Welcome to the September 2026 edition of Gordon’s Window with Gordon Donovan. This is your comprehensive guide to the latest shifts and challenges in global procurement and supply chain management. Save your own time reading research when you can get the key takeaways in one place!
Every source this month points to the same underlying issue, even when the surface topic differs.
Whether the subject is agentic AI, contract management, supplier onboarding, talent strategy or spend benchmarking, the finding is consistent: the architecture is wrong. Not the strategy. Not the intent.
The underlying structure of how procurement operates, how it holds data, how it sequences decisions and how it governs its own technology spend is no longer fit for what it’s being asked to do.
The first is the agentic inflection. Multiple reports this month mark 2026 as the year the conversation moved from AI as a productivity tool to AI as an operating model.
BCG quantifies the prize at 60 percent of buyer capacity freed, eight to 15 percent in cost savings and 30 to 60 percent faster sourcing.
But the prize only materialises when procurement redesigns its workflows for AI agents rather than layering AI onto existing processes. Most organisations have not done this. They are adding copilots to broken workflows and calling it transformation.
The second is the data and knowledge deficit. Sirion and Achilles both report the same constraint from different angles.
Seventy percent of AI projects fail in contracting because the underlying contract data is fragmented, unmonitored or disconnected from live systems. Supplier onboarding processes are 57 percent manual and supply market intelligence is ad hoc or absent in 42 percent of organisations. You cannot run agents on data you cannot trust.
The third is the governance gap. Gartner’s intervention this month is direct: AI is no longer being bought through a single category. It is being purchased across the enterprise, function by function, without a unified view of spend, risk or outcome.
Treating AI as its own category management domain is not a theoretical recommendation. It is a commercial necessity. The organisations that fail to do this will discover the cost of fragmented AI governance the same way they discovered the cost of maverick spend: in retrospect, at scale and expensively.
BCG – Executive Perspectives: From AI Assistance to Agentic Orchestration
BCG published this executive perspective as a companion to its broader AI in procurement research.
The core argument is architectural. Most procurement organisations are still running a point-based AI model: isolated use cases, fragmented processes, AI as an insight layer rather than an execution layer.
Leading organisations are moving to an agentic operating model in which AI agents orchestrate end-to-end workflows, and humans steer decisions, priorities and exceptions.
BCG quantifies the outcome difference clearly. Agentic procurement can free up 60 percent of buyer capacity and deliver cost savings of eight to 15 percent, on-time-in-full improvements of five to 15 percentage points and sourcing cycle-time reductions of 30 to 60 percent.
Transactional procurement costs fall 20 to 40 percent. Contract and policy adherence improves 20 to 40 percent. CO2 emissions drop 50 percent.
None of those numbers materialises through copilots and isolated automation. They require an AI-enabled operating foundation built on six elements: connected data and systems, a deep knowledge layer, AI capabilities and talent, an AI-first organisation, strong process governance and cross-functional collaboration.
The transition is not a technology deployment. It is a change management program with technology inside it.
Leaders are redesigning the work first, then applying AI to the redesigned process. That sequencing distinction separates organisations that compound value from those running expensive pilots that do not scale.
ProcureCon: 2026 Total Talent Asia Benchmark Report
The blended workforce is no longer a strategy. In APAC, it is the operating reality. Permanent employees sit alongside contingent workers, specialist freelancers, SOW providers and increasingly AI-augmented workflows.
The strategic case for managing them together has been accepted. However, this ProcureCon benchmark finds that acceptance and execution are not the same thing.
Only 18 percent of APAC organisations operate a mature, integrated total talent model. Nearly half (48 percent) have a unified structure on paper but cannot execute it consistently across borders.
That is the middle maturity trap: enough structure to feel organised, not enough consistency to deliver value.
The challenge is not strategy. It is orchestration.
Governance sits at the heart of the problem. HR exclusively owns total talent strategy in 50 percent of organisations. Procurement owns it in only 10 percent. Yet procurement manages much of the external workforce complexity: contingent spend, SOW governance, supplier panels and risk oversight. The mismatch creates accountability gaps that no governance model resolves.
