Phase 01 · Readiness
Readiness assessment
One week. Map data quality, access controls, and system architecture against a realistic AI use case.
- Data quality scorecard
- Access + policy review
- Use-case shortlist
- Go / no-go recommendation
AI Consulting
AI integration for teams that want to move — but need a safe starting point.
For companies that are AI-curious but hesitant to execute, we scope practical integrations tied to real workflows, risk controls, and measurable outcomes. We can lead end-to-end, co-deliver with your existing team, or ship an AI project your team can use and operate with confidence.
Free first-hour consultation
One hour. Two 30-minute sessions. Real work between them.
Most "free consultations" are 60-minute sales calls. Ours isn't. The hour is structured so you walk away with a tailored deliverable — a use-case shortlist, a risk register, and a go / no-go decision — not a recycled deck.
- 0130 min · live
Lay of the land
You walk us through your stack — data, design, access. Schema and a few PII-stripped sample rows is all we'll ever take away.
- 02async · we work
We build the analysis
We synthesize a representative dataset from your schema and ship a tailored notebook — no production data leaves your perimeter.
- 0330 min · live
Tailored walkthrough
15-min presentation of the notebook, 15-min discussion. You leave with a scoped pilot proposal — or a clear "not yet".
After Session 2: you're free to walk. No nurture sequence, no back-and-forth chase. If you want to engage further we scope a paid pilot; if you don't, the notebook stays with you.
Claim the hour →How we work
Three phases, scoped before any code is written.
We don't sell pilots that never reach production. Every AI engagement is scoped against a defined production rollout from day one — even if the pilot fails the evaluation framework and gets killed.
Phase 02 · Pilot
Pilot build + evaluation
4–6 weeks. Build a scoped pilot against a single workflow. Evaluation framework + governance guardrails included.
- Pilot system build
- Evaluation harness
- Risk + compliance review
- Go / no-go for production
Phase 03 · Production
Production rollout
4–8 weeks. Hardening, observability, on-call setup. Handoff to your team with runbooks and a 30-day office-hours window.
- System hardening
- Observability + on-call
- Enablement session
- 30-day office hours
Use cases we've shipped — or scoped to ship.
Practical, workflow-tied AI integrations. We avoid demoware. Every use case below maps to a measurable outcome — throughput, ticket resolution, time-to-answer, or cost reduction.
Internal Copilot
Analyst copilot over your governed semantic layer
An LLM-based assistant that answers metric questions in plain English by routing through your certified semantic entities — not raw warehouse SQL.
Use cases: ad-hoc analytics · metric Q&A · report drafting
Workflow Automation
Pre-merge data-PR review with policy guardrails
An LLM agent reviews dbt PRs for lineage impact, policy violations, and breaking-change risk before they reach a human reviewer. Cuts review queue time >60%.
Use cases: dbt CI · lineage gating · policy enforcement
Customer-Facing
Embedded analytics chat for your product
A scoped, governed chat surface inside your product that lets customers ask questions of their own data — behind your row-level security and your usage metering.
Use cases: SaaS embedded analytics · self-serve reporting
Pilot duration
4–6 weeks
Production rollout
4–8 weeks
Pilot price
From $8,500 fixed
First hour
Free, no fee
Governance guardrails
What every AI engagement ships with.
- Evaluation harness with measurable production-readiness criteria
- PII + data-egress policy review
- Row-level security verified end-to-end
- Audit log + replay capability for every model call
- Kill-switch and graceful-degradation paths
- Cost ceilings + rate-limit guards
- Model-version pinning and rollback plan
- SOC2 / HIPAA / GDPR posture documentation
Curious but not yet committed? Start with the free hour.
30–60 minutes with a senior engineer. We'll map your use case to a realistic production path or tell you it isn't one yet.