The starting point
A mid-sized training and education company. Operations are spread across more than 10 different systems: LMS, CRM, scheduling, payments, participant communication, content storage and analytics. Every new course run means dozens of manual steps somebody has to trigger in the right order.
Growing from 10 to 50 runs a month broke the whole thing. The traditional answer, hiring more operations people, was both slow and expensive.
The problem
- Fragile manual handoffs. Every handoff between systems had to be triggered by a person.
- Errors at scale. With several runs in flight at once, errors compounded. Missing sign-in links, wrong billing details, missed reminders.
- Headcount as the only lever. Operations grew linearly with volume, which was economically unsustainable.
The solution: an AI orchestrator
We built an AI orchestrator that connects 10+ systems and runs operations semi-automatically, with human-in-the-loop checkpoints at the critical decisions.
How it works:
- The agent sees the state of every connected system. It knows who is enrolled, who has paid, and who should get what and when.
- The agent proposes actions: send the invitation, update the CRM, raise the invoice, send the reminder.
- An operations person approves the critical actions in a single interface. For routine work the orchestrator runs on its own, but every action is logged and auditable.
We deliberately do not publish the stack. It combines an LLM agent layer, a deterministic orchestration layer, and an audit and checkpoint UI. We discuss the details with clients on a call.
What changed
- Preparing a run went from days of manual coordination to hours.
- Critical omissions (a missing link, an unpaid invoice) are almost eliminated, because the orchestrator never forgets.
- The operations team does less copying between systems and more work with people: participants, instructors and clients.
- Scale means the company now handles volume that would previously have required a substantially larger ops team.
Specific figures (before and after, the percentage reduction in manual steps) are being finalised with the client. We will update this page once they are approved.
What transfers to your company
- Start with one fragile handoff, not the whole company. The smallest place where it most often falls over.
- Human-in-the-loop is not a weakness, it is a safety net. On the first run, have a person approve every action. After a week you will see which actions are routine enough to automate and which are not.
- Log everything. Auditability is what turns an AI experiment into a production system.
How to bring this into your company
If you want to know what an orchestrator like this would look like for you, let us start with a 15-minute call. We go through whether it makes sense in your company and which of our workshops is the best starting point.