If you run a Tier 2 or Tier 3 subcontracting business in oil & gas, defence, energy or EPC work, you have probably had the same conversation land on your desk in the last few months: a prime contractor has flowed down a National AI Plan (NAIP) requirement, and somewhere in the pre-qualification pack there is now a section asking how you govern the AI your business uses. It was not there last year. It is there now, and it is scored.
This is a genuinely awkward position. You did not build an AI department. You do not have a compliance team sitting idle. And yet the gate is real, the deadline is usually "next month," and failing it can quietly cost you a spot on a panel you have been on for years.
So before you spend a dollar with anyone, it is worth being clear about what a fit-for-purpose compliance system actually needs to do for a business like yours — and, just as importantly, what it does not need to be.
The short version — five tests for any NAIP compliance system: it must produce scored tender evidence, be sized to what a Tier 2/3 sub can carry, map to the Australian regulatory streams, be sovereign and onshore where it matters, and leave you owning something your own team can maintain.
1. It Must Answer the Tender, Not Just Describe Good Practice
The single most common mistake is treating this as a "get certified" or "write a policy" exercise. Those things have their place, but the thing being scored is a pre-qualification response. An assessor sitting on the prime's governance team is reading your answers against the NAIP pillars and marking them.
A generic framework, a management-system certificate, or a safety report does not answer that question — it sits adjacent to it. The first thing to look for, then, is whether what you are buying produces actual tender evidence, mapped to the pillars an assessor scores, in the language a bid evaluator expects to read.
2. It Must Be Sized to What a Tier 2/3 Sub Can Carry
Heavy industry has a specific trap here: enterprise-grade AI governance frameworks are built for organisations with compliance staff, ongoing budget and a tolerance for controls that a prime can absorb but a sub cannot.
If your compliance system commits you to obligations you will never realistically maintain — or that you have to price back into the bid — you have made yourself less competitive, not more. What you want is someone who knows what a Tier 2/3 business can genuinely carry, and who sizes the controls to that reality rather than defaulting to enterprise theatre.
3. It Must Cover the Australian Regulatory Streams
A lot of the AI-compliance tooling on the market is built horizontally and globally — designed around ISO 42001, SOC 2 and ISO 27001, often from the US or EU. Those are legitimate standards, but they are not the same thing as the streams a NAIP tender actually tests:
- the Privacy Act's automated decision-making (ADM) provisions;
- the Digital Transformation Agency's requirements;
- SOCI obligations for critical-infrastructure work; and
- the prime-to-sub flowdown logic that governs how a requirement cascades down the contract chain.
If a solution cannot map to those specifically, it is answering a different exam. (For a plain-English walk through which of these apply to which kind of business, see Sovereign AI in Heavy Industry: What Actually Applies to You.)
4. It Must Be Sovereign and Onshore Where It Matters
On government and critical-infrastructure work, "where does your data live?" is a recurring, and sometimes disqualifying, question. A solution hosted offshore — however good — can create a sovereignty problem you then have to explain away.
For heavy-industry work touching defence, energy or critical infrastructure, onshore hosting (for example, AWS Sydney with data that never leaves the country) is not a nice-to-have; it is often the difference between a clean answer and an awkward one.
5. It Must Leave You With Something You Own and Can Maintain
Finally, look hard at what you are left holding after the engagement ends. There are broadly two failure modes.
The subscription-software model leaves you paying an ongoing licence for a platform someone internally still has to drive — and subs rarely have the staff to drive it. The bespoke-consulting model leaves you with a report that ages quickly and a bill that recurs every time you need it refreshed.
The better outcome for a sub is a compact, version-controlled evidence pack that your own team can maintain without a subscription meter running or a consultant on retainer.
The Market Splits Three Ways — and None Alone Fits Your Job
Once you start shopping, you will find the field falls into three camps. It is worth understanding each on its own terms, because each is genuinely good at something — just not, on its own, at the specific thing you need.
AI consultancies: real engineering depth, wrong shape of engagement
Firms of the Mantel Group / Quanton / Kinetic / Protiviti type have real machine-learning and engineering depth. If you wanted a custom AI tool built, they would out-resource almost anyone. But their track record skews to banking, retail, government and insurance; heavy industry is typically one line on the website.
They build models and capability on a scoped, time-and-materials basis — which turns you into a stakeholder in a build, typically priced from tens into hundreds of thousands, over months. That is a transformation program. Passing a pre-qual gate by Friday is a different job, and it is not the one they are structured to sell.
Domain specialists: one half knows the site, the other half knows the rules
Heavy-industry risk and AI-governance specialists are really two different animals that both look like rivals.
The safety and operational-risk firms know your site cold — WHS, EHS, hazard and incident, mining and O&G — and are increasingly bolting "AI" onto safety monitoring. But safety risk is not AI Plan compliance; a site-risk report will not answer a Privacy Act ADM or SOCI question.
On the other side, the pure-play governance and ISO 42001 specialists genuinely know the Act and the standards — but they have often never sat at a bid table or flowed a requirement down to a sub, and their deliverable tends to be a generic framework priced at a level a sub cannot fund. One half knows the site; the other half knows the rules; neither, alone, holds both.
AI SaaS and GRC platforms: mature engines that still need a driver
The compliance-automation platforms — Vanta, Drata, SureCloud, Scrut and the Australian ISO-42001 players — are mature, impressive engines with continuous monitoring and thousands of certifications behind them.
But a platform does not sit in the room, gather your data, make the risk call or draft the bid answer — someone does, and on a subscription model that someone is you. They are also built horizontal and global, with no native concept of NAIP streams, prime-sub chains or SOCI, and they leave you paying a licence indefinitely.
The through-line: builders cannot do tenders, domain experts cannot do AI, and software cannot do your industry. For a heavy-industry sub facing a scored NAIP gate, you need all three at once — and that combination is rare.
Common Questions
What is a National AI Plan pre-qualification requirement?
It is a scored section in a prime's pre-qualification pack asking how your business governs the AI it uses. The prime's governance team marks your answers against the NAIP pillars, so it is assessed as tender evidence — not as a general statement of good practice.
Does ISO 42001 or SOC 2 certification satisfy a NAIP tender question?
Not on its own. Those standards are built horizontally and globally. A NAIP tender tests specific Australian streams — Privacy Act ADM, DTA requirements, SOCI, and prime-to-sub flowdown. A certificate sits adjacent to those questions rather than answering them.
Why does data residency matter so much for a sub?
On government and critical-infrastructure work, data residency is a recurring and sometimes disqualifying question. Offshore hosting creates a sovereignty problem you then have to explain away; onshore hosting is often the difference between a clean answer and an awkward one.
What should I own after the engagement ends?
A compact, version-controlled evidence pack your own team can maintain — with no subscription meter running and no consultancy on retainer.
Coming Up in Part Two
In Part Two, I walk through where a solution that genuinely holds all three pieces fits, exactly where you would still need assistance in building your compliance system, and how a good engagement covers those gaps without turning into an open-ended bill.
Read Part Two: The Right Fit for a Heavy-Industry Sub →
Facing a scored AI governance gate in your next pre-qual?
James works with prime contractors and subcontractors across heavy industry to build defensible, audit-ready governance documentation. Fixed scope, clear outcomes.
Talk to James about your AI governance position