Cost-per-outcome: return on AI spend
Read the unit economics — what each digital worker saved versus what its AI usage cost.
Updated Aug 3, 2026
AI isn't free — each agent run spends provider tokens. The Agent performance tab (Analytics → Agent performance) shows whether that spend pays for itself, per digital worker.
The columns
- Saved (30d) — savings booked to that worker over the last 30 days.
- AI cost (30d) — what its runs cost in AI tokens, shown in £ (providers bill in USD; it's converted at an approximate rate).
- Net — saved minus cost.
- Return — the headline: £ saved per £1 of AI spend, e.g. "24×". This is the figure a buyer actually asks for.
Where else it shows
Each digital worker card in Automations carries a compact "Nx return · £x AI cost · 30d" line, so unit economics are visible at a glance, not buried in a report.
Reading it honestly
When runs use the free heuristic classifier (no AI key configured), AI cost is near zero and the return is effectively uncapped — so lean on Return once real AI is in use, and always pair it with the verified-vs-projected split so you're comparing confirmed savings against real spend.
Was this article helpful?