PROCUREMENT & SUPPLY CHAIN
Rebuilding supply chain and procurement for a world that won’t sit still — and why the sequence you fix things in matters more than the tools you buy.
It’s 9:14 on a Monday morning, and the plan is already wrong.
A vendor who confirmed delivery on Friday is now two days out. Nobody finds out at 9:14 — they find out at the 11 a.m. production review, by which point the missed truck has already cost four hours of line time. By lunch, procurement is calling three alternate suppliers, finance is asking why safety stock didn’t cover the gap, and the plant head is asking why nobody saw this coming.
Nobody saw it coming because the system was never built to show it early. The ERP was accurate. It just wasn’t fast enough to matter.
Swap the vendor, the plant, the industry — automotive, pharma, FMCG, discrete manufacturing, it doesn’t matter much — and the scene repeats with different actors and the same script. This isn’t a story about one bad Monday. For most manufacturing supply chains, it’s the operating model.
Procurement and supply chain leaders today are being asked to do three things at once that don’t naturally sit together: hold cost flat, hold risk down, and move faster — while tariffs, currency, single-source dependencies and talent turnover keep shifting the ground underneath. That tension, not whatever sits on today’s agenda, is the actual job.
Root causes
Where the pain actually lives
Across auto, pharma, discrete manufacturing and logistics, the surface symptoms differ. Six root patterns don’t — and most leaders relax a little once they hear someone else say it first.
Visibility that arrives too late to act on
Plans run on data that’s accurate and still late — a vendor slip, a quality reject, a demand spike surfaces in the review meeting, hours or days after it actually happened, not at the moment it happened.
Volatility that has become the baseline
Tariff shifts, freight disruption and currency swings aren’t the exception line-item anymore — they’re the operating environment. Single- and dual-source dependencies built for a calmer world are now the least-priced risk on the balance sheet.
Collaboration debt with the vendor base
Most OEM–vendor relationships are still PO-out, ASN-in. A vendor learns about a schedule change the same way it learns about a new order — after the fact. Multi-tier dependency stays invisible until it becomes a stockout.
Working capital squeezed from both ends
Safety stock grows to buy resilience. Days sales outstanding stretches because customers protect their own cash. Inventory and receivables quietly draw down the same rupee — often against two different, disconnected numbers held by finance and by supply chain.
Institutional memory walking out the door
A large share of experienced planners and buyers is at or nearing retirement age. Much of what they know — which vendor to call first, which exception is safe to ignore — was never written down as a rule, so it leaves with them.
Compliance and ESG moving from report to design constraint
Scope 3, SEDEX and traceability requirements are moving upstream into the sourcing decision itself, not staying in a year-end report. Most organisations still treat this as an annual audit exercise rather than a live input to buying decisions.
Most of what we call a “technology problem” in supply chain is actually a process and trust problem wearing a technology costume.
Fix the sequence — process, then collaboration, then technology, then AI — and adoption stops being a fight. Reverse it, which is what most organisations do under pressure, and every new system becomes one more silo.
The tool-first instinct
Why reaching for AI first keeps failing
Under pressure, the instinct is to buy something. Here’s what the data says happens when that instinct wins.
- 95% of enterprise generative-AI pilots deliver no measurable P&L return; of the tools organisations evaluate, only 20% reach a pilot and just 5% reach production. MIT Project NANDA, “The GenAI Divide,” 2025 — ~300 deployments, 150+ executive interviews
- In the same Gartner analysis of 600 supply chain decision-makers, more than half of leaders who relied on digital models for a recent strategic decision said they would have reached a better outcome without them. Gartner, 2023
- 72% of leaders have had to reopen an already-approved network decision at least once, adding delay — a pattern researchers call “decision debt,” and it compounds the same way technical debt does. Gartner, 2026
- Just 17% of leaders are pursuing an immediate, transformational redesign of process and workflow; 83% are layering AI incrementally onto what already exists. Gartner’s own read: the constraint is master data alignment and process maturity, not the model. Gartner, May 2026 — 140 senior leaders
AI layered onto a broken process and messy data doesn’t fix either — it just makes bad decisions faster, and harder to trace. That’s not an argument against AI. It’s an argument for where it sits in the sequence.
