What I Learned Trying to Win Canadian Government Contracts

I started JININSA TECHNOLOGY as a deliverable-based software shop: AI and agentic systems, cloud architecture, data pipelines. The kind of work I do every day, packaged as milestone-driven statements of work rather than a full-time seat. Government looked like an obvious buyer. Canada spends enormous amounts on exactly this, the opportunities are public, and the work fits.

So I did the naive thing first. I pulled the open tender feeds, filtered for software and AI, and started reading. What follows is everything I wish I had understood before I opened that first CSV.

Government buying is not commercial selling

The single most important thing I learned: a public buyer is not optimizing for the best solution. They are optimizing for a defensible decision. They spend other people's money under trade law and audit, so the private question in every evaluator's head is "will this choice get overturned, audited, or blamed on me?" Almost everything strange about the process flows from that one fact.

It means you are selling certainty and defensibility as much as capability. A slightly more expensive bid from a firm that clearly will not fail beats a cheap bid from an unknown. It means once a solicitation is live, the buyer goes silent and will only answer written questions in public, because talking to you privately would taint a process that losing bidders can formally challenge. And it means the decision is made by a rules-bound scoring exercise you have to play mechanically, not by the gut-feel relationship selling that works in the private sector.

The shape of the cycle

Every one of these follows the same arc, and knowing where you are on it changes what you should do:

Need identification. Someone inside has a problem and budget. Invisible to you unless you have a relationship or you read their strategy documents.
Market research. The buyer asks the market to educate them, usually through a Request for Information (RFI) or an industry day. Nothing is awarded.
Requirement definition. They write the actual requirement, often shaped by what RFI respondents told them, and pick the procurement vehicle.
Solicitation. The RFP or RFQ goes live. The clock starts and the buyer goes quiet.
Questions and addenda. Written Q&A, answered publicly. The answers usually reveal what the buyer actually cares about.
Submission. By the deadline, to the second. Late is dead, unopened.
Evaluation. Mandatory pass/fail screening first, then rated technical scoring, then price, then a combined formula.
Award, standstill, debrief. You can request a debrief even when you lose, and it tells you exactly where you scored down.
Delivery. Milestone deliverables or task authorizations against a ceiling.
Options and re-compete. Most contracts have option years, and the incumbent has an enormous advantage on renewal.

The thing that took me a while to internalize: the sale is mostly won or lost before the solicitation ever drops, at stages 2 and 3, but the decision is executed by a mechanical scoring exercise at stage 7. Those are two different games, and small firms tend to be good at neither. They show up cold at the RFP with no pre-positioning, then also fumble the compliance mechanics.

The staff-augmentation trap

My first exciting find was a federal notice for "AI-enabled software development." Perfect fit, right up until I read the fine print. It was a TBIPS requirement (Task-Based Informatics Professional Services), open only to firms already holding that supply arrangement, and it specified 915 days of level of effort for a single developer.

That is not a deliverable. That is a full-time seat billed by the day. The whole TBIPS family is essentially government body-shopping: the buyer is purchasing a person's time, not an outcome. If you have a day job and want milestone-based work, it is exactly the wrong shape. I learned to treat the words "resource categories" and "as and when requested" as a signal to move on.

The lesson generalizes. Read for the commercial model, not the topic. A notice can be dead-center on your skills and still be structurally incompatible with how you want to work.

The reference gate is what actually kills you

Here is the wall every new firm hits. Evaluation runs mandatory criteria first, as pass/fail. Fail one and nobody ever reads your brilliant technical section. And the mandatories are almost always some version of:

must hold a valid supply arrangement, or
must submit two or three reference projects of similar scope with verifiable client contacts, or
must demonstrate X years of corporate experience.

A brand-new company has none of these, by definition. It does not matter how good the work is. You are screened out before the evaluation of ideas even begins. This is the real reason direct bids felt hopeless: not that I could not do the work, but that the machine had no box for a firm without a track record.

