Decisions from data, AI where it earns its keep.
Most early-stage data work is either over-built — modern data stack for ten people — or under-built: gut-feel decisions on a product that already has signal. We build the version that fits today, with a clear path to the version that fits next year.
The moments this engagement was built for.
Founders usually call us when they have data they can't answer questions from, or a customer-facing AI feature that needs to actually work.
You can't answer “how is the product doing” in under five minutes.
An event schema, a warehouse, and the three dashboards leadership actually checks. The other twenty arrive when they're asked for.
Your sales team is asking for cohort and retention numbers you don't have.
Real cohort analytics, retention curves, and the unit economics view that closes a series A round.
You're adding an AI feature and don't want to ship a demo.
An honest read on what the model can and can't do, evals you trust, and a fallback plan when the model is wrong.
Your event tracking is duct-taped and nobody trusts it.
A clean event schema, a tracking plan, and the discipline to keep both up to date as features ship.
You need a customer-facing report or scoring model.
Pipeline, model, evaluation, and the production wrapping — built once, monitored, with a path to retrain.
The smallest data stack that earns its keep.
We don't install a modern data stack for the sake of it. We pick the smallest set of tools that solves your problem, and document the upgrade path.
- An event schema and tracking plan. Versioned in Git, reviewed at PR time, with a single source of truth for what every event means.
- A warehouse with the basics done right. Postgres or BigQuery, dbt models for the transforms, scheduled refreshes that don't silently break.
- The three dashboards that matter. A leadership view, a product view, a revenue view. Built in a tool the team will actually open.
- AI features that pass an honest eval. A small eval set built from your real customers, with a number we can defend before we ship.
- A cost ceiling on every pipeline. Warehouse, vector DB, model calls — instrumented, capped, alerted. No surprise invoices.
- Documentation a non-engineer can read. What each event means, what each dashboard is for, what to do when a number looks wrong.
Advisory to scope. Build to ship.
We almost always start with a one-week Advisory scope: read what's there, listen to the questions you want answered, and write a short plan that picks the smallest stack that solves it.
Build engagements typically run six to twelve weeks for a first version: tracking plan, warehouse setup, three dashboards, optional AI feature, and a clean handover to your team. We expect you to want changes after week four — that's normal, not a problem.
For AI work specifically: we won't ship anything we can't evaluate, and we won't leave you with a feature that has no fallback. If a model can't hit the bar, we say so before the demo.
Cost-based. Cash, equity, or a mix.
We price engagements at cost plus a small margin — designed to be affordable at zero revenue, with cash, equity, or hybrid structures available. Final shape gets agreed in the discovery call.
Data work usually pulls in the rest of the stack.
Once data is flowing, the next conversations are usually about product surfaces that read it, infrastructure that runs it, and the security of what you're collecting.
Product Engineering
Ship MVPs and production systems that don't need a rewrite at series A.
Read more →Security & Compliance
Enterprise-ready before your first enterprise deal.
Read more →AI-powered study planner
Mobile-first adaptive planner — AI reads where the aspirant is, what they struggled with, and what's next, then plans accordingly. Scheduling, not magic.
Read story →Tell us the question you can't answer. We'll tell you the stack.
A 30-minute call. We listen, we ask, we sketch the smallest data setup that answers the questions that actually matter.
Book a discovery call