Resilient Supply Chain Podcast: AI Exposes the Human Workarounds Behind Data Reliability
As supply chains adopt AI agents and more autonomous decision-making, a long-standing weakness in enterprise data is becoming harder to ignore: many systems have worked because people quietly supplied the context, judgement and restraint that the data itself lacked. In episode 144 of the Resilient Supply Chain Podcast, host Tom Raftery speaks with Sagar Paul, Chief Revenue and Growth Officer at The Modern Data Company, about what happens when those human corrections are removed and machines begin acting at speed and scale. For supply chain leaders, the issue is less about whether AI can automate decisions than whether the underlying data is trustworthy enough to support them. The full episode is available at www.resilientsupplychainpodcast.com
Humans Were the Hidden Control Layer
One of the central arguments is that enterprise systems were largely built for human judgement. People learned which fields were unreliable, which exceptions mattered and when apparently valid data needed interpretation before action.
Raftery characterises this as humans acting as an “error correction layer” across inconsistent systems. Paul agrees, arguing that people were applying both operational judgement and ethical restraint in ways that automated agents do not inherit automatically.
That distinction matters because AI changes the consequence of poor data. A bad dashboard may mislead one person. An autonomous agent can act repeatedly, across thousands of products or transactions, before a human notices the problem.
Semantic Ambiguity Becomes Operational Risk
The episode highlights a particularly difficult problem: data can be technically correct while still being operationally ambiguous. Paul uses the example of “order delivery date”. For finance, the field may relate to revenue recognition. For warehousing, it may refer to when a truck leaves the gate or an expected receipt date. Across an enterprise, Paul says AI could encounter 14 different interpretations of the same field, with the real meaning sitting in tribal knowledge rather than the underlying system.
The implication is significant. Standardisation of format is not enough. Data also needs business context, lineage, governance and a clear relationship to the decision it is intended to support.
More Data Does Not Automatically Mean Better Decisions
The conversation also challenges the assumption that real-time data is inherently valuable.
Paul argues that real-time information only becomes useful when organisations know which signals demand action. He uses a telecom example where operators face so many alerts from connected equipment that some are effectively ignored because teams cannot determine which ones deserve intervention. He describes the resulting flood as “digital exhaust”. For supply chain and operations leaders, the lesson is that speed without prioritisation can increase cost rather than improve control. More data can simply create more noise.
Procurement Shows the Commercial Consequence
The governance problem becomes especially concrete in procurement.
Paul describes large organisations negotiating complex supplier contracts, rebates and discount structures that appear highly attractive financially. Yet during execution, purchasing teams may have no practical mechanism to confirm that negotiated terms are being applied to individual purchase orders.
The result is leakage between negotiated value and realised value. Raftery pushes the issue further by asking who owns the conversion of contract terms into buying decisions: procurement, finance or the data team. Where responsibility falls between functions, automation may simply carry the same leakage forward.
Governance Must Follow the Decision
The broader strategic point is that data quality cannot be treated as an abstract technical standard. Paul argues that data should be judged according to who will consume it, what rights they have, what decision it will enable and whether it needs to be fresh, explainable or auditable.
That becomes more important as routine decisions move from monthly or periodic cycles towards daily or hourly automation. The faster organisations remove humans from the loop, the more important it becomes to examine the data infrastructure and governance underneath those decisions. For supply chain leaders, AI readiness therefore depends less on acquiring more sophisticated models than on exposing the assumptions and workarounds embedded in existing operations. Resilient automation will require clearer ownership, stronger data integrity and explicit governance around decisions that were previously protected by experience, intuition and human intervention.
