Supply Chain Leaders Weigh In on AI, Data Trust, and What Comes After the Pilot at Supply Chain Innovation Forum 2026
Almost every supply chain team has run an AI pilot by now,: a chatbot here, a forecasting experiment there. The more interesting question is what happens after: which of those experiments actually earns a permanent place in daily operations?
That was the focus of one of the breakout sessions at Supply Chain Innovation Forum 2026, led by Rob van Ipenburg, CEO of Quyntess, where a group of supply chain decision makers compared notes on what's working, what's next, and where they're still figuring things out.
The mood was one of momentum, paired with a clear sense of what still needs to be built to support it.
Productivity tools are established; agents are still emerging
Agents still feel somewhat abstract to a lot of users, more discussed than actually used, and there's often a lack of clarity about what these tools are and how they'd apply to daily work.
Data security remains the main constraint
Many organizations restrict external AI tools because of data security and privacy concerns, which limits some of the productivity gains they could otherwise capture. In response, companies are setting up governance policies, fenced-off or on-premise AI environments, and contractual agreements with providers to ensure data isn't used to train external models.
Data classification projects are underway in some organizations to determine what information can safely be shared with AI tools. One company has completed this kind of classification project specifically for tools like Copilot, and another has an internal AI department that curates approved models and sets usage boundaries.
Where AI agents could fit into existing platforms
One idea gaining traction is building AI capabilities directly into the supply chain platform itself: chat functions, pre-built agents for specific tasks, and the option for users to bring their own custom agents. The approach involves a secure environment (Amazon Bedrock was named specifically) that compartmentalizes customer data, connected through a standardized component called the "Model Context Protocol" (MCP), which lets both internal and customer-built agents interact with platform data while respecting user roles and permissions.
This had strong support, with particular interest in the analytics potential. Business intelligence dashboards are often too static for the ad hoc, exploratory analysis that root cause analysis requires. The technology is being released now. One constraint worth noting: the processing time required for agents to complete complex tasks still needs to improve.
Cross-company data sharing: multiple ideas on the table
One proposal explored companies opting in to share anonymized performance data to spot broader trends, such as a supplier's delays affecting multiple customers, and benchmark performance without revealing sensitive details like pricing. Two mechanisms came up: a badge system, where a supplier earns a performance score from aggregated data and can choose whether to publish it, and a trend comparison, where a company sees its own forecast trends plotted against the wider community.
Views on this varied. Some participants saw real value in shared visibility, pointing out that pooling this kind of data could help the wider market catch patterns that are hard to spot from inside a single company, like a supplier's performance dipping across several customers at once. Others favored a more cautious approach, weighing questions about data quality and representativeness against the benefits. Since companies legally own their data, any participation would stay strictly opt-in, and a clear incentive, such as access to supply chain finance programs for high performing suppliers, would likely help bring more companies on board.
What happens next
An MCP connector is being released so companies can connect their own AI agents to platform data, and more pre-built agents are in development for tasks like carrier selection and exception handling. The clearest recommendations to come out of this: establish AI governance policies and data classification rules early, look at integrated AI agents for analytics and process automation rather than just chat tools, and consider opt-in, anonymized data-sharing initiatives where clear rules and mutual benefit exist.
Three questions worth bringing back to your team
If any of these don't have a clear answer yet, Quyntess can help. Depending on where your organization stands, that might look like: