AI Demand Forecasting: Fix the Planning Basics First

By /

A better forecast is on almost every manufacturer’s wish list. Demand swings, supplier lead times keep stretching, and the planning spreadsheet that worked five years ago now needs constant patching. AI demand forecasting sounds like the fix.

It can help. But the latest research makes a point worth hearing before you buy anything: the limiting factor is rarely the algorithm. It is the data, the process and the people around it.

Big budgets have not bought autonomy

On September 24 Gartner predicted that by 2030 only 5% of organizations using some form of planning automation will make at least 10% of their planning decisions autonomously. The release notes that 83% of organizations in a Gartner survey had already spent at least $3 million on planning automation, including AI. That survey covered companies with at least $500 million in revenue.

Gartner’s analyst put it plainly: spending on automation does not by itself create AI readiness. The firm points to data quality, decision ownership, workforce skills and technology architecture as the things to fix first.

If large companies with multi-million-dollar budgets are moving slowly, a smaller manufacturer should not expect a tool to transform planning in one quarter. The good news is that the groundwork is mostly free.

People and data are the real bottleneck

A Gartner article from October 2 says only 17% of organizations have deployed AI at scale. It names human adoption as the main limit. More than half of organizations in Gartner’s 2025 survey reported difficulty fitting AI into existing operations, because they were layering it onto workflows never designed for it.

Middle-market manufacturers report the same thing. In RSM’s 2026 survey of 129 manufacturers, published July 21, 32% named data quality, availability and lineage as a barrier to wider AI use. Among those whose pilots had only moderate or limited success, 51% blamed data quality, 43% integration problems and 30% resistance from staff. And 84% agreed that leadership is more enthusiastic about AI than employees are.

Why forecasting is harder right now

Volatile demand makes forecast quality matter more. ISM chair Susan Spence said new orders have been moving in an “up-down pattern” since June, Manufacturing Dive reported. Supplier deliveries have slowed for ten straight months. When both demand and supply are uncertain, a weak forecast costs you twice: in stockouts on one side and excess stock on the other.

That is exactly when it is tempting to buy a model. It is also when a model trained on messy history will struggle most.

A readiness checklist before you buy

Before you evaluate any AI forecasting tool, check whether you can answer yes to most of these:

  • You have at least two to three years of order history by item and customer, in one place.
  • Stockouts, one-time orders, promotions and price changes are flagged, so the model does not learn from distorted history.
  • Your item master is reliable: units of measure, lead times, minimum order quantities and active or obsolete status.
  • You measure forecast accuracy and bias today, even roughly, so you have a baseline to beat.
  • Someone owns the forecast and the decisions that follow from it.
  • There is a regular planning meeting where sales, operations and purchasing agree one set of numbers.

If several of those are a no, start there. Those fixes improve your planning even if you never buy an AI tool.

Start with decisions, not software

Gartner’s advice is to begin with the planning decision itself. Classify decisions as strategic, tactical or operational. Decide whether AI should support, augment or automate each one, based on value, complexity and risk. Then measure whether decisions actually improve, not whether the technology was deployed.

For a mid-size plant, that might look like this:

  • Let AI suggest a baseline forecast for high-volume, steady items, with planners reviewing exceptions.
  • Keep humans in charge of new products, big customer orders and anything driven by a known one-off event.
  • Use the time saved to talk to customers and suppliers, which no model can do for you.

What this means for your business

  1. Measure your current forecast accuracy and bias for your top 50 items. Without a baseline, you cannot judge any tool.
  2. Clean and flag your order history: stockouts, one-offs and promotions.
  3. Fix the item master, especially lead times, which feed directly into reorder points.
  4. Review safety stock and order quantities with current data, using our free safety stock calculator and EOQ calculator.
  5. Pilot AI on one product family with a clear target, such as lower forecast error or fewer expedites, and a planner who owns the result.

If you want help getting your planning data and process ready for AI, our AI support and operations services start with a free 30-minute call, and we agree the scope before any work starts.

Sources

This article summarizes public reporting and research as of October 5, 2026. It is general information, not legal, customs or tax advice.

Scroll to Top