations. It is worth it when the gap is one of four kinds. It is not worth it when the gap is writing quality, which you fix with your own proposal staff.
A capability gap. The requirement asks for something your team has not built: a retrieval system over agency documents with citation to source, an evaluation suite with measured thresholds, a data platform that has to survive an authorization process, a model lifecycle with drift monitoring. You can describe it. You cannot show it.
A personnel gap. The solicitation names key personnel positions and your bench does not have a current, relevant résumé for one or two of them. Evaluators score currency, and a résumé whose most recent relevant project ended four years ago reads as exactly that.
An evidence gap. Your approach is sound and there is nothing in the volume that proves anyone has done it. No measured result, no artifact, no demonstration. On AI work in particular this is what separates a section that scores from one that reads like a plan.
A small business participation gap. Your subcontracting plan or the evaluation criteria call for meaningful small business content and what you have is a percentage attached to nothing specific. A named specialist with a scored technical scope is materially better content than a percentage.
If none of those four describe the problem, do not add a sub thirty days out. If one or more do, move immediately, because the schedule below has no slack in the first week.
What makes a thirty-day sub addition work or fail
Editorial weighting, illustrative rather than measured. The last row is low because a nondisclosure agreement plus a short letter of intent lets drafting start while contracts finishes the paper.
Days one to three: the handover
This is the whole schedule. A specialist who has the right material on day one produces a draft section in week two. A specialist who receives material in pieces over ten days produces a draft in week four, and week four is too late for a red team.
Send, in one package:
- The solicitation and every amendment, including the instructions to offerors and the evaluation criteria, not a summary of them. The specialist writes to Section M, and a paraphrase loses the words that matter.
- Your current technical approach draft, even if it is rough, plus the sections around the gap. The specialist needs to know what has already been promised so the new text does not contradict it.
- The template, style guide, page allocation and font rules. Text written to the wrong page budget is text that gets cut by somebody who does not understand it.
- The win themes and any discriminators already agreed. The specialist's section should carry them rather than introduce competing ones.
- The shape of the pricing model. Not your rates. Whether the estimate is labor hours by category, milestone prices, a level of effort against a stated period, and what the government asked for.
- The review calendar with your internal dates, and the name and contact of the engineer who can answer a technical question in a day.
- Anything you know about the incumbent, the program history and the customer's stated frustrations. This is the material that turns a competent section into a persuasive one.
Do this under a nondisclosure agreement, which takes a day, and a short letter of intent that describes the anticipated scope and states that each party bears its own bid costs. That is enough for drafting to start. The full teaming agreement can be negotiated in parallel during week two.
Days four to ten: scope, people and the first draft
By the end of the first week and a half, the specialist should have delivered four things, and you should hold them against that date rather than a softer one.
A written workshare scope. The tasks the specialist performs, mapped to your work breakdown structure and to the evaluation criteria the section addresses, with the deliverables and any contract data requirements named. This is the document your pricing team and your contracts group both work from, and it is what goes into the teaming agreement.
Named key personnel with résumés in your template. Not a bench summary. Named individuals, with a signed letter of commitment for each, an allocation percentage, and a period of availability after award. If the solicitation names key personnel positions, the résumés should be structured to answer the stated qualifications point by point.
A first draft of the technical section. Written in your template, inside the page allocation, in your voice. This is the deliverable that tells you whether the addition was a good idea. A first draft that arrives as a set of bullet points and a promise to expand them is a warning.
A risk register entry set for the specialist scope. The three or four risks specific to the technical work, with mitigation described in terms of what gets built and when rather than in terms of management attention.
Days eleven to eighteen: the demonstration, which is the actual discriminator
This is the part most primes do not ask for and should. On an AI or data pursuit, the difference between a section that scores and one that does not is usually evidence that the approach has been run rather than described. A specialist working from public data can build a working demonstration inside a week, and the artifacts it produces go straight into the volume.
What that looks like concretely. Take the public corpus closest to the agency's actual documents: a regulation set, a public rulemaking docket, an agency's published reports, an open dataset in the mission area. Build the ingestion, the chunking and the index. Stand up retrieval with citation back to the source paragraph. Write an evaluation set of a few hundred questions with known answers, and score retrieval and answer quality against it. Record the numbers.
Then the volume can say something specific: here is the architecture, here is a measured result on a public corpus of comparable structure, here is the evaluation suite the government inherits, and here is what changes when it runs against agency data. That is a paragraph an evaluator can score. The alternative paragraph, which says the team will use retrieval-augmented generation with appropriate evaluation, is a paragraph every offeror wrote.
