AI in Operations and Supply Chain: The Right Tool for the Right Problem

Plants.work · · upskilling
AI in Operations and Supply Chain: The Right Tool for the Right Problem

Learn which operations problems need machine learning and which suit generative AI and agents, from demand forecasting and inventory to suppliers and warehouse work.

A frozen food factory in Samut Sakhon supplies retailers nationwide. Last month a best-seller ran out of stock during a promotion, while another product sat in cold storage until it neared expiry. The planning lead knows there is enough data to forecast, but it is scattered across the ERP, sales team Excel files and customer emails. Operations and supply chain is where AI delivers the most tangible results, provided you match the tool to the problem.

There are two kinds of AI in operations, so do not mix them up

The first is machine learning and optimisation, for numerical work such as demand forecasting, safety stock calculation and route planning. The second is generative AI and AI agents, for text and coordination work such as reading supplier emails, summarising incidents or answering questions from SOPs. A common mistake is asking a chatbot to calculate stock levels, which is the job of a statistical model.

The overall results are backed by evidence. McKinsey's article Succeeding in the AI supply-chain revolution (April 2021) states that early adopters of AI-enabled supply chain management improved logistics costs by 15%, inventory levels by 35% and service levels by 65% compared with slower-moving competitors.

Five places to start

1. Demand forecasting that includes external factors

The forecasting model uses historical sales together with promotion calendars, festivals and long holidays. Generative AI helps read promotion plans from customer emails and turn them into data the model can use.

2. Inventory policy by product

Segment products by importance and volatility, then set safety stock and reorder points per group. Short shelf-life products need different rules from long-life ones.

3. Supplier risk agent

The agent monitors news, announcements and supplier emails, and warns about risk signals such as flooding at a source factory or several late deliveries in a row, summarising which orders are affected.

4. Exception management agent

When a truck is late or a delivery arrives incomplete, the agent gathers data from systems, drafts a message to the customer and proposes options, such as moving stock from another warehouse, for the shift lead to decide.

5. Frontline SOP assistant

Use RAG to pull answers from company SOP manuals and safety standards, letting staff ask in Thai on their phones, always citing the relevant manual section.

Example: the frozen food factory

The planning team starts by cleaning historical weekly sales data, then builds a forecasting model that includes each retailer's promotion calendar. An agent reads promotion emails from customers and adds them to the calendar automatically, while the exception agent sends a daily morning summary of orders at risk of missing delivery. The planning lead uses the morning meeting to decide from this list instead of chasing each department.

Clean data and clear processes matter more than the smartest model.

Cautions

How to get started

  1. Pick one problem you can measure in money, such as stockouts or expired goods.
  2. Check at least one year of historical data quality for that problem.
  3. Choose the right type of tool: statistical models for numbers, generative AI for text.
  4. Pilot with a small product group or a single warehouse for 8 to 12 weeks.
  5. Measure against the period before the pilot, then expand.

Businesses that want a team to set up the system from data assessment through delivery may consider SCALE, which works with manufacturing and logistics companies through a 90-day delivery plan. Contact SCALE at scale.co.th. Professionals who want to build automations themselves can try the course Build AI automation agents with Make and n8n and read AI workflow automation.