The question we get on almost every first call
Somewhere in the first twenty minutes, the other side asks a version of the same thing: what do you need from us to move? It is a fair question and it deserves a written answer, so here it is. Four things decide whether a partnership works from where we sit. A scope with an edge on it. A person who can decide. Data we can touch on a named date. And a partner willing to say the hard part out loud in the first exchange rather than the fourth. Nearly everything else is negotiable, including price, including workshare, including who holds the prime position.
We are publishing this because naming it saves everyone time. A capture manager can read this page in three minutes and know whether to send us a bid. A faculty researcher can tell whether a call is worth scheduling. And the partners who read it and think "that is exactly how we already work" are the ones we do our best work with. Filters cut both ways, which is the point of writing one down.
What moves a first conversation to signed work
Editorial weighting from our own partnering practice — illustrative, not a measured statistic.
A scope with an edge on it
The best scope we have ever received was three sentences long. It named four data feeds, said what came out the other end, and drew a line: "You own everything upstream of the case-management screen. We own the screen." We priced it the same day. The length was not what made it good. The boundary was.
A scope with an edge tells us what is ours and what is yours. It also tells us what is nobody's, which is the more useful half. Integration with a system that neither party controls, a data-sharing agreement that has not been signed, an accreditation that has to land before anything can go to production: those belong on the page as explicit exclusions or as a task with a named owner. Work that lives in the gap between two statements of work is the most common way a good technical program goes sideways.
On federal bids the boundary is not only a management preference. It is arithmetic with legal consequences. On an SBIR Phase I the small business must perform at least two-thirds of the research and analytical effort, and at least one-half on Phase II, under the SBA SBIR/STTR Policy Directive. STTR splits differently: the small business performs at least 40 percent and the research institution at least 30 percent. On services contracts the limitation on subcontracting at FAR 52.219-14 caps what can flow to firms that are not similarly situated. A partner who arrives with the split already sketched, even roughly, is a partner who has read the same rules we have.
A person who can decide
The single strongest predictor of a partnership closing is whether the person on the call can either make the decision or name, in one sentence, the person who will.
On the prime side that is usually the capture manager who owns the bid and the number, not the business development screener collecting capability statements. On the university side it is the principal investigator plus the name of the sponsored programs officer who will process the subaward. On a commercial or state engagement it is the executive who owns the budget line and the operational problem at the same time. When those two sit in different people, the work still happens, it just takes an extra month, and we plan for the extra month.
Here is the test we actually use. If we send a one-page scope by Friday, who reads it and who signs it? A good partner answers that with two names and no hesitation. It reveals whether an organization is exploring or buying.
Data we can touch, on a date
The schedule risk in an AI or data program is almost never the modeling. It is the data. We have never seen a machine learning task fail for want of an algorithm, and we have seen many slip because the extract that was promised in week two arrived in week nine, in a format nobody described, with three years missing.
So the question we ask early is blunt: what is the system of record, who owns it, what format does it export, and what date can we have a sample? A partner who can answer all four runs on schedule. A partner who can answer none of them is not disqualified; the first work order simply becomes a data assessment rather than a build, and both sides should know that going in.
Sensitivity is a separate axis from availability, and it is one we handle routinely. Open and public data moves immediately. Internal business data moves under a mutual NDA. Controlled Unclassified Information carries the marking and handling regime of 32 CFR Part 2002, the NIST SP 800-171 control baseline, and on DoD work the flowdown at DFARS 252.204-7012 with its 72-hour cyber incident reporting clock. Export-controlled technical data is where many partnerships stall, and it is a place we clear cleanly: Precision Federal is certified under the Joint Certification Program with an approved DD Form 2345, CAGE 1AYQ0, which is the mechanism DoD Directive 5230.25 uses to release militarily critical technical data to a contractor.
