Part 1 – AI in government: the procurement problem
There is no shortage of enthusiasm for artificial intelligence in the Australian public sector. Across every portfolio, agencies are exploring use cases, standing up pilots, and fielding briefings from vendors who promise transformational outcomes. The potential is exciting. The energy is real. So is the problem.
The problem is not technology. It is not policy, which is catching up. The problem is procurement and the time it takes.
By the time a standing offer panel is established, the technology listed on it is already a generation behind.
Australia's government procurement frameworks were designed for a world in which technology was stable, vendors were distinct, and evaluating capability was a matter of comparing specifications, implementation timeframes and delivery risk. AI invalidates all three assumptions simultaneously.
The pace of development in AI is unlike anything the enterprise technology market has previously experienced. A foundation model that represented the frontier six months ago may be superseded today and not incrementally, but potentially fundamentally. Procurement cycles that take 12 – 18 months to contract are, by definition, evaluating yesterday's technology. Agencies that lock into arrangements may find themselves contractually tethered to a solution or vendor whose AI capability has been leapfrogged before the first purchase order is raised.
The second problem is vendor distinction. Traditional procurement assumes you can identify and evaluate discrete suppliers. AI has made that increasingly difficult. The large language model underlying an agency's chosen product may be licensed from a third party. The infrastructure may sit on a hyperscaler. The fine-tuning may have been done by a boutique specialist. The integration may be the only genuinely differentiating layer. Buying 'an AI solution' from a single vendor is often a false construct but procurement frameworks still treat it as real.
The third assumption surrounds capability evaluation. This breaks down entirely when the technology changes faster than evaluation cycles. By the time an assessment panel has assessed a product, scored it, challenged the scores, gone through probity review, and awarded the contract, the product has been updated three times and the market has moved on.
A different model is needed
The agencies making the most progress on AI are not the ones with the most sophisticated vendor engagement. They are the ones who have invested in understanding what they are trying to achieve, separated that question from the question of which technology delivers it, and have built strategic relationships that can support them in navigating a market that will keep changing.
That means outcome-based procurement beginning with defining what success looks like. Not assessing products. It means procurement structures that allow flexibility across vendors and generations of technology. It also means engaging vendors who have no stake in which AI platform wins, only in whether the agency's objectives are met.
NCS works with government agencies on exactly this basis. We are not the vendor of an AI product. We are the partner who helps agencies define the problem, evaluate the options and the roadmap, build the capability to adapt as those options change tomorrow, and integrate whatever solution is chosen into the broader landscape the agency operates in.
The question is not 'which AI solution should we buy?' It's 'what are we trying to achieve, and how do we stay capable of achieving it as the technology evolves?'
This is a harder question. It requires more than a briefing from a vendor. But it is the right question. Agencies that are asking it are the ones that will have something to show for their AI investment in three years' time.
