AI cost surprises are the worst. This guide walks through the dashboard: daily consumption, breakdowns by model and user, exporting monthly invoices for finance, and setting up budget alerts.
Key takeaways
Go to Billing → Usage. The top chart shows the last 30 days of daily credit spend. Hover over any day for exact numbers. A red spike (consumption multiples of average) usually means someone ran a batch job or an integration went wrong — dig deeper to find out.
The breakdown table below the chart slices by dimension. Pick By Model to see Haiku vs Sonnet vs GPT-4o share (if expensive models dominate, consider downgrading). Pick By Member to find the top spender (someone batch-testing or over-calling). This is the key tool to root-cause month-over-month cost spikes.
Go to Billing → Invoices, pick a month, click Export CSV. You get a detailed line-by-line breakdown of every call that month — timestamp, user, model, credits, GBP amount. Hand the CSV to finance for departmental cost allocation.
When wallet balance hits zero, the Owner gets an auto-email. For earlier warnings, set a custom threshold in Billing → Settings (e.g. alert at 20% remaining). The safest setup: bind a card and enable auto top-up — when balance drops below threshold, it tops up automatically, preventing API rejections from low balance.