Delivery controls · six gated increments

Production delivery board

Platform and AI product work run in parallel. Each increment has an accountable owner, prerequisite and testable exit criterion; unmet gates stop promotion.

1. M1

   Increment

   Baseline and prove access

   Data platform

   Profile all 16 systems; classify fields; prove read-only connectivity; define customer, premise and asset keys; benchmark Fabric and Databricks with the same SAP/GIS extract.

   AI products

   Select two bounded workflows; capture four-week baseline, review volume, handling time, error rate and Finance-approved value equation.

   Accountable

   Programme architect + CISO delegate

   Entry dependency

   Read-only accounts; source SMEs; change windows

   Exit gate

   100% owners named; 16/16 source contracts drafted; benchmark and threat model signed

2. M2

   Increment

   Build controlled foundation

   Data platform

   Provision dev/test/prod; private connectivity; identity groups; secrets; catalogue; CI/CD; Delta landing; schema registry; dead-letter and replay framework.

   AI products

   Build two proofs with masked snapshots; instrument acceptance, override, latency, cost and failure paths.

   Accountable

   Platform engineering lead

   Entry dependency

   Platform decision; network and identity approval

   Exit gate

   IaC repeatable; zero critical security findings; contract tests block invalid data

3. M3

   Increment

   Production core feeds

   Data platform

   Operate SAP, GIS and HEAT ingestion; resolve customer/premise/asset identifiers; publish first conformed products and reconciliation dashboard.

   AI products

   Run one controlled pilot in a representative GA; retain human approval and a baseline cohort.

   Accountable

   Data platform lead + source owners

   Entry dependency

   M2 controls; SAP/GIS/HEAT extraction approval

   Exit gate

   P95 freshness within contract; ≥99% valid keys; control totals reconcile within 0.5%

4. M4

   Increment

   Certify metrics and release

   Data platform

   Add DMA, QAM, SCADA and Spot Billing; certify connection, work, incident, billing and throughput measures with lineage and row-level security.

   AI products

   Release the first quick win only if error and value gates pass; start service-request and HSE triage pilots.

   Accountable

   Data product owners

   Entry dependency

   Core entity keys; metric sign-off; support rota

   Exit gate

   Certified measures reconcile within 0.5%; access tests pass 100%; first service has runbook

5. M5

   Increment

   Evaluate Company Brain

   Data platform

   Deploy permission-aware retrieval and governed text-to-SQL beta; enforce citations, policy filters, query limits, abstention and audit logging.

   AI products

   Shadow-test one integrated model selected by data readiness: conversion, maintenance or leakage.

   Accountable

   AI product lead + security

   Entry dependency

   Certified models; approved corpus; evaluation set

   Exit gate

   ≥90% grounded answers on approved test set; 0 unauthorised disclosures; model value validated

6. M6

   Increment

   Handover and scale decision

   Data platform

   Enable service monitoring, cost allocation, backup/restore, quarterly access review, incident runbooks and ownership transfer; execute recovery test.

   AI products

   Audit realised value; scale products that clear gates; stop or redesign products that do not.

   Accountable

   Operations lead + executive sponsor

   Entry dependency

   Support model; benefit evidence; residual-risk acceptance

   Exit gate

   RTO/RPO test passed; ≥99.5% service target demonstrated; benefit owner accepts audit