Skip to content

Delta · AI FinOps

In development

Delta shows what every AI use case costs, who owns it and what it delivers. Budgets that hold, forecasts before month end, and cheaper models only where accuracy holds.

delta · spend overview
SpendBudgetsChargebackForecastAlerts

Month to date

£0

Day 15 of 30

Budget

£0

All use cases

Forecast · month end

£0

4% under budget

Cost per resolved ticket

£0.00

Down from £0.52

Cumulative AI spendActualForecastBudget
budget £26k1 NovToday30 Nov
Use caseSpendBudget usedAction
Support triage£6,240
71%
Switch to smaller model for triage
Quote drafting£4,980
100%
Cap reached · approval needed
Dispatch co-pilot£3,710
58%
On track
Invoice matching£2,140
44%
On track
Knowledge search£1,350
27%
On track

The problem

AI bills arrive after the decisions that caused them.

Most firms find out what AI costs at month end, as one number with no owner. By then the only lever left is to cut, and cutting blindly kills the use cases that were paying back.

0%

of organisations have constrained their use of AI because of what it costs to run.

Source: McKinsey, 2026

01

Spend lives in twenty places

Chat subscriptions, API keys on personal cards, a model bill inside another vendor's invoice. Nobody has the full number.

02

No owner per use case

When the bill doubles, nobody can say which workflow caused it, or who should decide whether it was worth it.

03

Cost without outcome

Tokens and seats tell you what you paid. They do not tell you what a resolved ticket or a sent quote actually cost.

What Delta does

The numbers every AI programme needs. In one place.

Seven views that turn AI from a cost line into something you can budget, allocate and defend.

Concept UI · sample data

Spend by use case and owner

Every model call is attributed to a use case and a named owner, so 'AI costs' stops being one line on an invoice.

Budgets and caps

Set a monthly budget per use case. At the cap, spend pauses for an approval instead of running on into next month's surprise.

Chargeback

Allocate AI spend back to the teams that used it. Departments see their own line, not a shared mystery.

Month-end forecast

See where the month will land while there is still time to act, not after the invoice arrives.

Alerts

Spikes, runaway loops and budgets at 80% reach the owner early, in a private channel.

Model-swap recommendations

Where a smaller, cheaper model scores the same on your own eval set, Delta says so. Where it doesn't, it stays quiet.

Cost per outcome

Not cost per token. Cost per resolved ticket, per quote sent, per invoice matched: the units your business runs on.

Model-swap recommendations

Cheaper only when it is just as good.

Delta tests a smaller model against your own eval set before it suggests anything. If accuracy drops past your threshold, it says nothing. Honest evals beat polished demos.

Support triage

Eval set: 200 real tickets

Concept UI · sample data

Current · large model

Accuracy
94.2%
Cost / item
£0.041

Candidate · smaller model

Accuracy
93.8%
Cost / item
£0.012

Recommend the switch. Accuracy holds (−0.4 pt, within the 1.0 pt threshold). Cost per ticket falls by about 70%.

Contract clause review

Eval set: 60 real contracts

Concept UI · sample data

Current · large model

Accuracy
91.5%
Cost / item
£0.310

Candidate · smaller model

Accuracy
85.4%
Cost / item
£0.070

Stay put. Smaller model drops 6.1 pts. Cheaper, but not worth it. No recommendation made.

Budgets and caps

A cap is a conversation, not a cliff.

When a use case reaches its cap, Delta does not quietly overspend or silently switch it off. It asks the owner, in a private channel, and records the answer.

  1. Step 1

    Budget set

    For example, quote drafting gets £1,500 a month, owned by the sales lead.

  2. Step 2

    80% alert

    The owner hears early, with the forecast for month end.

  3. Step 3

    Cap reached

    New spend waits. Nothing runs on unnoticed into next month.

  4. Step 4

    Human decides

    An approval request lands in Rendly. Raise the cap, or hold.

How it works

Your spend data stays home too.

Usage logs reveal what your teams are working on. Delta runs inside your infrastructure, like every Anoryx product, so that picture never leaves the building.

Self-hosted. No usage data sent to Anoryx or anyone else.

01

Connect

Delta reads usage and billing data from the AI tools and APIs you already pay for, from inside your infrastructure.

02

Attribute

Each call is mapped to a use case and an owner, then rolled up into spend, chargeback and cost per outcome.

03

Act

Budgets, caps, alerts and model-swap advice. Decisions that need a person go to Rendly for approval.

Starting point: the AI spend map from our Opportunity Sprint, an inventory of every AI subscription and API your teams already pay for.

Know what AI costs you. Before the invoice tells you.

Delta is in development. If AI spend is already a line you cannot explain, tell us. Early conversations shape what we build first.