AI Platform · FinOps
Cost per outcome, not cost per token
Token spend only matters against the unit it buys: a resolved ticket, a cleared compliance verdict, a research answer. Each surface has a daily ceiling that pauses it and pages its owner rather than quietly overspending.
Model spend today
₹4.3 L
Against plan
-18%
caching and routing
Surfaces at ceiling
0
ceilings pause, never overrun
Modelled annual saving
₹14.2Cr
audited monthly with finance
Spend by surface
Today, in rupees
Stoxo research₹1.84L
Voice concierge₹96k
Support desk₹62k
Compliance guard₹41k
Marketing studio₹28k
Retention₹19k
Unit economics
What the same outcome used to cost
Cost per research answer₹3.80₹9.40
Caching + routing cut it 60%.
Cost per resolved ticket₹1.10₹48.00
Human tier-1 cost was ₹48.
Cost per voice minute₹2.40₹14.00
Includes transcription and QA.
Cost per compliance verdict₹6.20₹380
Reviewer time, loaded.
Cost per creative variant₹18₹2,400
Agency rate card.
Cost per route
Caching is the biggest single lever, not model choice
| Route | Calls today | Cache hit | In / out per 1k | Est. cost | Effort |
|---|---|---|---|---|---|
| Stoxo research answer | 48,120 | 31% | ₹0.9 / ₹3.4 | ₹143 | high |
| Support reply draft | 92,300 | 58% | ₹0.08 / ₹0.3 | ₹15 | low |
| Compliance rewrite | 6,410 | 12% | ₹0.9 / ₹3.4 | ₹24 | high |
| Voice turn generation | 1,38,400 | 64% | ₹0.05 / ₹0.2 | ₹12 | low |
| Call transcription | 21,900 | 0% | ₹0.04 / ₹0.06 | ₹2 | low |
| Creative variant generation | 1,840 | 8% | ₹1.2 / ₹4.1 | ₹9 | medium |
| Member semantic search | 4,10,200 | 77% | ₹0.01 / ₹0 | ₹1 | low |
| Churn reasoning | 9,600 | 22% | ₹0.9 / ₹3.4 | ₹32 | medium |
Spend guardrails
Written as controls, enforced in the router
- ·A hard daily ceiling per surface; breach pauses that surface and pages its owner, it never overspends quietly.
- ·Cache-hit floor per route — a route falling below its floor is treated as a regression, not a cost blip.
- ·Reasoning effort is a budget line: high effort is allowed only on regulated and research surfaces.
- ·Every rupee reconciles to a gateway run id, so the ledger is auditable end to end.