How it works

Connect → Learn → Compound

Two onboarding paths depending on whether you already have a data warehouse. Either path converges on the same place: an AI analyst that knows your data infrastructure intimately because, in many cases, it helped build it.

The three stages

Week 1

Connect

We connect the agent (via MCP) to your warehouse and surrounding tools — BigQuery, Snowflake, or Redshift; Shopify, ad platforms, Klaviyo, Slack, Gorgias, your 3PL. Read-only by default.

Week 2

Learn

The agent builds a knowledge graph of every important table, dashboard, and script you rely on, plus the quirks only your team knows. We sit with you for 2–3 sessions to seed this.

Week 3+

Compound

First weekly insight digest goes out. Two starter dashboards live. Slack channel open. From here, every question asked, every analysis run, every quirk discovered gets written back to the knowledge graph — and the agent compounds.

Two onboarding paths

Path A · Plug In

If you already have a warehouse

Connect → Learn → Compound runs on the timeline above. First weekly digest in week 3. No setup fee on top of the monthly retainer.

Path B · Build & Run

If you don't have one yet

Weeks 1–2 we map sources and design the warehouse schema. Weeks 3–5 we build pipelines and seed the knowledge graph. Week 5–6 full reporting layer live. The agent learns your business during the build, so by go-live it already knows your data.

Governance — the "will AI break things?" answer

A hybrid AI/algorithm model:

  • →Read-only work — running read queries to analyse data, building reports, answering analytical questions — the agent does fully autonomously.
  • →Sensitive actions — deploying code, scheduling jobs, modifying live data, sending external communications — the agent proposes; a Tahbas engineer reviews and approves before anything runs.
  • →The human review layer is part of the retainer. You don't need to dedicate someone on your side to babysit the agent — that's our job, included.
  • →The agent cannot autonomously trigger writes to your production systems. Architectural, not aspirational.