Data quality has become one of the most reliable blockers in supply chain transformation. It comes up in almost every conversation about AI, automation, planning, visibility and orchestration. The product data is inconsistent. The supplier data is incomplete. Inventory visibility is patchy. Customer signals are fragmented. Historical data is distorted by disruption. External data is hard to integrate. Master data is not trusted.
All of that is real.
Poor data can distort planning, weaken service, increase inventory, slow automation and undermine confidence in every digital tool that sits on top of it. No serious supply chain leader should dismiss the importance of getting the foundations right.
But there is another side to the problem. Data can also become the reason not to move.
The promise of a perfect data foundation can turn into a holding pattern. Teams wait for the ERP programme to finish, for master data to be cleaned, for integrations to be rebuilt, for every exception to be resolved and for the model to become safe enough to trust.
In the meantime, the business still has to make decisions. Demand keeps moving. Lead times keep shifting. Suppliers miss commitments. Inventory sits in the wrong place. Customers expect answers faster than the organisation can provide them.
The question is not whether data quality matters. It does. The better question is which data problems genuinely prevent better decisions, and which ones have become an excuse for delaying progress.
Imperfect data does not always mean inaction
A useful example comes from the UK pork supply chain.
Sainsbury’s and Cranswick worked with Singular Intelligence to use AI to improve supply and demand sensing in a category under pressure from inflation, labour shortages, global disruption and uncertain demand. The challenge was not simply to create a cleaner data set. It was to improve planning and decision-making in a complex, volatile supply chain where poor decisions were leading to revenue losses, food waste and oversupply.
The solution combined available demand and supply signals. On the demand side, this included sales, availability and weather. On the supply side, it included cost risk, production risk and supply shock analysis. The reported outcomes included a 2 to 5% improvement in product availability, a 50 to 70% reduction in food waste for Sainsbury’s, and a 15.25% reduction in the oversupply of pork sent to the retailer.
The lesson is not that data quality can be ignored.
It is that imperfect information can still support better decisions when the business is clear about the decision it is trying to improve. That distinction matters. Too many AI and data programmes start with the abstract ambition to “fix the data” or “create a single source of truth”. Those goals may be valid, but they are often too broad to create momentum. A decision-led approach starts somewhere more practical.
Which decision is currently too slow, too manual, too reactive or too expensive when wrong? In the Sainsbury’s and Cranswick example, the decision was tied to supply, demand, availability, waste and oversupply. That gave the data work a commercial and operational purpose.
The real test is risk
Not all imperfect data is equal.
Some data problems create unacceptable operational, financial or compliance risk. A manufacturer cannot automate production decisions on unreliable quality data. A pharma supply chain cannot bypass validation because a model looks promising. A planning team cannot trust recommendations if the inventory position is materially wrong. A procurement team cannot make supplier risk decisions on records that do not reflect the actual supplier base.
In those cases, the data is not an excuse. It is the barrier. But other gaps can be managed. Teams may not have perfect demand visibility, but they may have enough signals to improve the current forecast. They may not have a complete view of customer behaviour, but they may have sell-in data, retailer inputs, availability signals, promotions, weather, market indicators or external demand proxies.
They may not be able to automate the decision fully, but they may be able to use AI to narrow the options, surface exceptions or recommend actions for human review.
That is where supply chain leaders need more nuance. The question should not be, “Is our data ready for AI?” It should be, “For this specific decision, what level of data quality is good enough to improve the outcome, and where do we still need human validation or stronger governance?”
Uncertainty is permanent
Amazon’s public explanation of its inventory planning evolution makes a similar point from a different angle. Amazon Science notes that customer demand cannot be perfectly predicted, even with advanced machine learning, and that vendor lead times vary because of manufacturing capacity, transport and weather. In other words, uncertainty is not a temporary data problem that disappears once the system is mature. It is part of the operating environment.
That matters because many supply chain technology conversations imply that the goal is to eliminate uncertainty. In reality, the more useful goal is to build systems and processes that make better decisions under uncertainty.
This is where the distinction between forecast accuracy and decision accuracy becomes important. A better forecast is valuable, but it is not the end point. The business still needs to decide where inventory should sit, how much risk to carry, which customers to prioritise, when to expedite, when to switch supply, and how to balance service, cash, cost and margin.
Better data helps. But the organisation also needs better decision rules, clearer escalation points, faster scenario modelling and a stronger connection between planning, logistics, procurement, finance and commercial teams.
The danger of waiting for perfection
There is a legitimate fear behind many data objections. Supply chain leaders know that bad data can lead to bad decisions. They have seen systems recommend actions that planners do not trust. They have seen automation fail because the underlying records were wrong. They have seen dashboards multiply without changing the operating rhythm of the business.
So the caution is understandable. But caution becomes a problem when it freezes progress. If the current process is manual, slow, fragmented and inconsistent, then waiting for perfect data may not be the low-risk option. It may simply preserve a decision process that is already failing.
That is why the best starting point is often not a full enterprise data transformation. It is a specific decision where the business can compare the current approach with an improved one.
Can AI help planners identify exceptions earlier?
Can external signals improve replenishment decisions?
Can better availability data reduce waste?
Can supplier risk signals improve escalation?
Can machine learning help position inventory more effectively?
Can a decision layer sit above legacy systems and recommend actions without replacing the system of record?
These are not excuses to avoid fixing the data foundation. They are ways to create value while the foundation improves.
What supply chain leaders should do next?
The most practical route is to separate data work into three categories.
First, identify the data problems that genuinely block progress. These are the issues that create material risk, destroy trust or make automation unsafe.
Second, identify the data improvements that would unlock the most operational value. Not every master data issue deserves the same priority. The most important data is the data tied to decisions that affect service, inventory, working capital, supplier performance, cost-to-serve and customer reliability.
Third, identify the decisions where imperfect data is good enough to start. These are areas where recommendations can be tested, validated, back-tested or kept within human approval before being scaled.
That approach changes the conversation. Data is no longer a generic barrier sitting in front of transformation. It becomes part of a decision architecture.
The supply chain teams that make the most progress will not be the ones that pretend data quality does not matter. They will be the ones that know where it matters most. Because in a volatile supply chain, the choice is rarely between perfect data and bad data.
More often, it is between acting with imperfect but improving signals, or continuing to make slow decisions with manual workarounds, partial visibility and delayed escalation.
That is the real test. Data should be good enough to improve the decision, trusted enough to support the level of automation, and governed enough to keep the business in control. Anything beyond that may be necessary over time. But it should not become the reason the supply chain stands still.
At the SupplyChain360 Summit, we will be exploring this directly in a session titled When Data Is the Barrier, and When It’s the Excuse.
The session will examine where poor data genuinely creates risk, where targeted improvements can unlock the greatest operational value, and how supply chain teams can move forward with imperfect data while maintaining trust and control.
Join us at The Belfry, Sutton Coldfield, on 3–4 March 2027 to be part of the discussion and hear practical insights from senior supply chain leaders shaping the next operating model for enterprise supply chains.
