#45 - The Procurement capacity gap, AI agents stuck in pilots, and Section 338 tariffs

Today we look at why Procurement is asked to deliver more and more with the same headcount, why so many agentic AI projects stall before reaching production, and what lies behind the new Section 338 tariffs between the US and Canada.

Together we will see where the capacity gap comes from, what it takes for a pilot to survive a real process, and why the map of trade keeps shifting under our feet.

More goals, same headcount: the Procurement capacity gap

Procurement teams are being given ever more ambitious goals, but the people to work on them stay the same. It is the clearest signal to emerge from the data of the last two weeks, and it raises the question every leader now carries around: where do I find the capacity I am missing?

A new CPO benchmark study, published on September 9 by ProcureAbility, quantifies a pressure that almost everyone already feels:

• 61% report higher savings expectations than in 2025
• 71% describe their team's capacity as moderately or severely strained
• 42% live with both: more results to deliver, and less room and capacity to deliver them

And the answer says a lot about where the function is heading. Only 19% plan to expand headcount. The rest look elsewhere: 46% invest in training the people already on the team, 31% rely on automation. For a manufacturing company, where a team of six or eight people has to oversee thousands of suppliers, the idea of hiring your way out stopped holding up a long time ago.

There is, however, a second brake, and it concerns automation itself: data quality. Half of respondents name it as the first obstacle to scaling AI, and 51% put it first among analytical gaps. A team buried in manual work rarely has clean, structured data, and without that data the tools that would free up time deliver less than expected. Capacity and data go hand in hand.

Why automation stalls before reaching production

If the answer is automation, an honest question remains: why does it so often stall before reaching everyday use? Money is not the problem: between April and September about $435 million flowed into enterprise agentic AI companies, concentrated on security and governance, the layer that decides whether an agent can run safely inside a real process.

Capital is chasing a by now well-known breaking point. International Data Corporation (IDC) and Lenovo estimate that 88% of companies never bring their agent initiatives into production, and Gartner predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027. The pattern repeats the same way: the pilot shines in the demo, then jams when the questions become real ones. Who approved? Which rules set the boundaries? Can you show that trail to Finance and to the auditors?

For a CPO, this moves the buying decision from the stage to reality. What matters about an AI agent is what it does on a Tuesday afternoon, when it handles a stalled purchase request or a negotiation with a supplier. A polished demo says little about any of this. It is governance that turns a promising pilot into capacity you can truly count on.

50% tariffs and record containers: no respite

None of this happens in calm waters. On August 22, US tariffs of 50% under Section 338 took effect on about $27.6 billion of Canadian goods. Section 338 is a 1930 provision never applied before, which allows duties of up to 50% on countries judged hostile to American trade. Canada responded with mirror counter-tariffs on steel, dairy products, wood, textiles and clothing, in force since September 8. For any manufacturing company with cross-border supplies, it is a concrete matter of re-sourcing and cost exposure already this quarter.

Meanwhile demand is racing: in July, global container volumes hit the all-time monthly record of 17.3 million TEU (the unit of measure for container traffic, equal to one standard 20-foot box), up 5.1% since the start of the year. Goods are definitely moving, and meanwhile the map of trade keeps reconfiguring.

Ahead are the teams that free up capacity without waiting for a bigger budget and that build governance into automation from day one. The pressure is real and the data confirms it, the tools exist: the work now is choosing the ones that stand up to a real process.

Compri helps manufacturing companies grow Procurement capacity without increasing headcount. Learn more 👉 compri.ai