06-reference

mostlymetrics procurement ai spend market maps

2026-11-12·reference·source: Mostly Metrics·by CJ Gustafson
procurementai-spendfinopsllm-routersagentic-financecfo

2026 Procurement Software and AI Spend Market Maps ("Always on Procurement")

Despite the title, this is the procurement and AI-spend chapter teaser for CJ Gustafson's upcoming Agentic Finance report (see [[2026-11-10-mostlymetrics-core-accounting-software-market-map]]), built around the claim that AI costs behave like cloud costs but land in every department.

Why this is in the vault

It is a CFO-side framing of AI token spend as a new always-on procurement and observability problem, with named vendor positions (Proxon, Tropic, Abacum) that are useful vocabulary for the AI-cost-governance conversation.

Mapping against Ray Data Co

The most concrete connection is RDCO's own cost discipline: the API-cost and spending-authority rules ([[project_rdco_spending_authority]]) are a one-person version of the metering-and-attribution layer CJ says finance must own. His point that spend attribution (vendor, model, team, workflow, owner against outcomes) is the missing layer is also an angle for [[project_credibility_for_phdata_sales]], since it is language CFO buyers use. Cross-read with [[2026-10-20-secret-cfo-ai-spend-payback-mailbag]] on AI spend payback. Medium: the thesis is sound but the Proxon and Tropic quotes come from report underwriters, so treat them as vendor views. No RDCO decision changes.

The core argument

Cloud costs sat with the CTO and mostly in COGS. AI inference is different in three ways: the amount is on a path to exceed cloud, it is multi-domain (sales, marketing and finance power users burn tokens), and it is unpredictable (plan sprawl, dozens of AI note takers). So the CTO cannot carry it alone.

Supporting points: Stripe's reported acquisition of OpenRouter and Ramp's router launch signal routers ("Kayak for models") going mainstream, but you only buy one if you can prove savings, which requires observability first. Proxon (Peter Pezaris) pitches spend attribution against results. An anecdote cites a $400K token bill in three days from one experiment. Tropic's Justin Etkin calls it always-on procurement: model changes upend forecasting, and with AI in cost of revenue, a 30 versus 50-60 percent gross margin matters much more. Abacum's Julio Martinez says AI behaves like headcount and should be planned by team. CJ's conclusion: finance must own metering AI spend and fairly passing cost through to customers.

Curation section

⚠️ Sponsorship

Masthead sponsor is Brex ("proudly powered by", affiliate-tagged partnerId=metrics link), a confirmed recurring pool member; Tabs did not appear this issue. The ten report underwriters (several quoted as experts) are paying partners, so their quotes are vendor perspectives. Report and podcast promos are CJ's own self-promo.

Related