If the data genuinely cannot leave your environment, say that in the first email. We build inside customer enclaves, we work against surrogate and synthetic corpora where the real corpus is unavailable, and we design evaluations that a partner can run on their own instrument and report back. What we cannot do is discover the restriction in week six.
Honesty about the constraint you would rather not mention
Every real program has one. The funding is not appropriated yet. There is an incumbent who is not going quietly. The customer likes the idea and has not agreed to pay for it. A previous vendor tried this twice and it did not work, and nobody wants to write that down. The deadline the executive announced is not the deadline engineering believes. An organizational conflict of interest under FAR 9.5 may bar one of the parties from the follow-on.
None of those ends a conversation with us. All of them change the design. A program with unappropriated funding gets scoped as a small, sharply bounded piece of work that produces something usable on its own, so that nothing is stranded if the money moves. A program with a failed predecessor starts with a short diagnostic, because the reason it failed is usually a data or definition problem that will bite the second attempt in the same place. A program with an incumbent gets a transition plan on page one.
Partners who lead with the awkward fact get better proposals and better prices from us, because we are pricing the real problem instead of the tidy version. That is estimating, not etiquette.
A funding path that exists on paper
We do not need the money to be in hand. We need the path to be nameable. A solicitation number with a close date. A task order under an existing IDIQ or BPA. A purchase order authority and a dollar ceiling. A state grant with a posted deadline. An internal capital line for the fiscal year with a named approver. Any one is enough.
Where a path does not exist yet, the honest ask changes shape. We are glad to help build the case, and the deliverable becomes an estimate, an architecture sketch, and a one-page justification a sponsor can carry into a budget meeting. That is a different piece of work than a build, priced differently and scheduled differently, and calling it what it is keeps both sides out of a slow-motion misunderstanding.
For research partners, the arithmetic of the federal path is worth stating plainly. Phase I awards commonly run between $150,000 and roughly $314,000 over six to nine months. Phase II typically runs to about $2 million across twenty-four months. The reason a modest Phase I is worth anyone's attention is what sits behind it: 15 U.S.C. 638(r)(4) lets an agency award a Phase III contract deriving from that work on a sole-source basis, with no dollar limit and no further competition. Partners who understand that structure treat Phase I as the front door rather than the prize.
The signals, and what each one costs when it is missing
| Signal | What the good version sounds like | Cost when it is missing |
|---|---|---|
| Decision authority | "I own the bid. Legal reviews, I sign." | Four to eight weeks of pleasant meetings |
| Scope boundary | "You own ingest and scoring. We own the UI." | Rework at integration, disputed change orders |
| Data access | "Sample extract, Parquet, from the warehouse, by the 15th." | The most common source of schedule slip |
| Sensitivity tier | "CUI under DFARS 7012. Our enclave, your accounts." | A month lost to an agreement nobody started |
| Funding path | "Task order under our IDIQ, ceiling already there." | Design work that cannot be executed |
| Stated constraint | "An incumbent holds the interface. Plan around it." | A late redesign at the worst possible moment |
What a good partner looks like, by type
The four things above hold across every relationship we have. What they look like in practice changes with who is on the other side.
- Prime capture managers. You know the customer, you own the win theme, and you need an AI/ML component that will survive technical evaluation. Send the solicitation and the piece you want covered. Note that FAR 15.305(a)(2)(ii) lets you cite a subcontractor's relevant past performance in your own proposal, and that a subcontracting plan is required once a contract exceeds $750,000 under FAR 19.702.
- Research institutions and faculty PIs. You have the science and need a small business to hold the award, run the commercialization plan, and handle the federal paperwork. Good partners bring a research plan, an honest statement of what is already published, and a tech transfer contact by name.
- System integrators. You have the contract and the customer relationship, and a data or model workstream that needs a bench with depth rather than headcount. Good partners give us the interface contract and the integration calendar early, then let engineering talk to engineering.
