DataPulse AI / Spend

Spend and Query Tuning

Know where the money went and where it is heading, catch a budget breach before month-end, and switch off what is idle, with every change approved and only measured savings counted.

Book a demo Read the Spend guide

Where the money went

Your bill, explained from your own system tables.

Cost by SKU (Jobs Compute, SQL Warehouses, All-Purpose, declarative pipelines), costliest first. Billing rows hold DBUs, so cost is DBUs multiplied by the list price.
Chargeback by tag, with the untagged share shown so you can see which cost has no owner.
Spend anomalies flagged by statistics: spend more than two standard deviations above its rolling 30-day mean. No AI is used.
A month-end forecast against your budget target, with the method one click away.
Every system-table query has a date filter, so it scans only recent partitions and stays cheap.
Every number has a How calculated link that opens its query.

The Spend page with spend, forecast, verified savings, spot ratio and idle waste cards, a spend trend with a forecast band, and chargeback by cost-center tag. — Spend · sample data, not a customer's bill

Budget guardrails

Budgets that propose the fix.

Guardrails check your budgets when you press Evaluate Guardrails and propose a fix, such as stopping a warehouse that has no running query, for a person to approve.

Approval mode is the default: each proposal waits in the Actions inbox with its risk, its daily saving and how to roll it back.
A warehouse stop is low risk only when the query history shows no running or queued query, and it is checked again right before it runs.
Auto-Idle mode (Business tier): low-risk actions on idle resources run without a person, and each is recorded and reversible. Cluster terminations are always proposed, never automatic.
Rightsizing proposals with CPU and memory p95 evidence (Business tier).

Guardrails do not cap your bill. They propose a change or, in Auto-Idle, act on idle low-risk resources.

Budget guardrails with two budgets, a breach that proposes terminating an idle cluster, rightsizing proposals and spend anomalies. Numbered markers label the parts. — Budget guardrails · sample data. Markers from the user guide: 1 the guardrail panel, 2 Evaluate Guardrails, 3 a budget, 4 a breach with its proposed fix, 5 review it in the Actions inbox.

Verified savings

A saving counts only after it is measured.

An executed action becomes verified only after its observation window has passed (7 days unless your admin changes it) and a person presses Verify. DataPulse then measures the effect against the window before the action. Nothing verifies on a timer, and estimates, including every AI estimate, never count as verified savings.

Query Tuning

Find the slow, expensive queries. Get a safer rewrite to test.

Top slow queries from system.query.history: SELECT statements from the last 30 days that ran 10 seconds or more, spilled to disk, or read 1 GB or more.
An AI optimizer rewrites a query with a model your admin allows. A safety check confirms the rewrite is a read-only SELECT.
The saving shown is labelled an AI estimate. DataPulse does not measure the effect of a rewrite: test it yourself.
Other users' emails are shortened unless you are an admin.

The Query Tuning page listing slow queries with duration, spill and bytes read, and the AI optimizer workbench. — Query Tuning · sample data

Known limits. Rightsizing, idle waste and the cluster side of guardrails read classic-compute system tables, so a workspace with only serverless compute has nothing to show there. The anomaly list flags individual billing rows and can be long: treat it as a list of leads. See the Spend guide.

See your own spend, read-only.

The Free Health Check runs the read-only Spend view and Pipelines board on your workspace for 30 days, with no licence fee.

Book a demo See pricing