What to automate first
Onboarding. Accounts, licenses, groups, permissions, documentation, and welcome emails in a single workflow instead of a checklist that someone follows with varying success depending on the day.
Offboarding. Deactivation, license release, email forwarding, and updated documentation – without anything being left behind. A forgotten account is both a security risk and an unnecessary license cost. See also password and access management.
Ticket triage. Categorization, prioritization, and assignment that happens automatically based on content, customer, or agreement – see service desk and help desk.
System reconciliation. Ensuring that ticketing systems, monitoring and documentation actually describe the same reality.
License management. Reconciliation between what is used and what is paid for.
Don't automate a broken process
The most common expensive lesson in automation is building a workflow on top of a routine that didn't work.
Automation makes a bad process faster, not better. Map out the workflow as it actually looks, fix what is wrong, and automate afterwards.
Three terms that get mixed up
Workflows are rules within a system: when a ticket is assigned a certain category, it is routed to a specific queue. These already exist in your ticketing system and are usually the first step.
Orchestration is the flow between systems: a closed ticket triggers actions in Microsoft 365, documentation, and billing. This is where a dedicated automation platform is needed.
RPA mimics a human in an interface that lacks an API. It works, but it is fragile – a changed button layout can break the flow. Use it when nothing else works, not as a first choice.
Where AI actually adds value
The distinction is simpler than the debate suggests: rule-based automation should handle the predictable, AI should handle what requires interpretation.
Creating an account according to fixed steps doesn't need AI – it needs a rule that runs the same way every time. Determining what a free-text ticket is actually about, summarizing a long thread, or suggesting a solution from the knowledge base are, however, tasks where AI adds value.
Therefore, be skeptical of workflows that use AI for decisions that should be rule-driven. A workflow that is sometimes right is worse than one that always does the same thing.
Start with a workflow, not a platform
The most common mistake after automating a broken process is starting too broadly.
Choose a task that is repetitive, error-prone, and performed frequently. Build it. Run it live for a few weeks. Only then will you know enough about how your systems actually behave to build the next workflow effectively.
Also, account for maintenance. An automated workflow is not finished once it is built – APIs change, license names are updated, and exceptions arise. Without someone owning the workflows, they will fail silently, which is worse than never having built them at all.
Where the value lies
Manual work that scales linearly is what puts a ceiling on growth. What takes ten minutes per customer takes twenty hours when you have a hundred.
Automation raises that ceiling – which is why it is usually more profitable for a growing business than for a stagnant one.
For an internal IT department, the calculation looks different, but the conclusion is the same: it is not the cost that decreases, but the queue. Time freed up from account management goes to what actually requires expertise.
