AI in production planning: how to access knowledge faster when the plan changes

AI in production planning can reduce the time a planner needs to make a decision even when the system itself does not create the schedule. When the answer is in an instruction, procedure, or technical documentation, the biggest delay is often simply reaching the right information. This article shows how to organize this step, where the role of APS or MES ends, and when explitia.Whisp may be the right answer.

If you are looking for information about automatic scheduling or resource optimization, see our article on AI in production planning. Here, we focus on a slightly narrower area in which the planner has operational data but still needs knowledge stored in documents to make a decision.

Planning is based on deadlines, orders, materials, and the availability of people and machines. When changing the order of jobs, however, it may also be necessary to check the permitted technology, quality requirements, a changeover instruction, or product-specific restrictions.

This is where AI in production planning can play a different role from APS. A planning system recalculates the schedule based on data and constraints. A knowledge assistant helps find information in company documentation when an employee needs it to assess the situation.

This distinction separates AI in production planning from the classic use of AI for schedule optimization. Here, we want to focus on what to do when the plan itself is not enough because the next decision depends on company knowledge.

A hypothetical example illustrates the problem well. A machine drops out of the plan and an order has to be considered for another workstation. Information about resource availability is available in the production system. Before moving the order, however, the planner has to check whether the selected product can be made on that machine, which parameters apply, and whether there are any additional quality requirements.

When documentation is scattered, the planner has to determine where the current instruction is stored, find the relevant section, and check whether it applies to that product variant. Sometimes a consultation with a technologist is also necessary.

In this scenario, AI in production planning does not replace the planner’s decision. It shortens the information-search stage, provided that the company has already built a reliable knowledge base.

explitia.Whisp is an AI assistant available in the Production Portal. The user asks a question in natural language, and the system searches for semantically related fragments of materials and prepares an answer based on the company knowledge base.

The knowledge base can include instructions, procedures, technical documentation, and spreadsheets. Whisp supports TXT, Markdown, PDF, Word, and Excel files, keeps conversation history, and allows scenarios to be created that define the assistant’s role.

For a project involving AI in production planning, the product boundaries are important. Whisp currently does not automatically retrieve live machine data, monitor production on its own, or perform operations on equipment. It works with prepared documents. If the problem is automatically recalculating the order of jobs, a planning solution is needed. If the problem is quickly finding the knowledge required for a decision, Whisp addresses exactly this class of need.

In the order-transfer example, AI in production planning can support the knowledge step: the planner can ask Whisp about the conditions for carrying out the operation at an alternative workstation. The assistant searches the knowledge base and prepares an answer based on the documents. The employee still assesses the situation and remains responsible for the decision.

Traditional search works well when the user knows the document name, procedure number, or a distinctive word used in the content. In a planner’s work, the question is more often situation-dependent, for example: under what conditions can this operation be performed at another workstation, or what requirements apply to this product variant?

Whisp uses semantic search, so it can find fragments that are meaningfully related to the question. In the context of AI in production planning, this changes the starting point. The user starts with the problem they want to solve instead of the file name.

This is especially useful with extensive instructions. Even an up-to-date document can still be difficult to use if the employee has to browse many pages to find one condition needed for a decision.

An assistant will not resolve contradictions in the source material. If two instructions specify different parameters or an old version of a procedure is still stored in the directory, the model receives ambiguous material. That is why AI in production planning requires identifying the documents that actually constitute the valid source of knowledge. The quality of the answer is directly related to the quality and currency of the knowledge made available to the system.

A pilot does not have to cover all plant documentation. It is better to choose one process, collect the questions that regularly return to planners or technologists, and map them to the materials they use. Then outdated versions should be removed and a content owner should be identified.

Before adding documents to the knowledge base, check:

If part of the knowledge exists only in the experience of technologists, it first has to be documented and approved because AI models work with knowledge made available in documents. They do not have access to expert knowledge that has never been recorded.

AI in production planning covers different applications, so the name of the technology should not be the first selection criterion. First, determine what is actually blocking the planner.

