First issue. Here’s what Indirect, Augmented is, and why it’s landing in your inbox now.
Indirect procurement is being transformed. Not in the future: now. On foundations you’ve already built (P2P, master data, contract governance), that AI is finally accelerating by absorbing the transactional load. The ecosystem moves fast: every month, a new model, a new tool, one more acquisition. This newsletter is the map to navigate that change.
The hero of this map is the augmented buyer: tooled when it helps, method when the tool isn’t enough, human judgment when method hits its limits. The contract fits in one line: one issue a month, 8 to 10 minutes to read, a month of news distilled, by a buyer obsessed with procurement and convinced by AI. No evangelism, no SaaS pitch. I explore with you where AI makes sense, where it doesn’t, and how to try to make the most of this fast-moving technology.
Well augmented, the indirect buyer becomes the strategic partner to the CFO everyone has been waiting three decades for. Badly augmented, he becomes a robot that approves purchase orders faster.
My read on procurement AI in three states: assisted, augmented, autonomous →
On the menu this month: why “AI costs too much” is the wrong question, a framework to decide (Build / Buy / Defer), three acquisitions reshaping the market, a workflow to try Monday, and two players to watch.
“AI costs too much”: the line that gives away you never priced the value.
One question runs through this whole issue: what does AI really cost, and compared to what. Here’s my short version. Since Microsoft cut its Claude Code licenses and Uber burned through its 2026 AI budget in four months, the debate has shifted to: “AI costs too much.” That’s the wrong question.
The cost of AI is one of the few things we can actually price today. The price per million tokens runs from $0.10 to $25 depending on the model (public vendor pricing, 2026), and it drops every quarter. That cost is known, bounded, and falling. What almost no one prices is the other side of the equation: the value. Which workflow, how much time freed, reinvested in what.
The approach of these giants was bound to fail. Instead of finding the right use cases and governing their agents (costs included), they handed unlimited credit cards to their teams, in the form of tokens. Uber even ranked its teams by usage volume. Then everyone is surprised the bill explodes.
Behind these companies’ “AI problem” sits a governance problem, and even a procurement one:
- Oversized models → challenge the need.
- Resources poorly used → assess TCO and ROI.
- Budgets exploding → governance and cost-monitoring.
- Contracts left unmanaged → negotiation.
You already make this trade-off every day, without naming it. An €80 purchase request routed through 4 approvers costs more to control than to execute: under €500, processing alone commonly runs €30 to €80 (Ardent Partners / APQC, P2P benchmarks). An €800k order through the same workflow is the opposite. The right level of process calibrates to risk and value, not to gross amount. Exactly the reasoning to apply to AI.
Value first, cost second. An organization that starts from cost buys the cheapest license and buries it in a PoC that gathers dust.
Identify the workflow that frees the most judgment time, price that time, then build with cost in mind. In that order. And that’s where indirect procurement is the right pilot. Not IT, not an AI committee: procurement already does this work. Run a should-cost on the AI itself: that’s exactly the job of an AI procurement audit. Manage your agents like you would manage a new hire: a scope, rights, a review. Governance enables augmentation rather than holding it back.
Well augmented, the indirect buyer sets the frame that makes value measurable. Badly augmented, he signs three SaaS and calls it a strategy.
The cost of AI was never the issue. The issue is the value you’re augmenting, and whether you priced it before opening the wallet.
Build / Buy / Defer.
Three acquisitions mark the month (detailed below, in the signals): each one a bet on the “Buy” branch of a framework that belongs on every indirect buyer’s desk in 2026. McKinsey sorts every agentic-AI decision into four branches. Read through a procurement lens:
Code it in-house. Reserved for genuinely differentiating use cases. Costs 3 to 5× more over 3 years once run and model-retrofit are counted.
Co-develop with evolution SLAs, oversight on fine-tuning, a reversibility clause. Without those clauses, it’s Buy in disguise.
Buy off the shelf, demanding an open-source benchmark as your BATNA. The default route for non-differentiating use cases.
Wait, as an active decision. Legitimate when the impact of waiting stays below the 12-month TCO.
You’ll recognize these branches in this month’s news. The three acquisitions are the Buy strategy seen from the vendor side: they buy the capability rather than build it. Lio, the a16z-funded vertical (below), is the Build strategy. And the same decision faces you, at your scale: for each AI use case, build it, buy it, or wait? The branch everyone forgets is Defer. BCG surveyed 32 retail CPOs: 72% are exploring AI, none could name a concrete use case. (BCG, Indirect Spend in Retail, January 2025.) On a technology whose token cost drops about 80% in 18 months (public vendor pricing, 2025-2026), waiting 6 to 12 months is often the highest-ROI scenario. 80% of indirect procurement use cases (spend categorization, invoice processing, RFQ generation) aren’t differentiating: the default answer is Buy or Defer, never Build.
Before your next AI committee, demand the “do nothing for 12 months” scenario, with the opportunity cost quantified. In 30 to 50% of cases, that’s the one that wins.
The month’s pieces on augmentation.
This month, one obsession on the blog: the cost of AI. Three angles to put it in its place.
Control your AI costs before they control you.
The inference bill is managed like any indirect spend line: before it drifts, not after. The line item you discover too late.
