Every supply chain has a dirty secret. Somewhere between a raw material leaving its source and a finished product reaching the customer, value is being lost - not stolen, not wasted through obvious negligence, but quietly eroded at stages that exist largely out of habit. These are pass-through stages: intermediary steps that add cost, time, and complexity without adding meaningful value.

For decades, identifying and eliminating these stages was a manual, expensive, and often politically fraught process. Consultants would be brought in, process maps drawn, and recommendations made - only to be shelved when the operational disruption seemed too high. Today, AI is changing that calculation entirely.

What Is the Pass-Through Problem?

A pass-through stage is any link in a supply chain that primarily receives, holds, or forwards goods or information - without transforming, verifying, or genuinely adding value to either. Think of a regional distribution hub that exists only because it was built in 1998 and the contract has never been reviewed. Or a procurement process that routes every supplier invoice through three approval layers regardless of value. Or a quality checkpoint that flags issues no downstream step actually cares about.

"Your margin is my opportunity." - Jeff Bezos. The quote has defined an era of supply chain disintermediation, and AI is now operationalising it at scale.

The concept is closely related to disintermediation - the removal of intermediaries from a supply chain. But where disintermediation has historically been driven by business strategy (Nike going direct-to-consumer, Dell bypassing retailers), AI-driven elimination of pass-through stages is data-driven. It is not a boardroom decision; it is an operational diagnosis.

Why This Problem Is Getting Worse, Not Better

Modern supply chains are longer and more complex than they have ever been. Globalisation layered on geographic spread. E-commerce layered on demand volatility. ESG requirements layered on top of both. According to recent research, 65% of logistics costs are now tied to last-mile delivery and inventory inefficiencies - and much of that cost is structural, baked into supply chain architectures that were designed for a different era.

As of 2025, only 23% of supply chain organisations have a formal AI strategy in place - meaning the vast majority are still running supply chains whose inefficiencies are largely invisible, or visible but unaddressed.

The problem compounds over time. Each pass-through stage introduces latency. Latency creates buffer inventory. Buffer inventory ties up working capital. And the longer a stage goes unchallenged, the more organisational infrastructure grows up around it - making it even harder to remove.

How AI Diagnoses Supply Chain Inefficiency

This is where AI's value proposition is most concrete. Supply chains generate enormous volumes of structured and unstructured data - purchase orders, lead times, carrier delays, inventory counts, quality rejections, invoice cycles, warehouse dwell times. Human analysts can review a slice of this data; AI can process all of it, continuously, and surface patterns that no human team could find at scale.

End-to-End Visibility

AI-powered dashboards now compile performance indicators across every node of a supply chain into a single, real-time view. Bottlenecks, redundancies, and dwell anomalies become visible not quarterly, but as they happen. What once required a multi-week consulting engagement to diagnose can now be flagged automatically.

Supplier and Stage Benchmarking

By continuously tracking supplier lead times, defect rates, contract compliance, and cost variability, AI systems can identify which stages are performing below the threshold of economic justification. Siemens' AI procurement platform, for example, evaluates over 15,000 suppliers against 200+ criteria - reducing procurement cycle times by 60% and generating 11% cost savings.

Anomaly Detection and Pattern Recognition

Machine learning models trained on historical supply chain data can distinguish between expected variance and structural inefficiency. A shipment arriving two days late during peak season is noise. A distribution node that consistently adds four days of dwell time without any value-adding activity is a signal - and AI flags it as one.

Scenario Modelling and Digital Twins

Before recommending the removal of a supply chain stage, the risk of doing so needs to be assessed. Digital twins - virtual replicas of the supply chain - allow organisations to model the downstream effects of removing a stage, renegotiating a supplier relationship, or rerouting a logistics flow before committing to a real-world change. Procter & Gamble's supply chain digital twin now simulates over 3,500 manufacturing facilities and 100,000+ shipping lanes for exactly this purpose.

‍

Also Read: Why Pharma Supply Chain AI Fails Without End-to-End Data Integration

Where the Inefficiency Usually Hides

AI-driven supply chain audits consistently surface the same categories of pass-through inefficiency:

1. Unnecessary intermediaries in procurement - brokers, agents, and spot-market platforms that charge for access without improving the underlying transaction

2. Over-engineered approval workflows - multi-layer sign-off processes calibrated to high-risk transactions but applied uniformly across all spend categories

3. Redundant quality checkpoints - inspection stages that duplicate work already done upstream, or that check for defects that have never actually occurred in a given supplier relationship

4. Regional distribution nodes maintained by legacy contract - warehousing and fulfilment assets that continue to exist because exit costs are never weighed against ongoing inefficiency

5. Manual data re-entry between systems - administrative pass-throughs where information is taken from one system, reviewed by a human, and entered into another, adding time and error risk without adding intelligence

Intel's experience is instructive on the last point. Its AI-powered fraud detection system now analyses 3 million daily procurement transactions, identifying suspicious patterns with 96% accuracy and preventing $47 million in procurement fraud annually - while detecting compliance violations 35 days earlier than manual auditing. The manual audit stage it replaced was not just slow; it was structurally incapable of matching AI's throughput.

The Disintermediation Question: Cut or Reconfigure?

Not every pass-through stage should simply be eliminated. Some intermediaries provide genuine value that is simply not being measured correctly - after-sales support, local market knowledge, supplier relationship management, or logistics capabilities that would be expensive to replicate internally.

