Every survey on supply chain AI lands on the same blockers: fragmented data and difficult integrations. What they say less often is where a large share of that data actually sits: in your suppliers' and carriers' systems, on the other side of a boundary you don't control. That boundary is also where risk management quietly stops working.
The bottleneck has moved
A couple of years ago, choosing the right technology was one of the bigger concerns for supply chain leaders. Today, integration is much higher on the list.
Oliver Wyman and Prequel Ventures surveyed 100 supply chain leaders across Europe, all at companies with more than $100 million in revenue, and compared the results with the previous survey wave.
Top challenges when implementing new supply chain technology
Challenge | Prior survey | Most recent survey |
|---|---|---|
Integration | 47% | 80% |
Data quality | 21% | 63% |
Resistance to change | 56% | 39% |
Lack of skilled workers | 50% | 45% |
The same shift showed up earlier in the decision process. Vendor availability fell from 60% to 26% as a top concern, while the ability to integrate the technology rose from 19% to 66%.
That makes sense. Evaluating an AI tool on paper is one thing. Connecting it to the systems, data and partners your operation actually depends on is another.
A lot of the data isn't yours to fix
Inside one company, fragmented data is already difficult enough. Order data might sit in the ERP, inventory in the WMS, movements in the TMS and product specifications somewhere else entirely.
But supply chain data does not stop at the company boundary.
It continues into supplier systems, carrier platforms, customs filings, freight invoices, EDI messages and customer portals. Some of those systems are modern. Some aren't. Some partners will give you structured data through an API, while others may still send a spreadsheet or document.
You can run a master data project across systems you own. You can assign someone to it, set standards and put a budget behind it.
You cannot do the same with a supplier's ERP or a carrier's EDI configuration.
That does not mean the data has to stay disconnected. It means the solution cannot depend on every external partner changing the way they work. You need to be able to connect, normalize and interpret information in the form it actually arrives.
And in the form it arrives next quarter. Partner data does not only differ from yours, it drifts. A supplier upgrades their ERP. A carrier changes a message format. Someone adds a field and nobody tells you. When that happens quietly, it is worse than a connection that breaks outright, because the data keeps flowing and stops meaning what it used to. Nobody gets an alert. The numbers just start being wrong.
Gartner pointed to the same underlying issue in a survey of 140 senior supply chain leaders, published in May 2026. Among the constraints holding back AI-powered orchestration, it highlighted inconsistent partner data, with information coming from trading partners often incomplete or unreliable.
This is one reason integration becomes more important as AI ambitions grow.
Why pilots are easier than rollout
Among European supply chain organizations already using or experimenting with AI, only around 15% reported having fully rolled anything out. Most initiatives were still sitting in proof of concept or pilot stages.
When respondents were asked what was getting in the way, the answers were fairly practical:
Poor or fragmented data: 67%
Legacy integration: 61%
Use-case gaps: 35%
Talent shortage: 33%
ROI and scaling hurdles: 30%
Strategy gaps: 22%
Further down the same list, regulatory and compliance risk came in at 11% and model risk at 10%. The blockers are not exotic. They are plumbing.
A pilot can be built around a clean subset of data. You can choose the process, clean up the inputs and keep the number of systems involved relatively small.
The real operation is different. Procurement, planning, manufacturing and logistics all have their own systems and definitions, and suppliers, carriers and other partners add another layer on top.
But the part a lab-grade pilot cannot show you is the long tail. The pilot runs on a handful of suppliers, a few hundred SKUs and one clearly defined use case. The real operation has hundreds of suppliers, most of them low volume and none of them standardized. Thousands of SKUs. Exceptions that only surface a few times a year. None of it is visible when you scope the pilot, and all of it arrives during rollout.
That is usually where the integration problem shows up, and usually later than anyone planned for.
Get the order right
There is another constraint here: companies still expect these investments to save money.
Every respondent in the European survey rated cost reduction as either an important or very important objective for supply chain technology investment. So while AI budgets may be growing, there is still pressure to show that the investment produces something useful.
There is a trap on that cost side worth naming. Building integrations in house often looks cheaper than paying an integrator, on the assumption that you do the work once and use it forever. In practice, you have not integrated to a partner's system. You have integrated to the current version of their API. Versions change, fields get deprecated, portals get replaced, and each of those events lands back on your own IT team. The savings from building it yourself get spent maintaining it.
Oliver Wyman's recommendation is to establish data ownership and governance before trying to scale automation and AI. That means agreeing on who owns core supply chain data, what good data looks like and what standards need to apply across planning, procurement, manufacturing and logistics.
But internal governance only takes you so far.
A supply chain architecture also has to deal with information coming from outside the organization, where formats, definitions and data quality will never be completely under your control, and where they will not hold still either.
Otherwise, you can end up adding another AI tool on top of the same fragmented environment you were trying to fix.
The question to ask first
Before starting the next AI initiative, there is a more useful question than which model to use or which vendor to shortlist:
Does an order line mean the same thing in all five systems that hold a piece of it, including the two systems you don't own?
If the answer is no, that is the project.
Everything else is downstream of it.
Sources
Oliver Wyman and Prequel Ventures, Supply Chain Tech Report 2026 (February 2026). Survey of 100 supply chain leaders across Europe at companies with more than $100 million in revenue, conducted at the end of 2025. Year-on-year comparisons are against the equivalent 2024 survey wave.
Gartner, Gartner Survey Shows AI is Not Driving Supply Chain Operating Model Transformation (6 May 2026). Survey of 140 senior supply chain leaders, conducted November 2025.