AI: the real problem isn’t the price.
Too expensive, compared to what? Why Spotify pays its AI bills without flinching while Microsoft and Uber panic. Same spend, two readings of value.
Why set up an HR process for AI agents.
What if we borrowed from HR to manage agents: a scope, rights, a review, an offboarding. Govern an agent like you onboard a new hire.
The month of three acquisitions.
One month, three acquirers with opposite profiles, one target: own the autonomous procurement stack. What each deal means, and what the three say together.
Vertice acquires Vendr (June 1). The merger claims the largest procurement-intelligence dataset in the world: $75B+ in indirect spend, 32,000 suppliers, real prices from 250,000 negotiated contracts. That fuel trains their negotiation agent “Ana.” For you: an agent trained on hundreds of thousands of real negotiations can’t be replicated in-house. It’s a data moat. (Vertice / PR Newswire, 2026.)
Coupa acquires Tonkean (May 21). Coupa (taken private by Thoma Bravo at $8B in 2023) adds AI-native intake and orchestration (natural language, 250+ connectors). Its 4th “autonomous spend” acquisition after Cirtuo, Scoutbee and Rossum. For you: if you already run on Coupa, the orchestration layer is coming to you, whether you budgeted for it or not. (Coupa, 2026.)
Deel acquires Sastrify (~May 5). Deel (global payroll and workforce, ~$17.3B valuation, IPO targeted 2026) folds Sastrify (Cologne, AI-driven SaaS buying and management) into “Deel IT.” An HR platform that starts managing your software spend: the line between indirect categories blurs. For you: your SaaS management could arrive bundled inside an HR contract, off procurement’s radar. (Axios / TechCrunch, 2026.)
Three very different acquirers (a procurement pure-player, a spend-management suite, an HR platform) are all after the same asset. Software no longer makes the difference; the price data and negotiation history do now. Concrete consequence: negotiating power shifts to whoever holds that data.
If it isn’t you, it’s your platform.
Before your next SaaS renewal, ask your vendor a simple question: where does the data from my past negotiations live, and can I export it?
An augmented buyer in action.
We start with the easiest use case: data enrichment. We’ll move further down the procurement process over the coming issues. What if you restarted that project you’d shelved for lack of data? If it lives in PDFs, extracting it has never been easier.
A project stalled for lack of data: rationalizing a panel, a circular-economy project, a category audit. The data exists, but it’s asleep in PDFs (company sheets, contracts, reports).
An AI connected to your Excel (Claude or ChatGPT wired to the file, to read, check and improve it). Your evaluation criteria and the expected output written into the context, before you run anything.
Don’t expect a perfect result on the first pass. The model returns a first version at 80%, with gaps and a few errors. That’s normal, and it’s the point. You validate on a sample, add two or three precise instructions (“the SIREN is in the footer,” “ignore subsidiaries”), and rerun. Two or three iterations, and the extraction becomes very reliable. The first pass is a draft, not a verdict.
Here’s my supplier file (missing columns: SIREN, financial health, certifications, category) and 12 PDFs. For each supplier, extract the missing fields from the PDFs and fill the table. Don’t guess: if a value is missing or uncertain, leave it blank and flag it. Start with 3 rows, I validate, then you do the rest.
A usable file in a few hours instead of a few weeks. The project you’d shelved becomes doable again.
The first results will probably be incomplete, even wrong. That’s expected. You have to iterate and build the most complete context possible: criteria, examples, extraction rules. Method before tool, always.
Start with the data. It’s the easiest use case, and the one that clears the way for all the others. Next month: sort a longlist of 8 suppliers into a ranked shortlist of 3.
Need a hand getting started or improving your first results? Write to me, I reply to every message.
The player to watch: Lio.
I’m launching this rubric, and fate lands on the company that shares my last name. I promise it isn’t staged. And no, it isn’t mine: we just share the name and the obsession with indirect procurement.
Lio (ex-askLio) is a multi-agent platform for indirect procurement: supplier research, negotiation, approvals and delivery tracking run in parallel. A $30M Series A led by a16z in March 2026, ~$33M raised in total, clients Munich Re, Brose, Novozymes. (Lio / a16z, 2026.)
What’s in it for you: concretely, its agents take over tasks your teams do by hand today. Finding and qualifying suppliers, preparing and running a negotiation, routing approvals, tracking deliveries. All of it on top of your ERP rather than inside it, so no new IT project. The CFO question: on which repetitive procurement tasks would a layer like this save you the most time?
The person to watch: Jason Busch.
Founder and CEO of Spend Matters since 2004, the world’s reference for procuretech analysis, and founder of Azul Partners. Co-founder of Public Spend Forum, ex-FreeMarkets, 1,000+ published analyses. (Spend Matters / Azul Partners, 2026.)
Why I follow him: a visionary, an AI enthusiast, and some of the sharpest analysis on the market. He doesn’t just read the deals (like our three acquisitions this month), he draws out the implications for the whole chain: buyer education, SaaS-vendor economics, procuretech consolidation. The read that gains altitude when the noise rises.
On your indirect procurement, which workflow would free the most value if you augmented it first? And have you priced it yet?
I collect the answers and turn them into an anonymized synthesis in issue #2. The 5 sharpest get their method detailed.
Reply to Alex →