Most enterprise procurement systems were designed for a world that no longer exists. A world where demand could be forecast by looking at last year’s sales. Where procurement operated like a train timetable — suppliers on the rails, prices (largely) fixed aside from negotiated savings (often lost year-over-year), and risk confined to weather delays and the occasional supplier bankruptcy.

But then the real world happened. Supply volatility. Category inflation. Shrinking cycle times. ESG complexity. Political dislocation. Export controls. Import duties. War. Demand surges that couldn’t be predicted by a macro in Excel.

In this world, deterministic systems — rule-based engines running on workflows, exception flags, and stale approvals — don’t cut it. They were designed to enforce compliance, not think.

Which is why they’re logically segmented into modules that in theory work together: spend analytics, category management/strategy, sourcing, advanced sourcing (optimization), supplier management, supplier/supply/supply chain risk management, contract management, e-procurement, accounts payable automation and working capital management (to name many of them).

Deterministic platforms (think S2P and P2P suites and best-of-breed modules) generally do what you tell them — as long as you told them in advance, in the right format, with clean data, and the wind blowing from the right direction. They operate in one scenario: the one you mapped during a six-month implementation.

They’re ideal when you know exactly what will happen. And they work if humans are in the loop and the FTEs can scale to the amount of work at hand without any black swans in the market.

But deterministic models break down at scale, especially under uncertainty and complex multifactorial decision making.

Here’s where probabilistic systems come in (and blow up the suite and module paradigm entirely).

They don’t just log activity. They interpret it. They don’t just follow rules. They adapt them. They don’t just support human users. They act like them — only much faster, with 1000x better memory, and a hell of a lot less passive aggression.

Probabilistic systems start from the premise that uncertainty isn’t the edge case — it’s the operating environment. These systems:

• Learn from structured and unstructured data (contracts, commodity prices, supplier logs, external news, ERP trails)
• Quantify confidence intervals and risk trade-offs
• Provide dynamic scenario-based recommendations, not fixed flows
• Make decisions when confidence is high, and escalate only when low
• Continuously refine their reasoning models based on feedback

This is more than analytics or source-to-pay in a box (or chevron!) It’s applied cognition. Autonomy with a leash. AI employees, not another layer of dashboards.

And most important, these systems are not bolted on to legacy workflows. They’re trained into the fabric of your categories, suppliers, and negotiation strategies. Their reasoning is embedded — not appended.

To bring this example alive, let’s look at a real-world mess: European retail.

Retail/Grocery Determinism in Procurement and Supply Chain Systems

Imagine you manage direct procurement for a European grocery chain — call it 1,000 stores, 25,000 core SKUs (representing 80% of revenue), and a category mix that spans produce, dairy, packaged goods, and private-label everything. You’re responsible for negotiating everything from butter fat and PET resin to corrugate and cling film.

Your tooling?

• SAP (including Ariba) for contracts and POs
• A patchwork of planning modules from vendors with names ending in “soft”
• A daily ritual involving Excel, perhaps a Bloomberg tab (if your trading desk bought one), and mild despair
• Price indices tracked by hand, often copied from PDFs, stored in email chains, or pasted into VLOOKUP’d hell

Your negotiation cycle is, well, cyclical:

• Quarterly contract reviews, unless something explodes
• Cost change requests from suppliers are reviewed “as they come in”
• You react to price hikes — usually after they’re already baked into the invoice
• Some buyers watch commodity markets obsessively. Others go by feel
• Weather disruptions? You hear about them from a supplier
• Currency spikes? Your finance team sends a memo
• You don’t have a unified view of which SKUs are exposed to which commodity, which contract has a price escalator, or which supplier is padding margins

Your team negotiates based on experience, not probability. And the outcomes? Good enough — until they aren’t.

Oh yeah – and a strategy and operations consulting firm with a name that starts with a B, A or M comes in every so often, does a study for you showing eight figures of savings, drops it in your lap – and exits before the first wave of savings ever gets implemented (if it does).