This division helps define where AI in production planning should begin and reduces the risk that AI integration in manufacturing processes starts with a tool that does not fit the problem. It makes sense where searching for and transferring knowledge consumes time. The lack of a good schedule, the lack of machine data, and the lack of an instruction are different situations, even though each can be described as AI in production.

If what you primarily need is data about current production execution, a separate direction is the explitia Production Portal and MES system, which can combine information from machines, PLCs, ERP, and operator forms, among other sources.

Generative models can produce answers that sound credible even when they contain an error. In the Generative AI Profile, NIST describes this phenomenon as confabulation and notes that it requires particular attention in applications that lead to decisions with significant consequences. For this reason, AI in production planning should have a clearly defined scope of human responsibility.

A tool such as explitia.Whisp helps users reach knowledge, but it does not replace approved procedures or an employee’s decision. For information affecting safety, quality, or technological parameters, the organization should define when the user must check the source or consult the decision with the process owner. The official product description also states that the tool does not replace approved procedures or user decisions.

For AI in production planning, the company should know where the model runs, where the knowledge base is stored, and who can use it. The tool can use locally running AI models, allowing the architecture to be adapted to the organization’s requirements. Security rules still have to be defined for the specific implementation.

The EU AI Act also requires measures that support the AI literacy of people using systems on behalf of an organization. Article 4 has applied since February 2, 2025, so users should understand the capabilities and limitations of the solution they use. The European Commission provides more information in its materials on AI literacy.

A pilot should start with the activities that create specific work, such as reviewing documentation, looking for the current version, or involving a technologist in repetitive explanations. Before launching the tool, it is worth measuring the baseline so that the new way of working can later be compared with it.

For an AI in production planning project, you can measure the time required to reach information, the share of questions that end with the correct source being found, the number of expert consultations, and answer accuracy as assessed by the process owner. The first step is to check whether the problem actually exists and whether the assistant reduces it.

The first pilot can cover one group of procedures or one type of planning decision. If AI in production planning works well within that scope, the knowledge base can be expanded. Weak results will show whether the obstacle is the documentation, the way questions are asked, or the choice of use case.

These criteria are more useful than simply checking whether a chatbot can generate an answer that sounds correct. For a production manager, what matters is the change in a specific work process and the ability to verify where the knowledge used for the decision came from.

Our article on preparing a company and its data for AI implementation provides more information about data quality, process context, security, and choosing the first use case.

If your company has problems with sensors, missing process-data context, or ERP integration with the shop floor, the first step should be to address those gaps. If delays come from access to instructions, procedures, and technical knowledge that is already documented in the organization, you can move directly to an AI in production planning pilot with explitia.Whisp.

The best candidate for a first implementation is a process in which planners regularly ask similar questions, the answers exist in documents, and finding them takes time or requires involving an expert. In that case, AI in production planning has a specific scope and its effect can be tested in the team’s real work.

Start with representative questions, map them to approved sources, and review the answers together with the process owner. Choose the number of questions according to the process scope and the variety of documentation instead of adopting an arbitrary threshold. If the test confirms usefulness, the next step can be to expand the knowledge base to further procedures or user groups.

explitia.Whisp has a clear role: it shortens the path from a question to knowledge stored in company documentation. Understood this way, AI in production planning does not compete with APS or MES. It complements the planner’s work where the next decision depends on information the company already has but the employee first has to reach. If this is the stage slowing down work in your plant, check out explitia.Whisp and evaluate one process in a pilot.

No. Whisp searches for knowledge in prepared documentation and creates answers based on it. Automatic scheduling requires APS functionality or another planning system working with data about orders, resources, and constraints.

Yes, if the task is to access knowledge stored in documentation. MES becomes necessary when the application requires current data about order execution, machine status, or process progress.

It is best to start with current instructions, procedures, technical documentation, and spreadsheets that answer questions that actually recur in users’ work. The pilot scope should be verifiable by the process owner.

No. A local architecture can help meet data-control requirements, but it does not remove the risk of an incorrect answer. NIST identifies confabulation as one of the risks of generative AI, so up-to-date sources and answer-verification rules are required.