Context Management for Business: The Right AI Context for Every Task

An AI tool answers a customer enquiry. The text reads well, the salutation is right, the tone fits — and the delivery date it quotes stopped being valid with the last price list. Such errors rarely originate in the language model. They originate before it, in the question of which information the AI saw for this one task at all.

That question is context management. It arises in every company that uses AI seriously, and it arises long before any tool is chosen.

What context management means in a company

Context management is the ongoing job of deciding which trustworthy information a person or an AI system needs for a specific task: where it comes from, who may use it, how current it is, and how a result can later be traced back to its source. Uploading every document into a chatbot is not what is meant. Context management always starts from one concrete task and works its way from there to the sources.

Knowledge management, AI context and context engineering

Four terms get mixed up here, in the German-language debate as much as anywhere. Knowledge management asks how company knowledge is captured, maintained and shared. Context management asks which dependable information a specific task needs, where it comes from, who may see it and how current it is. AI context is the package of facts, rules, examples and history that an AI system actually has in front of it for exactly this task. Context engineering is the technical practice of selecting, structuring and delivering that package. This distinction is our reading of the field and not a claim that the industry has settled on one definition.

A worked example: answering a customer email

Take an enquiry about a spare part, of the kind that arrives several times on any working day. A usable answer is made of five components, and every one of them sits somewhere else.

• The article and its successor, in the ERP.
• The price, from this customer's agreed terms.
• The lead time, from the supplier's current commitment.
• The rule for the amount above which someone counter-signs.
• Three earlier replies that carry the tone.

An employee who has been there a long time pulls this together in five minutes and notices when something does not add up. An AI tool without that access writes a fluent text with the wrong date. The difference lies entirely in the context.

The six questions that define a context path

For every task AI is meant to take on, we settle six points. They are deliberately cut so that an owner can answer them without using software vocabulary.

1. Task: what exactly is to be decided or produced?
2. Source: where does every detail that changes the result come from?
3. Responsibility: who may correct it, and which source wins in a conflict?
4. Permission: who or what may read it, and for what purpose?
5. Freshness: how quickly does it become wrong, and how does anyone notice?
6. Evidence: how will you know in everyday use that the context improved the result?

Where one of these questions stays open, the cause almost always sits in an unsettled responsibility.

What order alone already achieves

How much this groundwork is worth can now be put in numbers. In May 2026 a group at UCLA used LongMemEval-V2 to measure how well different setups make the experience of weeks of work inside an environment available again. Without any access to that experience, current frontier models reach 14.1 %. A purpose-built memory system that analyses every observation at write time and sorts it into three separate knowledge pools reaches 58.6 %. A standard tool allowed to search the same trajectories as plainly ordered files reaches 69.9 %. The best variant lands at 74.9 % and is that same standard tool, extended with a written working instruction, a table of contents and a few helper scripts.

The most effective single component was the working instruction. Remove that plain document and accuracy on the large run falls from 70.1 % to 64.1 %. For a company this means that an ordered filing structure and a current description of the workflow carry further than a database sitting on top of an unsorted archive.

These numbers do come from two artificial test environments in English, where there are no permissions, no conflicting sources and no data protection. They show how well a system finds the right place inside existing material, and say nothing about whether that material is correct and who may see it. On top of that, 74.9 % means one answer in four is wrong. A well-built AI context is therefore a help today for someone who can check the result.

When technology joins in

Only once the task, the sources and the responsibilities are settled is the question of tooling worth asking. We walk it in this order and stop as soon as the task runs reliably.

1. Remove obsolete and duplicate material and name an authoritative source.
2. Improve search, filing and permissions in the existing system.
3. Connect the task to live data through an export or an interface.
4. Add a search across approved documents.
5. Assemble the context for the task automatically and let the AI draft, checked by a person.
6. Permit bounded actions once reading is reliable and logged.

The first two steps cost no project at all, and in many cases the task is done with them. Where several programs, company-specific rules or sensitive data come together, the later steps start to make sense — the same line we draw for using AI in your business.

Where to start

We start with a single task that today depends on one experienced person. You do not need a requirements document for that. You show us a real case and explain in your own words how it runs today, who is involved and what should improve. We walk through it, ask about the exceptions, and record which information is missing at which point. What comes out is a context map for that task, a decision on authoritative sources and ownership, a recommendation for the smallest sensible intervention, and an estimate for implementation and upkeep.

If you have a task in mind where AI fails today for lack of context, let us talk it through. Book an intro call.