The Probabilistic Alternative

Now imagine your system — not a person, a trained AI employee — does the following:

Continuously monitors commodity indices for inputs across thousands of SKUs:

• PET, pulp, aluminum, wheat, palm oil, butter fat, resin, tinplate, soy, diesel, CO₂, corrugate, and freight — all dynamically linked to the exact line items in your SKU catalog
• Maps contracts to indexes and clauses, understands who you’re exposed to (and how), and knows which suppliers hedge and which pass everything through

Models market volatility based on a combination of:

• Live commodity feeds
• Historical spread behavior
• Currency fluctuations
• Weather anomalies
• Crop reports and freight bottlenecks

Identifies high-margin suppliers that have not yet passed on inflation, signaling a window for renegotiation before they do

Flags price softening in key inputs like soy protein or flex film — and recommends a contract reopening on relevant SKUs tied to those materials

Scores your leverage position:

• Contract duration
• Supplier dependency
• Alternate source availability
• Past concession history

Suggests timing and approach:

• Wait for volatility to settle?
• Engage now while you’re holding the pricing cards?
• Lock in longer terms because market exposure is rising?

And then — with human oversight — it initiates the renegotiation:

• Drafts the clause revision
• Benchmarks the input price shift
• Simulates impact on category margins
• Routes for escalation only if risk or legal thresholds are breached

No post-hoc spreadsheets. No reaction to a price hike email. Proactive, probabilistic, preemptive negotiation — driven by thousands of small signals, not quarterly gut checks.

This isn’t an “AI assist.” This is a superhuman–something (or someone if you care to personify the actor) that sees every SKU’s input dependencies, scans the market landscape, and knows the precise moment you should move first.

Let’s now shift our case study to CPG.

CPG Procurement Determinism

Imagine you lead packaging procurement for a $25B+ global CPG firm. Think beverage cans, condiment containers, cleaning product lids — basically everything a toddler might try to crush or lick.

You manage aluminum and tinplate contracts across three regions. Your suppliers include a couple of global giants and some niche converters. Contracts are indexed — sort of.

In your tech and data arsenal, you’ve got:

• SAP S/4HANA and SAP Ariba for POs and contracts
• A metal index subscription in someone’s inbox
• An internal “pricing tracker” spreadsheet built by a retired manager that nobody understands
• An unspoken agreement to negotiate when the market moves enough to feel uncomfortable

You’re exposed to Aluminum 5052 and tinplate stock, with pricing “tied” to the LME (but without visibility at the form, grade and alloy level – where over 50% of the total cost is). Moreover, half the clauses are vague, the escalator math changes by supplier, and freight surcharges come out of nowhere.

In reality, you’re not negotiating. You’re riding the commodity and supplier wave, responding and triaging as new information comes in (or when a supplier screams). For example:

• LME spikes → panic
• LME drops → wait and see
• Supplier calls → schedule a quarterly review

In practice, your team “reviews the data,” opens Excel, updates a VLOOKUP tab, argues over whether to push back now or in Q1, and then… sends a polite email.

Meanwhile, prices drift. Margins compress. Nobody’s sure if that last 6% hike was market-driven or “just because.”

The Probabilistic Alternative: Everything Everywhere All At Once

Now imagine a probabilistic procurement system — an AI employee if you will – that:

• Tracks live LME feeds and pricing at the form, grade and alloy level (not to mention primary and secondary pricing and regional scrap differentials for aluminum and tinplate). Gratuitous plug: MetalMiner (which enables this)
• Monitors conversion cost trends, freight rates, and energy inputs (e.g., natural gas volatility is no joke if your can supplier is in the EU, especially in the DACH markets, which decided for whatever reason to commit energy seppuku)
• Maps all supplier contracts to their exact index clauses, including base metal references, lag structures, escalator math, and known surcharge tactics
• Flags inconsistencies in pass-through logic across regions and vendors — surfacing when you’re overpaying compared to your own global benchmarks
• Analyzes market/index backwardation or contango to forecast future pricing risk
• Detects “margin stacking” patterns (supplier margin expands when raw inputs drop — neat trick)