Two findings capture the attrition-and-readiness tension.
Twenty-eight percent of employees leave because of a lack of flexibility in the workplace, yet only 16 percent of organisations prioritise flexibility as a retention investment.
Leaders rate their 2030 AI readiness at 7.14 out of 10, yet 42 percent report critical or emerging gaps in AI fluency and cross-market capability.
Confidence is running well ahead of capability.
Sirion: How to Get Contracting Right in the Age of AI
Deloitte estimates the cost of dormant contract knowledge at roughly $2 trillion a year globally. Sirion’s paper addresses that gap.
The core diagnosis is what Sirion calls the ‘memory gap’: every contract negotiation starts as though the organisation has never contracted before. The positions fought for, the terms avoided and the outcomes already known do not carry forward.
Eighty percent of organisations work from contract data that is not regularly refreshed. Sixty-one percent still rely on manual interpretation to understand their contracts, while 62 percent struggle to locate previously approved contracts.
The repository era has run its course. Sixty-seven percent of organisations now have a repository of signed contracts, but only 24 percent monitor contracts after signature.
Just 13 percent have digitised playbooks and only seven percent treat contract data as authoritative enough to drive transactions elsewhere in the business.
Stored is not the same as trusted.
AI makes this worse before it makes it better. Forty-six percent of AI proofs-of-concept in contracting are scrapped before reaching production. The actual reason is that the underlying contract knowledge was never made usable. AI does not repair weak data. It exposes it.
The organisations that have crossed this gap are posting real results: 37 percent of time reclaimed in legal, 33 percent reduction in vendor spend in procurement, 43 percent time savings and one to two revenue points in sales. One oil and gas company reported $235 million in hard savings.
Centralise, standardise, then act.
Achilles: Annual Risk and Sustainability Report 2026
2805 organisations across 12 countries responded to this Achilles survey, revealing supplier risk has moved from a specialist concern to a board-level priority.
Regulation has now overtaken carbon-reduction commitments and customer expectations as the dominant driver of sustainability action, with regulatory and legislative pressures accounting for 14.2 percent of reported drivers last year, up from 21.3 percent in 2026.
The regulatory push is reshaping what supplier due diligence means: organisations must evidence it continuously, not periodically.
Supplier-related disruption is widespread and the financial impact varies sharply.
Most respondents report their most significant disruption cost below $500,000, but in pharmaceuticals, a quarter of respondents report disruption costs exceeding $10 million.
In financial services, a meaningful share reports losses between $1 million and $10 million. The tail risk is real and sector-specific.
Thirty-eight percent are exploring or researching AI with no live pilots. Twenty-five percent say it is not a strategic priority.
Data quality appears to be the primary barrier across organisations of all sizes: where supplier information is fragmented across spreadsheets, regional platforms and disconnected systems, AI progress slows.
Organisations with structured risk management platforms show consistently higher maturity, not because their AI is better, but because their data is.
Zip: State of AI in Spend
The finding that leads this report of 1,050 respondents is blunt: only 17 percent of organisations can demonstrate clear, measurable ROI from their AI investments. Zip calls this group the ‘Builders’.
The gulf between Builders and the rest is not about tool selection. It is about depth. Fifty-five percent of Builders deploy AI widely across multiple processes. Among the laggards, it is four percent.
Among Builders, 84 percent use AI multiple times per day as an operational baseline. Among the trailing group, that figure sits at 40 percent.
The conviction follows: 71 percent of Builders are very confident in their ability to put AI to work, against 11 percent of those falling behind.
The shadow tooling problem is significant. Fifty-seven percent of knowledge workers are using AI tools their employer has not sanctioned. Among procurement and operations teams, 60 percent are running unsanctioned tools. Among IT, ostensibly the AI gatekeeper, 64 percent are.
