Skip to content

AI in warehouse and logistics: instructions, e-mails, stock

By the KI können editorial team8 min read

Written with AI and checked against the sources by a second AI. As of November 16, 2026. More in the notes on AI.

In the warehouse and in dispatch, a lot is routine that still eats time: the instructions for the new scanner, the e-mail to the customer because the truck is stuck in traffic, the list of what needs reordering. An AI can prepare part of it. This guide shows which tasks fit and where safety, data and co-determination set limits.

In a warehouse, a worker with a tablet stands between pallets of boxes, a tall rack behind him.

Key points

  • An AI drafts work instructions, checklists and customer e-mails from your bullet points. The facts come from you.
  • Risk assessments, dangerous goods and load securing are decided by the experts in your company, not by the AI.
  • For stock lists, an AI can build and explain spreadsheet formulas. Test them on figures you know.
  • Delivery addresses of private customers, and names and scanner data of colleagues, don’t belong in a public AI chat.
  • Anyone who wants to analyze employees’ performance with AI needs clear rules and, where there is one, the works council.

Which tasks fit in the warehouse

An AI like ChatGPT, Claude, Gemini or Copilot is good at turning bullet points into a clear text, shortening it or putting it into another language. But it knows neither your warehouse management system nor your customers nor the hazards at your loading dock. So tasks fit well where you supply the facts:

  • Work instructions and checklists from your bullet points
  • Simple versions and translations for colleagues who speak little German
  • E-mails to customers and suppliers, for example about delays or short deliveries
  • Spreadsheet formulas for minimum stock, days of cover and reordering
  • Questions for the handover between early and late shift

Tasks don’t fit where a mistake puts someone in danger or where the AI would need real data about people. More on that below.

In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson

Work instructions everyone understands

Ana is a trained warehouse logistics specialist and shift supervisor in a spare parts warehouse. From next week, incoming goods will be booked with handheld scanners. Her team speaks German, Polish and Romanian. She writes down her steps and has the AI turn them into instructions:

Sample instruction
Turn my bullet points into work instructions for receiving goods with a handheld scanner.
Ten numbered steps at most, one short sentence per step, simple words, no technical terms without an explanation.
Don’t change the order and don’t add any steps. If a step seems unclear or incomplete to you, ask me about it at the end.

Bullet points: take the delivery note, count the pallets, compare with the delivery note, report any difference to the shift supervisor at once, log in to the scanner, scan the delivery note number, scan every package, photograph any damage and mark it in the scanner, put the goods on the storage location shown by the scanner, scan the storage location, complete the booking.

Then give the instructions in Polish and Romanian as well.

Ana has each translation read by a colleague who speaks the language. An AI usually translates everyday language well enough, but one wrong word in an instruction can lead to a wrong move.

Handover between shifts

Things often get lost at handover: a damaged rack, a delivery still missing, a scanner that keeps freezing. So Ana had the AI suggest questions for a fixed handover, which her team is trying out:

Sample instruction
Suggest a template for the shift handover in a spare parts warehouse. Eight fixed questions at most that can be gone through in five minutes.
Areas: goods in, shipping, open orders, technology and equipment, safety and damage.
An empty field to fill in for each question. No questions about people, sick notes or the performance of individual colleagues.

After two weeks, the team discusses which questions help and which nobody fills in. Ana then changes the template herself. The completed handovers stay in the warehouse, on paper or in the system, not in the AI chat.

What you don’t hand over

  • Risk assessment: which hazards there are at racks, loading dock and loading zone, and what helps against them, is assessed by the company, supported by the occupational safety specialist.
  • Forklifts and other industrial trucks: anyone driving a truck with a driver’s seat or platform must be at least 18, trained and authorized in writing by the company. Pedestrian-operated trucks also require training. Instructions from an AI replace none of this.
  • Dangerous goods: labeling, packaging and transport documents follow their own regulations. Ask the experts in your company, such as the dangerous goods safety adviser.
  • Load securing: weights, lashing equipment and the number of straps come from manufacturer data and rules, not from an estimate in a chat.

An AI will still answer questions like these, and it usually sounds plausible. That is exactly the problem: you can’t tell from an answer whether a figure is right or made up.

E-mails about delays and short deliveries

Omar is a dispatcher at a small freight company. When a route is delayed, he has to inform several customers, often at the same time. He had a template built that he fills in with the facts each time:

Sample instruction
Write a short e-mail to a business customer whose delivery is delayed. Formal tone, 90 words at most.
Facts: delivery [order number], planned [date, time window], new time window [date, time window]. Reason: [reason in one sentence].
Give the new time window in the very first sentence. Apologize once, without stock phrases.
Offer that the customer can reach me at [phone number] if the new time window doesn’t suit them.
Don’t promise any compensation or any time that isn’t in the facts.

Omar fills in the placeholders in his e-mail program, not in the chat. That way, neither customer names nor delivery addresses end up with the AI service. With private customers this matters even more: name, address and phone number are personal data under the GDPR.

Before sending, he checks the time window against the route plan. An e-mail with the wrong time is worse than none.

Formulas for stock and reordering

Many warehouses work with spreadsheets alongside the system. An AI can build and explain formulas for you, even if you couldn’t write them yourself. Give it the layout of your spreadsheet, not the real data:

Sample instruction
I have a spreadsheet in Excel. Column A: item number, B: stock, C: average usage per week, D: lead time in weeks, E: safety stock.
Build a formula for column F that shows “Order” if the stock is lower than usage times lead time plus safety stock, otherwise empty.
Build a formula for column G for the days of cover in weeks, rounded to one decimal place. If usage is 0, G should show “no usage”.
Explain each formula in two sentences.

Test the formulas on three rows whose result you can work out in your head, and on one row with zero usage. If one is wrong, describe exactly to the AI what is wrong: “With stock 25, usage 10, lead time 2, safety stock 10, F shows nothing; I expected ‘Order’.”

A spreadsheet with an AI formula is an aid, not a replacement for the warehouse management system. If the system already has reorder points, the system applies.

Data about colleagues and customers

A warehouse produces a lot of data about people: who picked which item when, how long a driver waited at the loading dock, which shift was off sick. Data like this doesn’t belong in a public AI chat, meaning a freely available service with a personal account and no contract with your company. That applies even if you only want to “quickly analyze” something.

If a company wants to use software to analyze its employees’ performance or behavior, a works council (Betriebsrat), where there is one, usually has a right of co-determination. If an AI carries out such analyses or allocates tasks based on behavior or personal traits, the EU AI Act usually treats it as high-risk. The strict obligations for this apply from December 2027. More on this in the guide to the AI Act.

Questions you can ask before you work with AI in the warehouse:

  1. Which AI is approved here, and with which account?
  2. May order numbers or item data go in, or only general questions?
  3. Who reads work instructions before they go up?
  4. Whom do I contact if customer data was entered by mistake?

Check every result

Before an AI text goes up or goes out, go through this list:

  1. Do the order and content of each step match what you really do?
  2. Has the AI added anything you didn’t mention, such as a safety rule you don’t have?
  3. Do figures, times and numbers match the system and the route plan?
  4. Has someone who speaks the language read the translation?
  5. Is nobody identifiable in the instruction and the answer?

A sentence you can’t back up goes. Responsibility for the instructions stays with you and your company.

How to practice this

Good instructions for the warehouse are short, precise and free of names. In the app “KI können” you practice this in short lessons, including examples from warehouses and production, and put the instructions to work in your own AI. The trial lesson needs no account.

Sources

Information as of November 16, 2026. Laws and offers can change; this article is not legal advice.