Then it models negotiation timing, factoring in:

• Contract renewal windows
• Historical supplier concessions
• Current leverage (demand forecasts, alternate capacity, ESG risk profiles)
• Market sentiment and regulatory chatter (e.g., EU tariffs, CBAM adjustments)

Finally — based on this model — it triggers action:

• Alerts you that Aluminum 5052 spot prices have fallen 12% over 6 weeks with a flat or rising price curve the next few months
• Notes your suppliers have not adjusted the base cost
• Generates a clause revision tied to trailing 30-day average vs. trailing 90
• Simulates net savings across 37 affected SKUs
• Flags legal for review of revised pass-through logic
• Suggests bundling the ask with volume commitments to extract more favorable payment terms

And the only thing your team has to do?

Click “approve” or “edit offer.” And take the credit (no six or seven figure consulting study required).

That’s not “price intelligence.” That’s category-specific AI autonomy, built on commodity signals, contract memory, and actual strategic timing — not “wait until Q2” muscle memory.

Because in these categories, you don’t just negotiate price. You negotiate when.

And the system should know before your supplier does.

What Legacy Vendors Can’t (and Won’t) Do

If you think your legacy vendor is going to get you there, guess again. Sure, they’re adding co-pilots, LLMs and “agentic” capabilities – whatever that means.

But as a 20-year analyst, what I’ve seen are features that are mostly cosmetic, including chatbots that read help docs faster, classification engine sthat still needs manual override and dashboards that visualizes what already happened

In short, their “autonomy” is skin-deep, usually confined to recommendation engines that require a human to act. Worse, most are built atop rule-based orchestration engines that break under complexity. You can’t layer real autonomy on brittle plumbing.

And because they’re all about enforcing process, they can’t let systems make decisions — only support them.

In contrast, true probabilistic systems:

• Start with intent, not configuration. You don’t define workflows. You define goals. The system figures out how to get there
• Are agent-based. Each agent is trained on a category, risk model, or contract corpus. These agents aren’t chatbots. They’re domain specialists
• Handle ambiguity. They’re comfortable with imperfect data, incomplete signals, and novel scenarios. They don’t break — they infer
• Act autonomously. They only escalate decisions when thresholds for uncertainty or risk are breached. Otherwise, they execute
• Learn continuously. Every decision feeds back into the model. Every win or mistake sharpens the agent’s judgment
• Operate modularly. They can plug into Ariba, Coupa, SAP, or whatever else you use — but their brain doesn’t live in the old system. They build their own neural map of your categories, bills of materials (or ingredients), supply network and supply chain

And perhaps most critically: it’s not a module. Or a suite. Or a new dashboard or supply chain control tower .

No, this isn’t “procurement software.” It’s an autonomous operational infrastructure. Deployed not to digitize what already exists — but to create what couldn’t.

In short: it’s about radical autonomy.

In this world, it’s not about feature/function in the core tech (which raises a really interesting point about industry analysts like Spend Matters and Gartner – I’d argue the “bench test” as in processor speed in hardware is a more appropriate fit here than feature/function comparison going forward in this world).

It’s about outcomes.

How your sourcing and category engine thinks. How your risk and operating model speaks.

It’s a world where your system doesn’t ask “what now?” — it tells you “here’s what we’ve done.”

And if your legacy vendor says they can do this, all the power to them.

I’d have a few questions I’d ask them if I were still an industry analyst.

But I’m not anymore.

I’d rather build this stuff and take part in a revolution that represents the most significant shift in procurement and supply chain technology since the verticalization that took place around the industrial revolution (and de-verticalization post WWII).

Yes, it’s that big.

And it’s a gift to be alive and to play a small part in driving it – a once in a multi-generation opportunity in not just procurement and supply chain technology, but business and trade overall.