AI helps here too, by moving the question from "should we keep this stage?" to "what does this stage actually cost us, and what would it cost to replace it?" That framing shifts the conversation from political to economic - and makes the answer far less contestable.

The goal is not the shortest possible supply chain. It is the most value-dense one. AI provides the diagnostic layer to tell the difference.

In practice, this often leads to partial disintermediation: removing intermediaries from specific product lines or customer segments while retaining them where they genuinely outperform internal alternatives. This is the model now being adopted across distribution-heavy industries, where suppliers are identifying which customers and products to serve direct and which to deliver through third parties - based on AI-generated margin and fulfilment analysis, not assumption.

Agentic AI: From Diagnosis to Autonomous Action

The next frontier is not just identification but action. Agentic AI - systems that can autonomously pursue goals by planning multi-step operations and adapting to changing circumstances - is beginning to close the loop between diagnosing supply chain inefficiency and doing something about it.

GSK's deployment offers an early model. The company built an AI procurement system called 'GSK I Need to Buy', in which an AI agent analyses supplier quotes in real time, identifies savings opportunities, and triggers competitive sourcing events without human initiation. A separate negotiation agent automatically improves payment terms or lead times on sole-sourced purchases. The result: faster cycle times, stronger compliance, and measurable savings - without adding headcount.

At Target, an AI platform monitors 1,900+ stores in real time, processing 4.5 million data points hourly to detect inventory anomalies. Out-of-stock incidents have been reduced by 40%, and response time to supply disruptions has dropped from 2-3 days to under four hours. That speed difference is the difference between managing a disruption and being managed by it.

Also Read: Agentic AI in Pharmaceutical Operations: Beyond Track-and-Trace

The Implementation Reality: What Leaders Get Wrong

The technology is proven. The challenge is almost always organisational. Supply chain AI deployments frequently stall at the following points:

Data fragmentation

AI is only as good as the data it trains on. Supply chains where operational data sits in separate ERP instances, carrier portals, supplier spreadsheets, and legacy warehouse systems cannot be diagnosed holistically until those data sources are unified. This is the foundational work that precedes meaningful AI deployment - and it is frequently underestimated.

Change management resistance

Pass-through stages often have owners. Removing them means changing someone's job, renegotiating a contract, or restructuring a team. AI can identify the inefficiency, but the decision to act on it is organisational. The companies making fastest progress are those where supply chain AI deployment has senior executive sponsorship and clear programme ownership.

Treating AI as a point solution

Route optimisation, demand forecasting, and supplier scoring are valuable in isolation. But the real value of supply chain AI is systemic - across procurement, logistics, warehousing, and planning simultaneously. Organisations that deploy AI piecemeal often find they have optimised one node while leaving inefficiency in adjacent ones.

What the Numbers Say

The operational and financial case for AI-driven supply chain optimisation is no longer theoretical:

6. Manufacturers typically see 10-30% reductions in operational costs from AI-driven supply chain insights

7. Inventory right-sizing of 20-40% is achievable through AI-led demand forecasting

8. Organisations using decision intelligence to automate decisions and predict disruptions are outpacing peers by 17% in customer satisfaction and 34% in operational efficiency

9. By end of 2026, an estimated 60% of businesses will use AI-powered warehouse solutions - up from just 10% in 2020

10. Novartis used AI to eliminate chronic procurement bottlenecks across thousands of monthly purchase requests; GSK's agent-driven model captures value from spend categories that previously fell below the threshold of manual review

Where to Start

For supply chain and operations leaders who recognise the pass-through problem but are unsure where to begin, three starting points consistently prove useful:

1. Map before you cut

AI-powered supply chain visualisation tools allow you to produce a complete, data-driven map of every stage - including dwell times, cost per stage, defect introduction rates, and lead time contribution. Without this baseline, elimination decisions are anecdotal.

2. Benchmark against value, not activity

The question is not whether a stage is busy, but whether it is creating value. Reframe every supply chain stage in terms of what it costs to maintain versus what would be lost if it were removed. AI can run this analysis continuously and flag stages whose cost-to-value ratio deteriorates over time.

3. Start with procurement, where the ROI is fastest

Procurement is where AI adoption is most mature and where pass-through inefficiency - over-approval, redundant vendor tiers, manual spot-buying - is most immediately measurable. The deployments at GSK, Pfizer, Novartis, and Siemens have all proven that agentic AI in procurement delivers returns within a single quarter of deployment.

The Bottom Line

The pass-through problem is not new. Supply chains have always carried the weight of their own history - stages, intermediaries, and workflows that made sense at the time they were designed and have never been seriously revisited since.

What is new is the ability to see it clearly, continuously, and at scale. AI doesn't just speed up supply chain decision-making. It changes what decisions become visible in the first place. The inefficiencies that have survived this long have done so largely because they were too granular, too distributed, or too politically inconvenient to surface through manual analysis.

That cover is gone.

In 2026, the supply chains that are winning are not necessarily the most sophisticated. They are the leanest - and they know exactly where every unit of cost is sitting and why.

The pass-through problem is solvable. The question is whether your organisation is ready to let AI show you where to look.

VE3 Global helps enterprise organisations deploy AI across supply chain, procurement, and operations. contact us for more information