The fire that killed one phone maker and made another

Three issues ago I gave you the framework. Two issues ago I gave you the warning about what happens when you don't act on reclaimed time. This week I want to show you what the framework actually looks like when the stakes are real — using a story from outside procurement circles that most of you will half-remember, and a detail from inside it that almost nobody tells.

1. Walmart, Maersk and Vodafone are letting AI agents close supplier deals — and governance hasn't caught up. Reporting this week confirms what a lot of us suspected was coming: procurement is out ahead of every other function on agentic AI adoption, with negotiator agents now handling live, binding contract terms at volume no human team could match. My take: this is the RED category moving faster than most people's mental model of it. Fine for high-volume, low-complexity spend. Terrifying if anyone lets it near a strategic category without a human sign-off gate. Watch this one — the governance frameworks are the story, not the agents.

2. Only ~30% of organisations have mature governance for the agentic AI controls they're already deploying. New research this week puts real numbers on something I flagged in Issue #002 — most procurement AI pilots aren't stalling because the tools don't work. They're stalling because nobody built the guardrails before switching the tools on. Deloitte's parallel finding is worse: 85% of companies expect to customise their own AI agents, but only 21% have governance mature enough to do it safely. My take: if your organisation is excited about agents and hasn't had a single conversation about escalation thresholds, you're not early — you're exposed.

3. Chile's public procurement agency is using AI purely to make its own messy data usable. ChileCompra isn't chasing agentic negotiation — it's using AI to extract and classify information trapped in PDFs and unstructured formats so its existing systems actually work. My take: this is the unglamorous, unsexy 80% of AI's real value in procurement right now, and it's the bit nobody puts in a keynote. Structured data problems, solved quietly. More of this, less of the demo-stage theatre.

On 17 March 2000, lightning struck a power line outside Albuquerque, New Mexico.

The resulting power surge caused a fire inside a Philips semiconductor plant — a fire that lasted about ten minutes and was put out by the plant's own sprinklers. Philips' initial internal assessment: a week of lost production, easily absorbed.

That single plant supplied radio-frequency chips to two of the biggest phone makers on earth: Nokia and Ericsson. Both were told the same thing, at roughly the same time, through the same channel. A minor fire. A week's delay. Nothing to escalate.

Here's where the two companies split.

Nokia's component purchasing team didn't take the initial reassurance at face value. One of their own supply chain managers pushed for more detail, kept asking, and eventually got Philips to admit the real damage was far worse than a week — millions of chips, months of output, gone. Nokia responded immediately: redesigned some chips to be sourced elsewhere, leaned harder on alternative suppliers, and had senior executives personally engage Philips to secure a larger share of whatever capacity remained.

Ericsson, by most accounts, took the initial reassurance at face value and didn't escalate with the same urgency. By the time the scale of the problem was fully understood inside Ericsson, the chip shortage had already started biting into a product line the company could not afford to lose.

The numbers that followed tell the human story.

Ericsson later estimated the disruption cost its mobile phone division somewhere in the range of $400 million in lost sales. Within two years, Ericsson had exited handset manufacturing altogether, folding its phone business into a joint venture with Sony. Nokia went on to become the dominant handset maker in the world for most of the following decade.

Same fire. Same supplier. Same initial information. Completely different outcome.

Why this belongs in a newsletter about AI and procurement

Because it's tempting to read this story and think: if only they'd had better tools. They didn't need better tools. Both companies had the same information source. What separated them was entirely a GREEN and AMBER task from the framework I gave you in Issue #003 — supplier risk interpretation, and the judgment to keep pushing on a signal that didn't sit right.

AI does not change that judgment call. What AI changes is how cheaply and quickly you can gather the raw signal in the first place — the layer of intelligence-gathering that used to require an analyst, a network of contacts, or, in Nokia's case, a manager persistent enough to keep phoning a supplier back. That gathering work is now something any procurement professional can do in fifteen minutes with the right prompt. The interrogation of what it means is still entirely yours.

Four takeaways:

This week's tool: Perplexity AI — for the pre-meeting supplier brief

I ran the exact intelligence-gathering exercise this newsletter is built around, using a real current supplier relationship and the pre-meeting brief prompt from the AI Playbook appendix.

What it got right: financial health signals, recent leadership changes, and market sentiment came back in about twelve minutes, cited, and organised in a way that would have taken me the better part of an hour manually pulling from LinkedIn, trade press, and company filings separately.

What it didn't get right: it has no memory of my relationship history with that supplier, no sense of which of their account managers has a track record of underselling problems, and no way of knowing that a "minor" issue from this particular supplier has, in my direct experience, tended to be understated by roughly the same margin every time. That context lives only with the practitioner who has sat across the table from them before.

Honest verdict: AMBER, and correctly so. It will not make the Nokia call for you. It will get you to the point where you can make it fifteen minutes faster than you could last year — which, as the story above shows, may be the only fifteen minutes that matters.

The framework only earns its keep when reclaimed time gets spent asking the second question, not just processing the first answer faster. If this issue helped you see where that applies in your own supplier base, grab the free ebook this newsletter is built on — it has the full pre-meeting intelligence framework and prompt templates in Chapter 7, ready to run this week.

→ Get The Procurement Professional's AI Playbook (free) at thesmartersupplychain.com

The Smarter Supply Chain — practical AI for procurement and supply chain professionals, written by a practitioner with 28 years in the field. No theory. No hype.