How an engagement actually runs
Most Forecasta engagements run 3–6 weeks, fixed price and fixed scope. This is the walkthrough of a typical sprint.
Scope lock and first deliverable
We confirm the scope, data access, and success criteria in writing before any building starts. By the end of week one, you see a named, working deliverable — a working retrieval pipeline on a sample of your documents, a first backtest of a forecasting model, a draft agent workflow — not a plan. You should never wait two invoices to see output.
Build against the real thing
Work moves from the sample to your actual data and systems. We run a short check-in — usually 20 minutes, not a status meeting — at the end of each week: what shipped, what we found, what changes next. You get access to whatever is running so far; nothing is held back until the end.
Validation, monitoring, and handover
Before we call anything done, it's validated against a held-out test — a backtest for a forecasting model, an evaluation set for a retrieval system, a review sample for an agent's decisions. Monitoring and alerting go in before launch, not after. Handover includes documentation a future engineer, ours or yours, can actually follow.
What we need from you
Access to the relevant data and systems, and one point of contact who can answer questions within a day. That's it — we don't need a dedicated project manager on your side.
If scope changes mid-sprint
We tell you immediately, with the cost and timeline impact stated plainly, and agree a change before continuing. We don't absorb scope creep silently and we don't spring a surprise invoice at the end.
What happens if it's wrong
Every deliverable is validated against a held-out test before we call it done. If something doesn't hold up, we say so and fix it before handover — not after you've found it in production.