AI and Automation: Understanding What Should Be Automated and What Should Not

The question of what to automate is more important than the question of how to automate, and it receives far less attention. Automation tools have become accessible enough that almost any business process can be automated in some form. This accessibility is valuable, but it creates a risk that did not previously exist: the risk of automating the wrong things and investing in systems that deliver efficiency on processes that should have been redesigned or eliminated, while leaving the highest-leverage opportunities unaddressed because they were less obviously automatable.

The four categories of work

A useful framework for thinking about automation divides all organisational work into four categories. The first is rule-based and repetitive: work that follows a defined procedure, handles structured inputs, and produces consistent outputs. Data entry, invoice processing, appointment scheduling, report generation. This category is the strongest candidate for automation — AI and workflow tools handle it reliably, the output quality improves or stays the same, and the human time recovered is real. The second category is rule-based but variable: work that follows procedures but encounters enough variation and exception that it requires judgement to handle well. Customer queries, document review, quality inspection. AI can handle the routine end of this category effectively — the standard query, the common document, the in-specification product — but requires human oversight or escalation paths for the exceptions. Partial automation with well-designed handoff points is the right approach here. The third category is judgement-intensive: work that requires assessing ambiguous situations, weighing competing considerations, and making decisions that cannot be fully specified in advance. Strategic decisions, client relationship management, performance assessment, creative problem-solving. AI can provide information and analysis to support judgement-intensive work, but the judgement itself should remain human. Automating this category produces either oversimplified decisions or confident-seeming outputs that substitute an algorithm's pattern-matching for genuine reasoning about a specific, contextual situation. The fourth category is relationship-based: work whose value comes from genuine human connection — trust-building, empathy, presence, accountability. Leadership, client relationship management at the strategic level, conflict resolution, culture-building. AI has no meaningful role in this category except as a tool that supports the people doing the relationship work — by handling information preparation so they can be more present in the relationship itself.

The test before automating

Before committing to automating any process, three questions should be answered honestly. First: is this process in the right category for automation? Rule-based and repetitive is yes. Relationship-based is no. Second: is the process stable and well-understood? Automating a poorly understood process produces a fast version of something that should have been redesigned. If the process is not well-understood, understand it first. Third: what would stop if this automation failed? If the answer is something critical, the automation needs human oversight and fallback procedures designed in from the start, not added as an afterthought.

The human role in an automated organisation

The goal of automation is not to remove humans from the organisation. It is to concentrate human effort where it creates the most distinctive value. In a well-designed automated organisation, the people who were previously executing rule-based repetitive work are doing something genuinely more valuable: building relationships, exercising judgement, solving novel problems, and improving the systems that handle the routine. This is a better use of capable people, and it is what automation makes possible when it is implemented with this goal explicitly in mind — rather than as a cost-reduction exercise that treats human effort as an expense to be minimised.

For further reading on this topic, check out our guide on How to manage ESIC claims and ensure your employees can access benefits.

For further reading on this topic, check out our guide on The Real Cost of Waiting on AI (And Why the Window Is Closing).

For further reading on this topic, check out our guide on How to know when it's time to pivot and when to stay the course.