The framework
Get the sequence right
The organisations getting durable results do the opposite of reaching for a tool first. Four layers, in order.
Process
Find the real constraint before you fix anything
Find the constraint first. At any moment, one or two points limit the flow of the whole chain, and most improvement effort goes into steps that were never the bottleneck.
- Name the current constraint explicitly — machine, vendor, approval step, or missing information.
- Separate decision rights from data access: who can actually act on an exception, and do they have that authority today?
- Agree one shared definition of “on time,” “critical” and “at risk” across functions.
Collaboration
Move vendors from transaction to co-planning
Shift key vendor relationships from PO-and-ASN exchange to shared, live production and material plans, with tranche commitments instead of binary confirm/reject.
- Map multi-tier dependency for the top-risk categories, even informally.
- Build risk-sharing into contracts instead of pushing all volatility onto the smaller partner.
Technology
One source of truth before more dashboards
Integration beats another point solution. Real-time beats retrospective — an exception alert decays fast.
- Master data discipline — one item master, one vendor master, one customer master — is unglamorous and non-negotiable.
- Most “integration failures” are master-data failures wearing a different name.
AI
Only once the first three are real
This is where prediction, matching and bounded autonomous exception-handling earn their keep — proposing a re-route, a re-sequence, a reorder within pre-approved guardrails. Not because AI isn’t powerful, but because it amplifies whatever foundation it’s given.
Proof, not theory
What the sequence looks like in practice
CASE · ELECTRONICS MANUFACTURING
Lenovo didn’t start its resilience push with an AI pilot. It started by pulling planning, sourcing, manufacturing, delivery and quality into a single control tower — Supply Chain Intelligence (SCI) — one shared, real-time view across a network of 30-plus manufacturing sites spanning 180 markets. AI and machine learning were layered on top of that unified base, not ahead of it.
Non-negotiables
The foundation AI actually needs
This is the part most transformation roadmaps skip — worth being blunt about with peers who are all under pressure to “do something with AI” this year.
Data
- Single, owned master data for items, vendors and customers — AI prompted against three versions of the truth confidently produces a fourth.
- Data lineage and freshness: know where a number came from and how old it is before trusting a recommendation built on it.
Process
- Standard definitions across functions for the terms that drive decisions — codified, not tribal.
- Documented decision rights: which recommendations AI can surface versus which it’s allowed to act on, and who owns the override.
Governance
- Explainability as a requirement, not a nice-to-have — a recommendation a planner can’t interrogate is one they’ll quietly ignore.
- An audit trail for every AI-influenced decision, and a human-in-the-loop model that reviews flagged exceptions, not every output.
Bring this to your next roundtable
Questions worth sitting with
Three prompts, better than conclusions
- What’s the one exception in your supply chain that, known three days earlier, would have cost you nothing — and cost you plenty by the time you found out?
- Where in your vendor base do you still find out about a problem the same way you’d find out about a new order — after the fact?
- Where are you most tempted to reach for an AI tool before you’ve fixed the process underneath it — and what’s stopping you from checking?
The organisations that pull ahead over the next few years won’t be the ones that adopted AI first. They’ll be the ones that got honest first — about where the real constraint sits, about how transactional the vendor relationships still are, and about whether the data would survive being questioned by a system that doesn’t share its assumptions.
AI is a multiplier. Point it at a good foundation and it compounds. Point it at a shortcut, and it just breaks things faster.
— A working note for procurement and supply chain leaders
- Gartner, analysis of 600 supply chain decision-makers, 2023 — via Supply Chain Management Review, “What comes after visibility?”
- Gartner, survey of 506 supply chain leaders, December 2024 — Gartner Newsroom, May 2025.
- The Hackett Group, 2025 U.S. and European Working Capital Surveys.
- MIT Project NANDA, “The GenAI Divide: State of AI in Business 2025.”
- Gartner, 2026 decision-debt survey, and Gartner survey of 140 senior leaders, May 2026 — via Supply Chain Management Review / Alation.
- Lenovo, Supply Chain Intelligence control tower — Lenovo Newsroom and Gartner Asia/Pacific Supply Chain Top 10, 2023.