Two things break the deadlock. First, build references from small commercial work - even modest fixed-scope projects, instrumented for a measurable outcome, converted into a reference with a willing client contact. Second, and more powerful, is the path I will come back to below.

Pre-qualification pools are the small firm's friend

Not every purchase runs the full open cycle. Some vehicles are two-step: you qualify once into a pool, then compete only against other pool members for individual jobs, often on compressed timelines. Federally these are the supply arrangements (TBIPS, SBIPS) and the AI Source List, which pre-qualifies suppliers into bands for AI work up to set dollar ceilings. Provincially, Alberta runs standing pre-qualification requests for things like application services and IT advisory.

The trade is that you front-load the pain: a heavy qualification submission up front, in exchange for a stream of lighter, smaller-field competitions afterward. For a firm that cannot win a big open RFP cold, this is usually the better structural bet. Being listed is not a payday, though. A pool guarantees nothing until a call-up or second-stage competition comes your way.

The RFI is the cheapest move on the board

The best-fit thing I found was not a contract at all. It was an RFI from a city transit agency exploring AI-based, camera-driven condition assessment of bus-stop assets. Computer vision, an edge pipeline, a reporting layer, privacy-by-design because they explicitly did not want to store images of people. A system shaped exactly like something one engineer can scope and deliver in milestones.

An RFI has no award. Your "bid" is a written document answering their question list. You cannot lose it. But the payoff is real: RFI respondents shape the requirements of the eventual RFP, and the vendor whose framing ends up in the RFP has a structural edge when it comes time to score it. You also land on the bidder list for the follow-on. A weekend of writing buys you influence over the requirements and a seat at the next stage. That is the best return on effort in the entire process.

The subcontracting breakthrough

Even with all of that, the direct-bid odds stayed low, because of the reference gate. So I went looking at the problem from the other side, and the data reframed everything.

Ottawa has spent more than $800 million on AI in three years. The firms winning that work are large staffing and advisory generalists. They win the vehicles and the big contracts, but they frequently lack deep, hands-on AI and agentic engineering talent in-house. That gap is precisely what a small specialist has.

And here is the mechanism that removes the reference gate entirely: a subcontractor does not need to hold the supply arrangement or the corporate track record. The prime does. The standard instrument is a teaming agreement - a written agreement on scope and price, no new legal entity, no shared equity. The prime carries the qualifications and names you in their bid as the specialist who de-risks their technical score.

That flips the whole positioning. My product is no longer a winning prime bid, which I cannot produce yet. It is specialist capability that makes someone else's already-qualified bid stronger. I go from "unproven firm fighting the reference gate" to "the reason a prime's proposal scores higher on technical merit." Much easier sale.

Two honest caveats I had to respect. Teaming agreements are, legally, agreements to agree - work-share is not guaranteed if the prime wins. And you have to be named in the actual bid, not swapped in after award: a 2025 Procurement Ombudsman review found that in a third of sampled contracts, none of the originally proposed personnel actually did the work, so the rules on resource substitution are tightening fast. The clean path is to be in the proposal from the start, ideally with a fixed-price deliverable carved out so it stays milestone-based and does not turn into a per-diem seat.

There is also a lower-threshold path I had missed entirely: ProServices, the mandatory federal vehicle for professional services below the trade-agreement threshold (roughly sub-$100K). Lighter qualification, small enough dollar values that a firm like mine could realistically prime them directly. Small value is the feature, not the bug. It is the on-ramp.

What I would tell past me

Read tenders for the commercial model, not the topic. Staff-aug and deliverable work look identical from the title.
The mandatory criteria, not the technical section, decide whether you are even in the game. Solve the reference gate first.
Pre-qualification pools and RFIs are cheap, high-leverage, and boring. Do them anyway. They are how you get inside the fence.
Stop trying to win tenders as a prime before you have a track record. Attach to someone who already has one.
Your first small win is worth more than its dollar value, because it makes you an incumbent and a reference for everything after it.

None of this is the fast, heroic version of starting a company. But it is the version that matches how the buyer actually buys, and that is the only version that works.