Walmart's AI Is Already Negotiating. My Local Council Is Still Deciding Who to Trust.

I've been trying to understand something for a while now: how differently is AI actually landing in procurement depending on who's buying? Walmart is already letting an AI negotiate directly with 2,000 suppliers at once. That's one end of it. The other end is a lot closer to home and a lot less flashy — a document from the Municipal Association of Victoria, the body that represents local councils here, laying out how those councils should even go about vetting an AI vendor in the first place. Not negotiating with AI. Just deciding who's allowed to sell it to them.

That gap is the actual thing I wanted to write about. Not "here's what MAV's document says" for its own sake — I doubt many people reading this have any reason to know what MAV is, and they shouldn't need to. What I actually wanted to understand is where government procurement sits on this curve compared to where a company like Walmart already is, and what that gap tells you about how AI adoption really moves through an economy — fast at the top of the private sector, and much more cautiously everywhere decisions have to be defensible to the public.

What the council document actually does

MAV published a scoring rubric in October 2025 that Victorian councils use to assess AI vendors before they're allowed onto a shared procurement register. Six categories, weighted, each scored 0–4. Five of them are genuinely lenient — a vendor can score well with nothing more than a credible, dated roadmap instead of a finished capability. One isn't: Regulatory & Legislative Compliance, weighted the highest of the six, with a hard rule that anyone scoring under 50% shouldn't be on the register at all. No roadmap gets you out of that one.

| Category | Weight | Roadmap accepted instead of current capability? | |---|---|---| | Strategic Relevance & Capability | 20% | Yes | | Governance, Ethics & Human-Centred AI | 20% | Yes | | Regulatory & Legislative Compliance | 25% | No | | Data Usage, Security & Technical Standards | 15% | Yes, if evidence is "in place or in progress" | | System Adaptability & Improvement | 10% | Yes | | Sustainability & Implementation Support | 10% | Yes |

That one non-negotiable gate against five negotiable ceilings tells you exactly where a council thinks the real risk sits — not in whether the AI is any good, but in whether it's legal. Which is about as far from "let it negotiate 2,000 supplier deals overnight" as procurement gets.

Why I actually read it the way I did

I've sat on both sides of a rubric like this — as a buyer setting weighted criteria of my own, and as a bidder. My business applied for a government logistics tender last year and didn't get shortlisted. The documentation load alone was heavy enough to strain a small team, and once the field clearly included established incumbents, our real odds had narrowed before evaluation even started.

Put that next to MAV's rubric and something genuinely useful falls out of it: a hard binary gate, like the compliance floor here, can actually help a smaller vendor — either you clear it or you don't, and there's no room for an evaluator's comfort with a familiar name to creep in. It's the open-ended documentation slog that quietly favours incumbents, because a big established supplier has already amortised that cost across dozens of past bids. If I were advising a small AI vendor on whether a register like this is worth the effort, I'd tell them the strict-looking parts matter less than how much of the process is a clean yes/no versus an endless paperwork exercise.

The one clause I actually think is genuinely sharp

Buried in the guidelines is a set of AI-specific conflict-of-interest disclosures that go further than most of what I've seen published on AI procurement anywhere in Australia. Vendors have to disclose if their system was trained on data from parties who have business before the council. They have to disclose if their own compensation is tied to specific outcomes the AI could be nudged toward. And they have to name, in writing, the exact point where protecting their own algorithm limits how transparent they can be about how it reaches a decision.

That last one is the interesting part. Most AI procurement guidance treats transparency as something to be scored. This treats the limits on transparency as something the vendor has to actively confess to, up front.

What this actually tells me about the gap

Reading this next to what Walmart's already doing, the honest picture is: private sector AI adoption in procurement is running way ahead on the "make it work at scale" problem, and government procurement is still mostly working through the "how do we even trust it enough to let it near a decision" problem. That's not councils being slow for no reason — a public tender needs someone who can be held accountable for the outcome, and no council is handing that off to a system nobody's finished vetting. But it does mean the two conversations about "AI in procurement" happening right now are barely the same conversation. One is about negotiation efficiency. The other is still about admission criteria.

If you're a small AI vendor trying to sell into government, that's the practical takeaway: figure out early which parts of an evaluation are a hard gate and which reward a well-written roadmap, because that's where your limited time is actually worth spending. And if you're just watching this space the way I am — it's worth knowing you're watching two different clocks, not one.