The same move works outside language models. For a data platform, build the schema and load a public dataset of the right shape, then show query performance and the lineage model. For a modernization, build the interface against a public API and show the accessibility conformance testing. The point is identical: replace a claim with an artifact.
Two constraints keep this honest. The demonstration uses public data only, and the volume says so plainly. And the numbers reported are the numbers measured, with the corpus and the method described well enough that a technical evaluator could repeat them. An unsourced accuracy figure does more damage than no figure at all.
Days nineteen to twenty-four: basis of estimate and price
The pricing team has to defend the specialist's number, and a number arrived at by applying software labor categories to an AI scope will not survive price realism. The basis of estimate the specialist owes you has five parts.
Task decomposition. The work broken into elements small enough that each has an estimating rationale, and each mapped to a deliverable.
Data work stated separately. Acquisition, cleaning, labeling where labeling is needed, schema design and pipeline construction. On most AI programs this is the largest single block of effort and the one most often folded invisibly into development hours, which is exactly what makes an estimate indefensible.
Evaluation effort stated separately. Building the evaluation set, the scoring code, the human review protocol where one is needed, and the regression runs. Evaluation is not a testing line item borrowed from a software estimate.
Iteration cycles counted. Model and pipeline work does not converge on the first attempt. Say how many cycles are assumed for each element and what triggers another one. An estimate with no iteration assumption is an estimate that will be exceeded.
Compute and tooling as a separate line. Training and inference cost, the environment, and the licensed tools, with the sizing assumption written out.
Under each element the labor mix should match the technical approach. A pipeline-heavy scope is data engineering hours, not research hours. A retrieval and evaluation scope is a different mix again. When labor mix matches scope, the estimate reads as built rather than back-solved, and that is what price realism is checking.
| Day | Prime owes | Specialist owes | Failure mode if skipped |
|---|---|---|---|
| 1 to 3 | Solicitation, draft approach, template, page budget, win themes, review calendar, an engineer to call | Signed nondisclosure agreement, named team, first questions | Drafting starts in week three and misses red team |
| 4 to 10 | Answers inside 24 hours, work breakdown structure, pricing model shape | Workshare scope, key personnel with commitment letters, first draft section, risk entries | Contracts and pricing both stall for want of a scope document |
| 11 to 18 | Feedback on the draft, access to any releasable reference material | Working demonstration on public data, measured results, figures for the volume | The section describes an approach instead of proving one |
| 19 to 24 | Rate structure guidance, price realism expectations | Basis of estimate with data, evaluation, iteration and compute broken out | Pricing cannot defend the sub's number and cuts it arbitrarily |
| 25 to 30 | Red team comments, final page allocation, signature-ready teaming agreement | Revised text to page, final résumés, representations, compliance matrix entries | Last-week edits by someone who does not understand the technical content |
Days twenty-five to thirty: red team, page fit and paper
The last week is compression and compliance. Three things go wrong here and all three are preventable.
The first is that red team comments come back to the prime's proposal staff and the specialist never sees them. The specialist wrote the section and is the only person who can cut it without breaking the technical logic. Route the comments directly and give a 48-hour turn.
The second is page fit. The specialist's section will be over. Cutting technical text to fit a page budget is a skill, and the wrong cuts remove the exact sentences that were scoring. Tell the specialist the final allocation, let them make the cut, and review the result rather than making the cut for them.
The third is the paperwork tail: representations and certifications, the subcontracting plan data, past performance information the specialist may cite, insurance certificates, and the signed teaming agreement. None of it is hard and all of it is the thing that is discovered at 9pm on the due date. Ask for the full package by day twenty-six.
What you are actually buying, and what stays yours
A specialist added at this stage is not a body shop and should not be priced like one. What you buy is a scored section, named people who will be there, an artifact that proves the approach, an estimate your pricing team can defend, and a partner who performs the scope after award.
What stays with you is everything that matters to your business. The customer relationship is yours and the specialist does not contact the contracting officer or the program office except through you. The prime position, the program management and the contract are yours. The code and data produced under the subcontract carry whatever assignment or license the agreement specifies, agreed before submission rather than after, and the specialist's pre-existing tooling is listed on a schedule so nothing is ambiguous at delivery. Your template, your voice, your win themes.
On price shape, we work two ways. Fixed-price milestones where the scope is defined well enough to write acceptance criteria as tests, which is most build work. A committed team at a stated allocation where the work is discovery, sustained engineering or task-order response. For the proposal phase itself, bid costs sit with each party in the ordinary way.
The first step is one email with a one-page brief: the solicitation number, the gap in the technical approach, the shape of the workshare you have in mind, and the submission date. We return a scoped, priced statement of work, a workshare description you can drop into a teaming agreement, and the names of the engineers who will do the work.