This is not a security problem at this stage. It is an unmet demand signal. The workforce has moved faster than governance.
On productivity: 89 percent of respondents report a net positive productivity effect from AI after accounting for correction time.
Seventy-two percent say their productivity would take at least a moderate hit if their primary AI tool was unavailable for a workday. The dependency is already structural.
The gap is widening fast enough that the organisations still waiting to transform are heading for a difficult outcome.
Gartner: How AI Enables a More Proactive Procurement Strategy
By 2028, organisations that rebalance strategic sourcing towards shorter contract cycles will outperform peers by institutionalising proactive, AI-enabled market testing, according to this report.
Gartner identifies the core problem as market engagement inertia. Only 52 percent of sourcing teams consistently evaluate alternatives before purchase. Forty-three percent conduct broader market research such as peer engagement, while 40 percent run product review analysis.
Forty-six percent waive competitive bidding if the supplier is already on contract, while nearly half of teams become involved only at or after negotiation.
AI breaks the historical argument against frequent tendering. Pre-negotiation processes that once took more than 12 months can now be completed in 20 weeks, and that timeline will keep shrinking.
The cost argument for long-term contracts was always partly about the friction of running events. Remove the friction and the argument dissolves. Sourcing cycle time and contract duration become strategic choices rather than operational necessities.
Fifty percent of CPOs cite resource constraints as a top challenge, while 57 percent face increasing pressure to identify new sources of savings and 63 percent say cumbersome workflows hinder execution.
AI addresses all three simultaneously by reducing tendering effort, expanding market coverage and freeing buyer capacity for the decisions that require judgment.
Proactive procurement, one in which procurement routinely tests the market without waiting for stakeholder initiation, becomes viable for the first time.
Gartner – Capture ROI From Procurement AI: Build Tech Readiness
AI is now the number one topic in procurement, and CPOs are accountable for ROI. This Gartner research note is about building the technology readiness to deliver it.
The Gartner 2026 CPO Survey was conducted among 213 respondents from North America, Western Europe and Asia-Pacific across organisations with annual revenues of at least $500 million.
The C-Suite AI Survey covered 1,303 respondents across multiple functions and regions.
The findings are direct: AI investment is accelerating faster than procurement’s ability to turn it into measurable value.
Gartner distinguishes three AI types that matter for procurement: machine learning, generative AI and agentic AI. Each has different characteristics, different use cases and different maturity levels. Conflating them produces poor procurement decisions and unrealistic expectations.
Gartner recommends a buy-build-blend strategy: buy commercially available solutions, build proprietary agents on enterprise platforms, and blend the two based on the need for differentiation and speed to value.
The most important recommendation is counterintuitive: do not wait for a perfect data foundation before deploying. Build the AI foundation on a use-case basis. Identify specific use cases aligned to business objectives, clean up the necessary data for those cases and start there.
Waiting for enterprise-wide data readiness is delay disguised as due diligence.
The CPO’s job is to build AI literacy within the function, define the buy-build-blend strategy in partnership with the CIO, and then sequence deployment from highest-impact use cases outward.
BCG: How AI in Procurement Drives Competitive Advantage
BCG’s August 2026 procurement paper is the most direct statement of the transformation argument building across this month’s sources.
The case starts with what AI is in most organisations today: a support layer for existing, people-centric workflows. That use case improves procurement. It does not transform it.
BCG’s modelling is specific: redesigned AI-enabled procurement can free up 60 percent of buyer capacity. Cost savings range from eight to 15 percent, OTIF improves five to 15 percentage points and sourcing cycle times fall 30 to 60 percent.
Seventy percent of the effort required to reach those outcomes must go into people, organisation and process redesign. The technology is the smaller part.
BCG’s five-step approach moves from copilots to high-impact workflows, then to end-to-end workflow redesign, foundational data and governance investment, and finally the hybrid human and AI operating model.
The copilot step is not the destination. It is the starting line.
As always, reach out to discuss more. I’m always happy to hear your thoughts!
