About/Methodology
Methodology

How we deliver, end-to-end.

The Foundry Cortex deployment methodology. Audit-led, phased, vendor-neutral. From the 14-day readiness audit to production deployment in 16-24 weeks, with measurable recovery within 90 days of model go-live.

The four phases.

Phase 1 — Audit (Weeks 1-2)

14-day Readiness Audit. Sensor coverage mapped, data quality scored, ROI quantified per asset class, vendor stack recommended. Board-ready executive brief delivered at the day-14 readout.

Outcome: go/no-go decision with full visibility on cost, timeline, and projected recovery.

Phase 2 — Data foundation (Weeks 3-10)

Historian deployment or extension. Sensor estate retrofitting on top-decile critical assets. IT/OT integration bridge between PI and the modelling environment. CMMS data quality cleanup.

Outcome: clean industrial data flowing from sensors to model substrate within 6-8 weeks of contract signature.

Phase 3 — Predictive + twin (Weeks 9-20)

XGBoost / LightGBM models on top-decile assets first. Anomalib vision-AI on critical inspection points. Chronos forecasting layer once foundation is mature. Industrial twin scope-bounded to single mill or line.

Outcome: first true-positive interventions captured within 90 days of model go-live. Recovery measurement begins.

Phase 4 — Agentic + handover (Weeks 16-24)

LangGraph or SymphonyAI IRIS Flows agent layer on top of predictive forecasts. Auto-generated work orders, technician dispatch, exception escalation. 4-week shadow mode followed by production autonomy. Operational handover to client team.

Outcome: 70-85% of routine maintenance work orders flow autonomously. Ongoing managed-service retainer takes over the optimisation work.

Beyond the four phases.

Phase 5 and beyond — humanoid robotics deployment, multi-site scaling, advanced agentic operations, custom Cosmos workflows — happens on a per-operation cadence. The four phases above are the standard playbook; Phase 5+ is bespoke and scoped against the operational maturity that Phase 4 produces.