Policy, leave and payroll agents took over the routine queue — answering from policy documents and executing workflows with approval checkpoints. Routine queries reaching HR teams fell by 40–60%.
The situation
Most of what an HR service desk handles is not difficult; it is repetitive. How much leave do I have, what does the policy say about this, why is this deduction on my payslip. The answers exist in documents and systems the employee already has access to, but finding them is harder than asking a person.
So the queue fills with questions that have documented answers, and the HR team spends its time on lookups instead of the cases that genuinely need a human.
What we did
We built agents for the three highest-volume areas — policy, leave and payroll. They answer from the organisation’s own policy documents, with citations, so an employee can see where an answer came from. Where a question turns into an action, such as submitting or amending a request, the agent executes the workflow but stops at defined approval checkpoints.
Those checkpoints are the point. HR actions have consequences — pay, entitlement, employment record — and the design question is not whether an agent can complete the task but where a human has to confirm before it does.
The outcome
What made the difference
Grounding every answer in a citable source, and being deliberate about which actions an agent may complete alone. Earned autonomy — start with answering, graduate to acting, always with a checkpoint where the stakes justify one — is the model we apply across agent work, and it is set out in the AGENTX and SENTRA frameworks.
We publish the metrics and withhold the names — our clients’ competitive advantage isn’t ours to advertise. Talk to us about the queues worth automating in your organisation.
