Enterprise AI adoption is a paradox. Adoption is everywhere; impact isn’t. According to McKinsey’s State of AI 2025, 88% of organizations use AI in at least one function, but 80%+ report no measurable EBIT impact. This guide unpacks why — and where supply chain leaders are actually winning. For the full data, see our Industry Stats hub.
The short answer
- 88% of organizations use AI somewhere
- 71% use generative AI in at least one function
- 23% are scaling agentic AI systems; another 39% are experimenting
- 80%+ see no measurable EBIT impact yet
- 70% of large organizations will adopt AI-based supply chain forecasting by 2030 (Gartner)
Why the EBIT gap exists
The pattern McKinsey identifies is clear: workflow redesign is the differentiator. Companies that bolt AI onto existing processes get small productivity bumps that disappear in the noise. Companies that redesign the underlying workflow around what AI can do see EBIT impact at the line-item level.
In supply chain, that distinction looks like this:
- Bolted on: Demand planner runs the same monthly forecast process, but uses an AI tool to draft a summary. Net change to forecast accuracy: ~0%.
- Redesigned: Daily AI-driven forecast cycle replaces the monthly meeting. Planners now spend time on exceptions, not baseline generation. Forecast accuracy improves 15-25%; planner productivity 2-3x.
Where AI is actually working in supply chain
Demand forecasting (especially new product introductions)
For SKUs with limited history, ML models that use analogs, attributes, and external signals materially outperform statistical methods. This is where most planners see AI deliver value within a quarter, not a year. The Geneva Forecast API handles the algorithm selection automatically — one call returns the best of 45+ models for each SKU.
Inventory optimization
Dynamic safety stock and reorder point optimization — driven by lead-time variance, demand volatility, and service-level targets — typically free up 10-20% of working capital while holding service levels constant.
Transportation and routing
AI-powered route optimization continues to deliver 5-15% on transportation cost. The math is well-understood; the variable is integration quality with TMS and order systems.
Supplier risk monitoring
News, financial filings, weather, and geopolitical signals can be ingested in near-real-time and scored for supplier disruption risk. This is where generative AI is starting to add specific value — summarizing supplier risk narratives at scale.
Document and exception handling
POs, invoices, ASNs, customs documents — generative AI now reliably extracts, validates, and routes these for exception handling.
The new pattern: LLM agents calling production forecasting
The most important shift in 2026 isn’t generative AI for executive summaries — it’s LLM agents that can call real production tools. This is what makes “AI for supply chain” actually work instead of theatrical.
Geneva Forecast exposes a Model Context Protocol (MCP) server. That means Claude, GPT, or any agent that speaks MCP can call Geneva natively — generating production-grade forecasts from natural language, without custom integration code. A planner can say “forecast my top 200 SKUs for the next 13 weeks, using the last two years of weekly history and the holiday calendar,” and the agent picks the right model, fits it, and returns the result with confidence intervals.
This is the workflow redesign McKinsey is talking about. The planner stops being a model-fitter and starts being an exception handler. Productivity goes up 2-3x; forecast accuracy goes up 15-25%.
See Geneva Forecast API + MCP →
Where AI is overhyped in supply chain
- End-to-end “autonomous planning.” Most demonstrations are scripted. The exception handling that actually happens in S&OP is not yet automatable.
- Causal reasoning for promotions. AI models can correlate, but rarely identify what actually caused a lift. Human + AI still beats AI alone.
- Generative AI for executive summaries. Looks impressive, generates noise. Most planners trust their own numbers more.
What to do in 2026
- Pick one workflow to redesign, not bolt onto. Demand sensing or supplier risk are the highest-ROI candidates.
- Set a measurable success metric upfront. Forecast MAPE, working capital, on-time-in-full — not “productivity.”
- Resist the platform pitch. A focused tool that solves one problem in 8 weeks beats an enterprise suite that promises everything in 18 months. The Geneva Forecast API is one such focused tool — production forecasting in a single API call.
- Build a feedback loop. AI without a learning loop just locks in last quarter’s mistakes.
Frequently asked questions
What percentage of organizations use AI in 2026?
McKinsey’s State of AI 2025 found 88% of organizations use AI in at least one business function, up 10 points YoY. 71% regularly use generative AI in at least one function.
Is enterprise AI delivering ROI?
For most, not yet. 80%+ of organizations see no measurable enterprise-level EBIT impact from generative AI. Workflow redesign is the differentiator between the 20% that do see impact and the 80% that don’t.
Where is AI working best in supply chain?
Demand forecasting (especially NPI), inventory optimization, transportation routing, supplier risk monitoring, and anomaly detection in PO/invoice data.
How fast will AI adoption grow in supply chain?
Gartner projects 70% of large organizations will adopt AI-based supply chain forecasting by 2030. Adoption is accelerating, but value capture lags adoption by 2-3 years on average.
How do LLM agents access forecasting?
Geneva Forecast exposes an MCP server. Claude, GPT, or any agent that speaks MCP can call Geneva natively to generate production-grade forecasts from natural language inputs — letting AI agents handle demand planning workflows end-to-end without custom integration.
See the full set of AI adoption and supply chain benchmarks — refreshed weekly from verified primary sources — at RoadMap Industry Stats.