AI
Token economics, not just totals
Split every dollar into rate, volume and mix. When the bill jumps 28% on lower usage, the waterfall says so — and names the model and token class that did it.
Blended rate $0.94 → $1.32/M on 9% lower volume
AI · Cloud · SaaS in one ledger
Technology spend, understood
Most tools tell you the bill went up. This one tells you whether that was price, usage, or what you moved between them — and shows the arithmetic.
Read-only keys · free beta · no card required · demo data in one click
Month-to-date
$17.0K
▲ 3% vs last month
AI spend
$6,969
▲ 23% MoM
Savings identified
$11.7K
8 open findings
Daily spend — last 30 daysstacked by category
AICloudSaaSanomaly flagged
Connects toAnthropicOpenAIAzure OpenAIGoogle Gemini / GCPMicrosoft 365 + CopilotSnowflake CortexSaaS register + CSVNext: Cursor · GitHub Copilot · ServiceNow Now Assist · Databricks · AWS
The problem
Your AI spend doesn't arrive as an AI bill
It meters in tokens, credits and assists rather than instance-hours — so tools built for cloud cost can't read it. And most of it lands on invoices nobody labels “AI”, owned by teams who never see it as AI spend.
Snowflake Cortex
bills as credits
- Lands on
- the data-platform invoice
- Budget owner
- Data engineering
- In an AI budget line
- No
ServiceNow Now Assist
bills as assists
- Lands on
- the ITSM invoice
- Budget owner
- IT operations
- In an AI budget line
- No
Microsoft Copilot
bills as AI credits
- Lands on
- the M365 invoice
- Budget owner
- Workplace IT
- In an AI budget line
- No
Salesforce Agentforce
bills as flex credits
- Lands on
- the CRM invoice
- Budget owner
- Sales ops
- In an AI budget line
- No
A company that believes it tracks AI spend is usually tracking only its API vendors. SpendIQ starts with AI — then brings cloud and SaaS into the same ledger, so the total finally reconciles.
28
deterministic rules
10 AI · 14 cloud · 4 SaaS, plus anomaly detection
7
connectors live
AI vendors, cloud, M365 with Copilot, Snowflake
FOCUS 1.2
open cost standard
The FinOps Foundation schema, not a proprietary one
1
AI-assisted step
Everything else is arithmetic you can audit
Product
Three kinds of spend, one place
They behave differently, so they are measured differently — and reconciled to the same total.
Cloud
Evidence before advice
Recommendations carry the utilization that justifies them — CPU, memory, disk, liveness — so rightsizing is a decision you can defend, not a guess from a billing line.
Idle VMs flagged with the CPU and memory that prove it
SaaS
Seats you're paying for and nobody uses
Licences reconciled against real activity from Microsoft Graph and a register you control, with an owner against every subscription.
Seats reconciled against real sign-in and Copilot activity
Why the bill changed — last 30 days+$5,757 (28%)
Volume fell. The increase was almost entirely rate — the price per token moved, not how much you sent. That is a pricing conversation, not an engineering one.
What it actually does
Every item here is live today and shows in a demo.
Attribution that names names
Which key, which project, which person. And when invoiced cost cannot be tied to a key, the remainder is shown — not hidden.
Alerts that name the mover
Not just "+49%". The email says Azure went from $438 to $1,181 overnight, so the first message already has the answer.
Findings with a lifecycle
Approve, dismiss, snooze, assign. Dismissed findings stay dismissed — re-analysis never resurrects a decision someone already made.
Anomalies against a baseline
Each day is judged against its own trailing history, per service and per key. A runaway agent loop surfaces the morning it happens.
Every finding links to its proof
A recommendation about a key opens that key. One about a resource opens that resource. The evidence is one click away, always.
Try it before you connect anything
One click loads 90 days of realistic, clearly-labelled demo spend. Evaluate the whole product without sharing a single credential.
How it works
From keys to a number you trust
01
Connect read-only keys
Anthropic, OpenAI, Azure, GCP, Microsoft 365, Snowflake. Reporting scopes only — nothing can spend money or change infrastructure.
02
Spend lands in one ledger
FOCUS 1.2-aligned, idempotent, restatement-safe. The first sync backfills 90 days so there is history to compare against on day one.
03
Decide, with the arithmetic in front of you
Every finding shows the evidence that produced it. Approve, dismiss or snooze it — by a named person, with a reason, recorded.
Anomalies, with a baseline
A day is flagged against its own trailing history, per service and per key — so a runaway agent loop surfaces the morning it happens, not on the invoice.
Seats you can actually reclaim
Assigned licences checked against real activity, with an owner on every subscription so the cancellation conversation has someone to have it with.
Security
Read-only by construction
Adopting SpendIQ never widens your blast radius. Isolation is enforced by the database, not by remembering to write the right query.
Reporting keys only
Connectors ask for read scopes. Nothing SpendIQ holds can spend money or change infrastructure.
Row-level security
Each tenant reads only its own rows, enforced by Postgres policies under a role with no bypass.
Encrypted per tenant
Secrets are encrypted with a key derived per workspace. One tenant's key never decrypts another's.
Roles that mean something
Owner, admin, member, viewer. Every action checks a capability; unknown roles get nothing.
A full security review was completed, and every Critical and High finding closed, before the first external user. Security summary available on request for your review team.
See the whole product before you connect anything
One click loads 90 days of realistic multi-provider spend on clearly-labelled demo connections — with a runaway-agent spike and low cache-hit patterns baked in, so the intelligence engine has something to find. Purge it whenever you like. It never mixes with real data.
Questions we get asked
Is this "AI-powered"?
Mostly no, and on purpose. Findings come from 28 deterministic rules with the evidence attached, so you can check them. Exactly one step uses a model: rating how hard each recommendation is to implement, with a one-line rationale. Everything else is arithmetic you can audit.
What happens to our credentials?
They are reporting keys with read-only scopes — nothing SpendIQ holds can spend money or change infrastructure. They are encrypted at rest with a key derived per workspace, so one tenant's key never decrypts another's.
How is this different from our cloud provider's cost console?
Three ways. Their console sees only their own spend, not your AI vendors or SaaS. It has no idea what a token costs you per feature or per person. And it shows you a number rather than asking anyone to make a decision and recording who did.
Do you support AWS?
Not yet — Azure and Google Cloud today, AWS next via CUR 2.0 / FOCUS exports. If AWS is most of your spend, tell us and we will talk about timing honestly.
Can you prove the savings?
We show you exactly where your spend goes, what looks wrong, and the evidence for it. Realised-savings tracking — proving the money was actually saved after you acted — is on the roadmap, and we will not claim it until it is built.
What does it cost?
Nothing during the beta. No card, no time limit, no feature gating. Connect one provider and there is something real to look at within minutes.
Create your workspace
Free while in beta. Connect one provider and the first sync backfills recent history, so there is something real to look at within minutes.
Takes a minute · no credit card · invite your team after