Case studies

Real agents, real fixes

What broke, what I tried and what worked, from client agents. Clients stay anonymous.

Time-based actions

The agent knew the opening hours. It couldn’t tell if the office was open.

A client’s phone agent read the opening hours back perfectly, then forwarded calls at the wrong times. A bigger model fixed it but made every reply 1.5 seconds slower. The fix kept the model, the speed and the cost.

Latency kept. GPT-5.1 took 2.1 s: 0.6 s
Model and API cost per call: Same
Small workflow, on the client’s own n8n: 1

Flow design

The flow had a node for everything. That was the problem.

A receptionist agent was split into 12 nodes, one for each department and request. It passed every test, then went live. Replies took over 3 seconds and the agent started making things up. The fix was fewer nodes, not more.

Nodes in the flow, before and after the fix: 12 → 4
Latency on live calls before the fix: 3 s+