Routine is the word for all those activities that are necessary but get nobody anywhere. Sorting tickets. Maintaining master data. Answering standard requests. Moving attachments. Filling in fields. In every industry and every size of company, a substantial share of working time goes into such activities. They are necessary — but they are not the reason anyone became a specialist.
Where routine actually comes from
Routine does not come from a single activity, but from its repetition in identical form. Reading an invoice is not routine. Reading a thousand invoices a month, all in the same structure, is. Prioritising a ticket is a decision. Sorting the same five ticket classes every day is mechanical. Answering an email can be demanding. Covering standard requests with the same three text snippets is not.
This is exactly where AI automation comes in: at the repeated, formal, rule-based processing — not at the substantive judgement.
Which activities AI reliably takes over
Four categories are production-ready today:
- Classification: sorting incoming items by type, priority, responsibility — from tickets through emails to documents.
- Extraction: pulling structured data from unstructured documents — invoice fields, contract clauses, shipping information.
- Pre-filling: populating fields, drafts or documents with expected values that a human only checks.
- Summarisation: turning long texts, email threads or minutes into manageable overviews.
In all four cases the substantive decision stays with the human — AI delivers the preparatory work, the human confirms or corrects. That is the pattern that works in practice.
Why relief does not mean reduction
The worry is understandable: if AI takes over routine, does work disappear — and do jobs go with it? The answer depends on what is meant by “job”. A clerk who today spends 70 per cent of her time sorting and pre-filling cases has 30 per cent for actual case work. With AI preparation, that ratio flips.
In most of the companies we have worked with, the result was not staff cuts — but a shift in what staff actually do: less mechanics, more advice. Less data entry, more customer contact. Less ticket closing, more problem understanding. The roles stayed, but they became more interesting — and more productive.
What a pragmatic entry looks like
Pragmatic means not the biggest lever first, but the clearest. An activity that
- occurs daily,
- is standardised,
- is measurable (count, duration, error rate),
- and whose transfer to AI is legally unproblematic (no personal decisions with serious impact),
is a good candidate for entry. Typical candidates: ticket classification in IT service, document pre-capture in accounting, response suggestions in customer communication.
The first four to six weeks show whether the model meets the quality requirements. If yes, roll out. If not, adjust. Either way, a concrete understanding emerges of what works in-house — and that is the basis for everything that follows.
What remains in the end
What remains is the work that only humans can do: moderating a conflict with a customer well. Defending a creative idea. Answering an unusual request with professional judgement. Exactly this work is the reason companies hire specialists in the first place — not the kind AI conveniently absorbs.
AI takes the tedious. What remains is the essential.