In operations, AI earns its keep in three places: forecasting demand more accurately, handling the paperwork around orders, shipments and invoices, and answering questions about stock and orders from live data. Done right, it means fewer stock-outs, less cash tied up in inventory and fewer hours spent chasing documents. Done wrong, it means an expensive model sitting on messy data that nobody trusts. The difference is almost always data quality and human approval, not the algorithm. Below we compare traditional operations with AI used wrong and AI used right, and show where to start.
What can AI realistically do in operations?
The evidence is strongest in supply chains. McKinsey reported that early adopters of AI in supply-chain management improved logistics costs by 15%, inventory levels by 35% and service levels by 65%. In separate research on AI-driven forecasting, McKinsey found that applying AI to supply chain forecasting can reduce errors by 20% to 50%, which can cut lost sales and out-of-stock products by up to 65%, warehousing costs by 5% to 10% and administration costs by 25% to 40%.
Large companies are moving in this direction. Gartner predicts that 70% of large organisations will adopt AI-based forecasting to predict future demand by 2030, and in McKinsey's State of AI 2026, cost savings from AI are most often reported in supply chain and service operations.
How do traditional operations compare with AI-assisted operations?
| Operations task | Traditional way | AI used wrong | AI used right |
|---|---|---|---|
| Demand forecasting | Last year's sales plus a guess | A model trained on incomplete or wrong stock data | A model on clean history and known events and planners adjust it |
| Reordering stock | Someone notices shelves are empty | Automatic purchase orders with no limits | AI drafts purchase orders from forecasts and a person approves them |
| Order tracking | Calling the warehouse or the courier | A chatbot guessing order status | Staff and customers ask and get answers from live order data |
| Documents | Typing invoices and shipping papers by hand | Trusting extracted data without checks | AI reads and fills documents and flags anything unusual for review |
| Delays and exceptions | Found when the customer complains | Alerts nobody owns | AI flags late orders early and routes them to a named owner |
| Operations reporting | A weekly spreadsheet built by hand | Summaries from a public tool fed with exports | Managers ask questions on live ERP data within their permissions |
How is AI forecasting different from traditional forecasting?
Traditional forecasting usually takes past sales, applies an average or a growth rate and adds the planner's experience. It works when demand is stable. It struggles with promotions, new products, seasonal peaks such as Ramadan and back-to-school, and sudden changes in price or supply.
AI forecasting models can take many more signals into account at once, such as each product's history, prices, promotions, seasons and holidays, and learn how they interact. They do not replace the planner. A good process lets the model produce the baseline forecast and lets planners adjust it for what the model cannot know, such as a new contract or a supplier problem.
Why do AI projects in operations often disappoint?
- The data is not ready. Gartner predicts that through 2026 organisations will abandon 60% of AI projects that are not supported by AI-ready data, and that 63% of organisations either lack the right data management practices for AI or are unsure. Wrong stock counts and duplicate item codes break any model.
- Nobody defined the return. In a 2026 Gartner survey, 55% of chief supply chain officers said they were unclear on the returns from their AI investments, even though 67% of supply chain digital investment now goes to AI.
- Jumping straight to autonomy. Fully automatic planning is further away than vendors suggest. Gartner predicts that by 2030 only 5% of organisations automating supply chain planning will make at least 10% of their planning decisions autonomously.
Where should you start with AI in operations?
- Documents first. Purchase orders, delivery notes, supplier invoices and shipping papers are repetitive and easy to check. McKinsey experts estimate that generative AI can cut the time to produce shipping documentation by up to 60%. For invoices and e-invoicing in Egypt and Saudi Arabia, see our guide to AI for finance and operations.
- Questions second. Let staff ask where an order is, how much stock is left or which supplier is late, and get the answer from your live system.
- Forecasting third. Once the item data and stock counts are clean, start with your most important product group and compare the model with your current method for a few months.
- Suggestions, not automatic actions. Let AI draft purchase orders and flag exceptions. Keep approval with a person until the numbers prove it can be trusted, and even then keep limits.
What should stay with people in operations?
- Approving purchases above an agreed value.
- Supplier relationships, negotiations and switching suppliers.
- Decisions on exceptions that affect customers, such as which orders to prioritise in a shortage.
- Quality and safety checks.
What data do you need before adding AI?
- One clean item list with no duplicates and consistent units.
- Accurate stock counts, reconciled regularly.
- Order and sales history long enough to cover your seasons.
- Supplier lead times and minimum order quantities.
- All of this in one system, or in systems that are connected. If you are deciding between systems, read CRM vs ERP and our guide to custom ERP and CRM development cost.
How do you measure AI in operations?
- Forecast error by product group, against your current method.
- Stock-out rate and lost sales.
- Inventory days or cash tied up in stock.
- Order cycle time from order to delivery.
- Time spent per document and the share of documents processed without correction.
Our guide on how to measure AI ROI shows how to set the baseline first.
This article is part of our series The AI-Ready Company. Read the opening article, Traditional vs AI: what really changes and why most companies use AI wrong.
Want your team to ask your ERP where an order is or which items are running low, and get the answer in one sentence? That is what we build with Amrak. Tell us how your operations run today. Talk to us.
