Databricks cost + AI readiness · one read-only pass

The work to cut your Databricks bill is the work that makes AI trustworthy.

One read-only pass finds the spend you can recover now, then turns the same telemetry into a metric reconciliation and agent-readiness roadmap. Nothing leaves your workspace.

Read-only access Runs inside your tenant No raw data required
Sample scan output read-only
Addressable annual spend
$0.0Millustrative
across one workspace, 90 days of system-table telemetry
Oversized SQL warehousesresize · 6 found
Over-refreshed dashboardsreschedule · 11
Repeated query patternsmaterialize · 4
Unattributed serverlesstag · 3 teams
Warehouse idle rate vs comparable tenants80th pct
The pattern

Your spend grows. Nobody can explain why. And the same mess is what blocks your agents.

The cleanup that fixes the bill is the same cleanup that makes AI safe to deploy. Most teams pay for it twice because they treat them as separate projects.

01

Cost climbs without a cause. Warehouses, dashboards, and query habits keep doing expensive work, but no one can attribute the bill to the workloads and behaviors that drove it.

02

Every team counts differently. Finance, Sales, and Product each define revenue, churn, and active customers their own way, so the dashboards quietly disagree with each other.

03

Agents would ground on the mess. You want Genie and AI agents, but they will answer from inconsistent metrics, stale dashboards, sandbox tables, and workflows that nobody formally owns.

04

It is one problem, not two. The inconsistent metrics and unowned workflows inflating your spend are the exact gaps that make an agent unsafe. Fix once, solve both.

One pass, three outcomes

From a single read-only diagnostic.

The same telemetry that explains your spend is the substrate for trustworthy AI. We read it once and deliver all three.

Act 01 · now

Recover spend

Find oversized warehouses, over-refreshed and duplicate dashboards, and repeated query patterns that automatic table maintenance never touches. Savings explained in business terms, with evidence behind every line.

Behavioral waste, not table tuning
Act 02 · next

Reconcile metrics

Surface every conflicting definition of revenue, churn, and active customers across your dashboards and SQL. Name the canonical one. Generate the governed metric view that consolidates them.

The on-ramp to governed semantics
Act 03 · ready

Get agent-ready

Score each workflow for grounding, freshness, ownership, and control gaps. Then emit a portable, standards-aligned grounding pack your agents can actually consume.

Works with Genie, or any agent
What you get

Evidence-backed artifacts, not a slide of opinions.

[ 01 ]

Near-term savings report

A recoverable-spend figure with a peer benchmark, broken down by warehouse, dashboard, query, and attribution, with the evidence and expected impact for each finding.

[ 02 ]

Reconciliation diagnostic

Your conflicting metric definitions, the canonical version for each, and draft metric-view DDL to consolidate them across dashboards, SQL, and AI tools.

[ 03 ]

The Operating Graph

Critical tables, jobs, dashboards, warehouses, owners, and dependencies, with an evidence link on every inferred edge, labeled inferred until your owners confirm it.

[ 04 ]

Grounding pack

A portable, standards-aligned context bundle, ready for your MCP server or agent stack, for Genie or any LLM working over Snowflake, BigQuery, Redshift, or Databricks.

[ 05 ]

Executive briefing

One page for the CDO, CIO, or AI sponsor: business impact, savings opportunity, the top risks, and a 30 / 60 / 90-day roadmap.

[ 06 ]

Remediation backlog

Prioritized P0 and P1 items with suggested owner, evidence, impact, and next step. Pick two or three to implement or measure in a follow-on pilot.

Where this fits

Not another cost tool. Not another catalog. The pass that connects them.

Cost optimizers stop at spend. Semantic layers store the metric you already reconciled. Observability tools watch what is already running. We do the assessment that connects all three.

01

One pass, three layers

We are upstream of the tools you already evaluate. We find the waste, reconcile the metrics, and produce the readiness verdict, then hand the output to your governance, semantic, and observability stack.

02

Peer benchmarking

We compare your warehouse idle rate, refresh cadence, and metric sprawl against comparable tenants. Your platform vendor, by design, can only see you.

03

Vendor-neutral grounding

The grounding pack conforms to the open semantic standard, so it is portable across engines and agents. You are never locked to one platform's idea of trust.

04

We are the "before"

We feed the catalog, the metric views, and the monitoring tools. We do not replace them. The diagnostic that gets you ready is the gap nobody else sells.

Marketplace and trust center

One product site, one technical review hub.

Brickcost is the product customers install and buy. Operating Graph AI is the broader platform story behind the roadmap. The Marketplace package, setup docs, connector contracts, MCP server notes, security posture, privacy terms, and support policy live at insights.sig.ai.

Use brickcost.sig.ai for buyer conversion and POC narrative. Use insights.sig.ai for Databricks Marketplace review, procurement, security diligence, and technical onboarding.
Data handling

Designed to clear security review on the first pass.

The diagnostic is built to minimize friction. Read-only, in your tenant, and honest about what it has and has not confirmed.

Read-only access. No production mutation, ever.

No raw business data required. Telemetry, not your tables.

SQL literals redacted, names masked to team level. By default.

The scanner can run entirely inside your workspace. Data stays in your tenant.

Least-privilege service principal. Scoped to system tables.

Every finding is evidence-linked. Labeled inferred until your owners confirm it.

The 2 to 4 week diagnostic

From access to an evidence-backed plan in a month.

Week 0
01

Setup

Confirm the workspace, system-table access, and target domain. Configure a least-privilege service principal or your own scanner.

Week 1
02

Diagnostic run

Ingest telemetry, build the Operating Graph, and produce the first findings and readiness scores.

Week 2
03

Validation workshop

Review the top findings with your data, governance, and business owners. Confirm what is real and what to ignore.

Weeks 3 to 4
04

Remediation plan

Final report, graph, backlog, and grounding pack. Choose two or three fixes to implement or measure next.

Fixed-fee diagnostic. Credit it toward an annual subscription if you convert.
What comes next

Today, Databricks. Later, commitment-aware placement.

Most teams run two or more platforms. Once the Operating Graph spans them, the same pass can show which net-new or uncommitted workloads belong on which engine, and where commitments are over-bought, without recommending moves that strand committed spend.

Databricks Snowflake BigQuery Redshift Azure
Start here

Start with a number, not a meeting.

Run the read-only scan inside your own workspace and get an evidence-backed estimate of recoverable spend. No data leaves your tenant, and the result doubles as the first page of your AI-readiness roadmap.