AI is being used in forecasting and planning for demand estimates, inventory decisions, and supply-chain coordination. But a deployment is not proof of business impact: the available examples range from a company describing its planning workflow to a vendor reporting a performance gain, and those are different kinds of evidence.
How organizations use AI in forecasting and planning
Forecasting turns historical and current data into estimates of future demand or other operating conditions. Planning uses those estimates to choose actions, such as how much inventory to hold, how to align supply with demand, or which scenarios to prepare for. AI can contribute forecasts or decision-support signals within that process; it does not necessarily make the final operating decision.
As an Amazon Associate I earn from qualifying purchases.
Two documented examples illustrate the range: Ericsson describes machine learning as part of a broader integrated business-planning environment, while Google Cloud describes a demand-forecasting model deployed by retailer OTTO. Neither example should be treated as representative of all AI planning systems.
What the documented deployments show
Ericsson: machine learning inside integrated business planning
In an interview with the Infosys Knowledge Institute, Ericsson leaders described an integrated business-planning transformation connecting demand, supply, inventory, and sales and operations planning (S&OP). The company developed an in-house machine-learning forecasting solution and incorporated it into its planning environment alongside SAP IBP. Forecasters could compare internal and SAP forecast inputs, while business users selected scenarios. Infosys Knowledge Institute’s account of Ericsson’s transformation describes the workflow but does not report a quantified forecasting-accuracy improvement.
#1 Best Overall
Diego Moreno, Ericsson’s head of supply-chain planning, described machine learning as an evolving capability rather than a replacement for the existing process: “My personal expectation is that we will take the lessons learned while we and how we did in our in-house development, but we are not there yet to use machine learning as to replace this one.” The wording reflects his expectation in that interview, not a general conclusion about when AI can replace human planning.
OTTO: retail demand forecasting
Google Cloud says German ecommerce retailer OTTO implemented its Time-series Dense Encoder (TiDE) model on Vertex AI and Google Kubernetes Engine for demand forecasting. Google reports an improvement of up to 30% in forecasting accuracy, along with lower inventory costs and better product availability. This is a vendor-reported case result: the cited page does not establish the measurement period or an independent evaluation. It should not be directly compared with Ericsson’s qualitative account. Google Cloud’s collection of organizational AI use cases describes the case.
What independent field research adds
A 2026 study in the Journal of Operations Management analyzed approximately 575,000 algorithm forecasts, user adjustments, and sales observations across 91 SKUs in three general-merchandise categories and 485 store locations over 84 weeks. The authors examined how people adjusted algorithm-supported demand plans. This is evidence about forecast use and human adjustment behavior in the studied setting, not a universal benchmark for AI accuracy or a direct comparison with either company deployment.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteThe study also notes an identification limitation: it did not directly identify individual users, using SKU-store groups as a proxy for individual forecaster responsibility. Its scale provides a useful view of field activity, but the limitation matters when interpreting who made particular adjustments or why. See the study, “Is AI an Algorithm by Any Other Name? Behavioral Reactions to AI- and Model-Based Demand Planning Algorithms”.
How to judge a reported AI deployment
When evaluating a case study, separate the fact that a system was deployed from the claim that it improved results. Look for the details that make a result interpretable:
- Task and scope: Is the system forecasting demand, informing inventory, or supporting a broader planning process? Is it deployed across a business or described as one part of a workflow?
- Model and data: Does the source identify the model, platform, or forecast inputs? Those details help define what was actually implemented.
- Human role: Can planners review inputs, adjust forecasts, or choose scenarios? Ericsson’s account explicitly describes comparison and scenario selection.
- Outcome and measurement: What metric changed, over what period, and against what baseline? A percentage without a period or evaluation method is difficult to interpret.
- Source of the claim: Is the result reported by the organization operating the system, a technology vendor, or an independent study? Attribute claims accordingly; these evidence types are not interchangeable.
On that basis, Ericsson’s example establishes a described deployment and human workflow, not a measured accuracy gain. Google Cloud’s OTTO case supplies a vendor-reported accuracy figure but not the measurement period or independent validation on the cited page. The field study supplies substantial observational scale and caveats, but it is not a head-to-head evaluation of those deployments.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the “11 deployments” count can and cannot establish
The AI Weekly category page for “AI in forecasting & planning: real deployments” showed six entries when opened, rather than eleven. Because it is a changing directory, that observed count is only a snapshot, not a fixed portfolio total. The directory alone therefore does not substantiate eleven separately named deployments. AI Weekly’s forecasting and planning category can be checked for its current listings.
The examples and study above answer how AI is being used and what kinds of evidence are available; they do not form a verified set of eleven cases. Across them, there is no common outcome metric or independent, like-for-like comparison that would support ranking the deployments by impact.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
