Surprise bills
A voice product's bill doubled as calls got longer. Nothing warned anyone until the invoice arrived.
$350 → $700 / moBilling-grade cost attribution for companies where AI is the COGS. Every request tagged to a customer, team and feature, governed before the bill exists, and routed to the cheapest model that passes your quality bar.
OpenAI · Anthropic · Bedrock · Vertex · your own GPUs
API bills show tokens, not customers. Seats show licences, not people. GPU clusters show nothing at all. Three ways AI spend goes dark.
A voice product's bill doubled as calls got longer. Nothing warned anyone until the invoice arrived.
$350 → $700 / moTwenty AI seats across four tools. The CEO can't see who spent what, or whether it made the company money.
52% of companies have no AI cost ownerAPI bills at least show tokens. A private inference cluster reports GPU-hours and utilisation, never cost per customer.
56% run or plan inference on private cloudSources: Broadcom private-cloud survey · industry AI cost-ownership survey · UnitSense design partners
UnitSense sits in front of your providers as a gateway. It acts on every request as it happens, instead of reporting on it a month later.
The tag is the chargeback. The month's tagged spend reconciles to the provider invoice, to the cent, so finance can defend the number.
Budgets are table stakes. Custom policies switch models, downgrade a tier or block a use case, before the provider is called. Every action lands in the same ledger as the spend.
Frontier and open-weight models compared on cost, quality and latency, on your own traffic. Scenario modelling tells you the margin impact before you switch.
| OUTCOME | VOLUME | PER OUTCOME |
|---|---|---|
| PRs merged | 1,240 | $4.00 |
| Cases resolved | 8,600 | $0.80 |
| Homes sold | 310 | $7.00 |
Panel numbers are illustrative, not customer data.
Same H100, two very different costs per token, depending on utilisation. The meter that already sits in front of your API traffic goes in front of your own GPUs too: usage times hardware cost, per customer, owned and API side by side.
Live for API traffic today · GPU join on the roadmap
Today's tools report GPU-hours and percent busy. None report cost per request, or per customer.
How we connect: gateway (LiteLLM, Bifrost) · Python and JavaScript SDKs · our open-source agent · provider admin APIs · CSV import · invoice ingestion
…and the rest of your stack. Where a vendor exposes usage, we pull it. Where they don't, we ingest the invoice.
Priced on the AI spend we govern, so the subscription stays a small fraction of the bill.
Map every AI tool, provider and invoice. Set budgets and cost policy. UnitSense live in week one.
Small teams start on their own. Attribution, budgets, weekly digest.
Gateway in the request path. Budgets enforce, requests route, every dollar reconciled to the invoice.
Dedicated gateways, partner deployments, metering on GPUs you own.
Everything built for you during onboarding is included. Sales-assisted today; self-serve tier opens 2028.
Bring your last AI invoice. We'll show you what it would look like tagged, governed and routed, and whether discovery is worth your time.
Prefer slides? margin-deck · governance-deck.