Which gaps justify adding a subcontractor this late
Editorial weighting, illustrative rather than measured. The last row is low because writing quality is fixed faster in house.
How the two teams should be wired for four weeks
The mechanics of a short pursuit matter more than the org chart. Three arrangements do most of the work.
One technical counterpart on each side, named. Your chief engineer or solution architect, and the specialist's lead engineer. Every technical question routes through that pair. When questions route through business development on both sides, the answer takes three days and arrives changed.
A shared document location with the current template in it. The specialist should be writing into the same file structure your proposal manager compiles from, not emailing revisions. Version confusion in the last week is what causes a section to ship a draft older than the red team comments.
A standing fifteen-minute check, not a weekly meeting. Short and frequent beats long and scheduled on a thirty-day clock. Most of these calls end in five minutes, which is the point.
One more thing worth stating explicitly at kickoff: the prime decides the final content. A specialist with strong architectural views is who you want writing the section, and also who can spend two days defending a design choice the evaluation criteria do not reward. Saying up front, in writing, that the prime arbitrates technical disagreements on proposal content removes that risk entirely, and a partner who accepts it readily is telling you how they will behave on the program.
Three ways a late addition goes wrong
The specialist is asked to review rather than write. A review produces comments. Comments produce a rewrite by someone who did not do the work. If the gap is technical content, ask for technical content and give the page allocation to the person who can fill it.
The teaming agreement is negotiated before drafting starts. Contracts and drafting are parallel tracks. A nondisclosure agreement and a two-page letter of intent are enough to start, and every day spent waiting for a fully negotiated agreement is a day taken from the demonstration.
The specialist's price is cut without changing the scope. This happens in the last week, quietly, and it produces a subcontract nobody can perform. If the number has to come down, cut a task and say which. A specialist who accepts a silent cut on a defined scope is telling you what performance will look like.
Bottom line
Thirty days is enough time to add a specialist subcontractor and come out with a stronger volume, but only if the handover happens in the first three days and the specialist writes rather than reviews. Give the solicitation, the draft approach, the template, the page budget and an engineer who answers within a day. Expect a written workshare scope, named key personnel with commitment letters and a drafted section inside ten days. Spend the middle week building a demonstration on public data so the section proves the approach instead of describing it. Spend the fourth week on a basis of estimate that breaks out data, evaluation, iteration and compute. Leave the last week for red team and paper. The addition is worth doing when the gap is capability, personnel, evidence or small business content, and it is not worth doing when the gap is writing.
Frequently asked questions
Yes, if the handover happens in the first three days. The schedule that works is: nondisclosure agreement and full document package in days one to three, workshare scope and named key personnel with commitment letters by day ten, a drafted technical section by day ten and a working demonstration by day eighteen, the basis of estimate by day twenty-four, and red team plus paperwork in the final week. The teaming agreement is negotiated in parallel rather than before drafting. What makes thirty days fail is receiving the solicitation and template in pieces over the first two weeks.
The solicitation and all amendments including the instructions and evaluation criteria, the current technical approach draft and the sections around the gap, the proposal template and style guide, the page allocation, the agreed win themes, the shape of the pricing model, the internal review calendar, and the name of an engineer who can answer a technical question within a day. Also anything known about the incumbent and the customer's stated frustrations. Sending a summary instead of the solicitation costs a week, because the specialist writes to the evaluation criteria and paraphrase loses the words that matter.
On AI and data pursuits it is usually the highest-value week of the thirty. A working demonstration built on public data of comparable structure produces measured results, architecture figures and evaluation code that goes directly into the technical volume, and it lets the section prove an approach rather than describe one. Two rules keep it defensible: use public data only and say so, and report only numbers actually measured, with the corpus and method described well enough for a technical evaluator to follow.
Five parts. A task decomposition where each element maps to a deliverable and has its own estimating rationale. Data work broken out separately, covering acquisition, cleaning, labeling, schema and pipelines, because on most AI programs it is the largest block of effort. Evaluation broken out separately, covering the evaluation set, the scoring code, human review where needed and regression runs. A stated number of iteration cycles per element with what triggers another. And compute and tooling as their own line with the sizing assumption written down. The labor mix under each element should match the technical approach.
When the gap is writing quality, which your own proposal staff fixes faster. When the requirement is small enough that the contracts, subcontracting plan and representations work costs more than the scope is worth. When nobody on the capture team can be freed to answer questions within a day, since an unanswered specialist stalls immediately. And when the intent is to have the specialist review the existing text rather than write new content, because a review produces comments and comments produce a rewrite by someone who did not do the work.