Forecasting remains one of the most difficult parts of supply chain planning. When forecasts are consistently wrong, the impact quickly shows up in margin, inventory, and working capital. Yet improving forecasting accuracy has become more attainable in recent years.
Two developments are particularly important: there is now more usable historical supply chain data available, and AI has significantly reduced the effort required to build and improve forecasting models. Both depend on the same foundation: a continuous and reliable record of what has happened in the past.
Getting forecasting wrong is costly. Research estimates that inventory distortion, the cost of both missing and unwanted inventory, is about 6% of retail sales, with roughly two-thirds of this impact coming from empty shelves and the remainder from surplus stock [1]. The split reflects how differently the two failures present themselves. An empty shelf is immediately visible: to the customer standing in front of it, to the store manager, and to the account team fielding the complaint. Excess stock is far less confronting; it sits quietly in a warehouse and surfaces later as markdowns, obsolescence, and tied-up working capital.
Better forecasting can address both sides of this equation. More accurate demand signals can reduce missed sales while also enabling a leaner inventory position. And with recent advances in data and AI, building the capabilities to achieve this is becoming increasingly feasible.
COVID is finally becoming history for forecasting
The COVID years distorted supply-chain data as thoroughly as they distorted our day-to-day operations. Demand patterns changed abruptly, supply became unpredictable and many relationships between variables temporarily broke down.
That effect is now gradually receding. In many supply chains customer orders have turned back to normal patterns, with more consistent seasonal patterns and long-term trends. This matters because forecasting methods need several comparable years of data to distinguish recurring patterns from isolated events.
The longer and cleaner the historical record becomes, the better models can identify what is genuinely predictable and what was simply an anomaly.
AI has lowered the barrier to better forecasting
At the same time, AI is making sophisticated forecasting capabilities more accessible. The most well-known recent advances in AI have taken place in the area of large language models, which are not directly designed for forecasting. Their indirect impact on forecasting, however, is still significant. In particular, AI is lowering the resources required to develop and improve forecasting capabilities in two ways:
- Selecting and developing the right approach
AI models can help identify the most relevant error metrics based on the specific forecasting problem and suggest appropriate models given the available data.
- Reducing development effort
AI-assisted coding has significantly reduced the effort required to build, test and iterate tailored analytical models. What previously required a dedicated permanent forecasting team can now be developed with a much smaller investment of time and resources, making more advanced forecasting accessible to organizations that may not have the scale to maintain a large specialist team.
Data quality determines what is possible
The value of any forecasting model ultimately depends on the quality and continuity of the data behind it. Forecasting is inherently time-bound. A model learns from what has happened before and uses those patterns to anticipate what comes next.
This is what makes gaps in historical data particularly damaging. A system bug that silently misrecords transactions for a quarter, or a migration in which historical records were not carried across, will each punch a hole in the period a model depends on. Even a relatively long dataset can therefore lose much of its value if important periods are unreliable.
Companies should not treat data quality as a technical detail. Those that spend resources on a resilient data warehouse are seeing the returns now that analytical possibilities are expanding. As everyone climbs aboard the AI train, it is worth remembering that good data is the track it runs on.
Better forecasting starts with better foundations
The opportunity is clear. AI is lowering the barriers to developing and deploying forecasting capabilities, while supply chain data is becoming increasingly useful as the disruptions of recent years move further into the past.
Companies investing in reliable data infrastructure, consistent data retention, and clear ownership are building the foundation for better forecasting tomorrow. At Argon & Co we help organizations turn supply chain data into better forecasting and planning decisions. By assessing data quality, forecasting, and the underlying planning processes, we help clients identify where AI can create value and what foundations need to be in place to make that value sustainable.
Sources: [1] https://www.ihlservices.com/product/inventory-distortion-study-2026/
Share:
