AI module · complements the 12 weeks · about 4 hours

Logistics and supply chain with artificial intelligence

Mini MBA Logistics and Supply Chain

Module objectives
  • Use AI and analysis tools to improve forecasting, procurement and stock monitoring.
  • Measure the reliability of a forecast (mean absolute percentage error) before using it.
  • Keep procurement decisions and transport information under human control and secure.

This module complements the 12 weeks of the programme. It adds no exam: certificates and the diploma are earned as before. Tools are described as known in November 2026; they change quickly.

1Demand forecasting and stock management

Forecasting methods range from moving averages to statistical or machine learning models. Generative AI mainly helps to write the formulas or scripts, explain a method, and spot seasonal trends in a consumption history.

Conditions for reliability:

  • Clean data: real consumption (not the quantities distributed during a stock-out), consistent units, stock-outs flagged.
  • Error measurement: the mean absolute percentage error (MAPE) compares forecast and reality over past months.
  • Field judgement: an influx of displaced people, an epidemic or an unusual rainy season are not in the history.

In anticipatory action, forecasting models (floods, drought) trigger funding before the peak of a crisis; they rely on thresholds validated in advance by experts, not on a chat assistant.

2Procurement: analysing bids without delegating the decision

AI can:

  • Prepare a tender dossier from your template: clear specifications, criteria and weightings.
  • Build the comparison table of bids (price, lead times, technical compliance, warranties) from the documents received.
  • Flag missing documents or inconsistencies (unit price × quantity different from the total).

But evaluation and award remain the responsibility of the procurement committee, under the rules of the organisation and the donor. Supplier bids are confidential: use only an approved professional account, and never a consumer tool during the confidentiality period. Any extraction error (a misread lead time) can distort the scoring: check every figure against the original bid.

3Transport, access and imagery

Useful tools: maps and access data shared by the Logistics Cluster, satellite imagery analyses (for example from UNOSAT or specialised services) to assess roads, bridges or flooded areas, and GPS vehicle tracking.

AI can help calculate routes, compare costs and lead times, or summarise access reports. Precautions:

  • A map or image may be out of date: check the acquisition date and confirm with teams on the ground.
  • Convoy routes, schedules and positions are security information: never in a consumer tool.
  • Cost optimisation must not override security and local acceptance.

4Warehouses, customs and documents

  • Stock: alert formulas for expiry dates (first expired, first out), reorder points and inventory discrepancies.
  • Customs: AI can help you understand a procedure or prepare a packing list, but customs codes, exemptions and licences are checked with the freight forwarder and the authorities. A wrong code can block a shipment for weeks.
  • Dual-use goods and sanctions: checks done with the official lists and compliance procedures.
  • Health products: product quality follows the requirements of regulatory authorities and of the organisation; AI does not replace prequalification.

5Data, responsibility and the environment

A supply chain produces a lot of data: purchase orders, delivery notes, distribution lists. Distribution lists contain personal data: apply data minimisation and never send them to an unapproved tool.

Responsibility: if a wrong forecast causes a stock-out of medicines, the responsibility remains the organisation's. Document the method, the measured error and the decisions taken. Finally, AI can help estimate the carbon footprint of transport or compare local purchase with import: these estimates depend on emission factors that must be cited and checked.

Exercise: measuring the error of a forecast

Situation. A model forecast the monthly consumption of hygiene kits. Over three months: forecast 1,000, actual 800; forecast 1,200, actual 1,000; forecast 900, actual 900.

Questions. 1) Calculate the absolute percentage error for each month (|forecast − actual| / actual). 2) Calculate the MAPE. 3) Do you use the model as it is?

Answer key. 1) 200 / 800 = 25%; 200 / 1,000 = 20%; 0 / 900 = 0%. 2) MAPE = (25 + 20 + 0) / 3 = 15%. 3) The model overestimates two months out of three: it can serve as a basis, but the bias must be corrected (check whether the "actuals" were reduced by stock-outs), the error tracked every month, and a safety stock calculated from that error kept.

Key points

  • A forecast is only as good as its data: real consumption, stock-outs flagged, consistent units.
  • Measure the error (MAPE) before using a model, and track it every month.
  • AI prepares the comparison table; the procurement committee evaluates and awards.
  • Convoy routes, schedules and positions: never in a consumer tool.
  • Customs, sanctions and product quality: checked with official sources.

Self-assessment quiz

6 self-assessment questions, corrected immediately. They do not count towards the exams.

1Forecast 1,000, actual 800. Absolute percentage error:
2Monthly errors of 25%, 20% and 0%. MAPE:
3Who awards a contract after the bids are analysed?
4Can you ask a consumer assistant to optimise a convoy's exact route with its schedule?
5Why do quantities distributed during a stock-out distort the forecast?
6A customs code suggested by AI must be:
Going further

The Mini MBA AI and Humanitarian Action (Claude and Gemini) covers these topics in depth over 12 weeks.

See the AI Mini MBA