- Commercial and state buyers. You have a working problem and a budget. Good partners can describe the current process end to end, including the manual steps everyone is slightly embarrassed by, because those steps are where the value is.
- Investors and boards. You need an independent technical read on a company, a model, or a claim. Good partners tell us what decision the read supports and when it has to be in hand, and give us access to the actual system rather than a deck about it.
What the first ten business days look like
From first email to a scope you can act on
When we say no, and why a fast no is worth having
We build software, data, machine learning, and cloud systems. When the deliverable is a fabricated part, a bench chemistry result, or a staffed seat rather than an engineered outcome, the right answer is someone else, and we will say so within a day instead of letting a bid drift toward a deadline on a maybe.
We also decline when the request is a staffing requisition wearing an engineering label. Two engineers under someone else's ticket queue is a different business from a firm that owns a deliverable and stands behind it, and the second is the one we run. We decline when FAR 9.5 puts us on the wrong side of a conflict for a follow-on we both want. And we decline when the only way to hit the stated date is to promise capability we would then have to invent under a deadline. The proposal that wins on a promise and loses on delivery costs more than the bid we passed on.
Every no comes with the reason and, where we can, a direction. Capture managers who have gotten a fast no from us come back with the next bid, because a partner who declines cleanly is a partner whose yes means something.
What we owe you in return
A filter that only runs one direction is not a partnership, so here is the symmetric half. We answer fit questions within one business day, in writing, with the reason. If it is a yes, you get a one-page scope inside 48 hours with the deliverable, the named people who will do the work, the hours behind the number, and a fixed price where the scope supports one.
We name our people. Our bench is a standing group of named engineers, licensed professional engineers, and domain specialists across defense, health, energy, transportation, and public-sector data, and the names in a proposal are the names that show up on the calls. Our practice is led by a former professor in technology who ranks in the top 200 of more than 200,000 on Kaggle and holds seven cloud certifications, backed by twenty years of production federal system delivery across five consulting firms, three of them federal. Precision Federal is SAM.gov active, CAGE 1AYQ0, and JCP / DD-2345 certified.
We tell you when the answer is not us. And we hold the same standard on ourselves that this page asks of a partner: our scope has edges, our decision-maker answers email, and we say the awkward thing in the first exchange.
The four lines that start it
- What the work is. One or two sentences, plus the solicitation number if there is one.
- Who decides. The name of the person who reads the scope and the person who signs it.
- What data exists. The system, the format, and the date we could see a sample.
- The date. Proposal close, task-order start, or the board meeting the answer has to reach.
That is the whole intake. Four lines in a plain email beat a thirty-slide deck every time, because those are the lines we would have to extract from the deck anyway.
Frequently asked questions
A clear workshare boundary, a named decision-maker on the prime side, an early answer on data sensitivity, and disclosure of the constraints that shape the schedule. Splits also have to satisfy the applicable rules, including the limitation on subcontracting at FAR 52.219-14 and the SBIR performance percentages, so the arithmetic belongs in the first conversation.
One business day for a yes or no with the reason, and 48 hours after that for a one-page scope with named staff, hours, and a price. Fast answers matter most near a close date, which is when most partnering questions arrive.
Yes, routinely. We build inside customer environments, work against surrogate or synthetic corpora, and design evaluations a partner can run on their own instrument. The requirement is simply that the restriction is named up front rather than discovered mid-program.
A research plan, a PI who is engaged in the technical work, and a tech transfer or sponsored programs contact by name. The statutory split gives the small business at least 40 percent of the effort and the research institution at least 30 percent, and the small business holds the award and the commercialization obligations.
Precision Federal is certified under the Joint Certification Program with an approved DD Form 2345 under DoD Directive 5230.25, which is the release mechanism for militarily critical technical data. CUI is handled to 32 CFR Part 2002 marking rules against the NIST SP 800-171 control baseline, with the DFARS 252.204-7012 flowdown and its 72-hour incident reporting requirement where the contract carries it.