Delivery methodology
Discover (4w) → Build (12–26w, 2-week sprints) → Harden (4–6w) → Run. The methodology is named, written, gated and audited — designed to survive a PMO audit and a TDRA / CBUAE / FSRA review.
Discover
4 weeks
Build
12–26 weeks · 2-week sprints
Harden
4–6 weeks
Run
90-day hyper-care · continuous
G0
G1
G2
G3
Kick-off
Production-readiness mid-Build
Go-live approval
Hyper-care exit
What this methodology refuses to do
Most AI programmes fail at the methodology level long before they fail at the model level. A 14-month strategy phase produces two decks and zero shipped models. An Agile sticker on a waterfall plan produces governance theatre. An open-ended T&M arrangement produces an unbounded burn-rate that the PMO eventually has to explain to the audit committee.
We will not run a strategy phase longer than four weeks. We will not call a ceremony Agile if it is not. We will not contract T&M without a written exit ramp. And we will not declare a system production-ready without a signed regulator-evidence pack at G3. The methodology below is what we will commit to in writing.
Four phases, in order
Each phase ends in a named gate. Each gate produces a regulator-evidence pack version.
Phase 1
Stakeholder map, use-case scoring, data-readiness audit, model-feasibility spike, reference architecture, SoW drafting, G1 stage-gate review.
Closes at G1 — feasibility memo signed
Phase 2
Sprint-zero (CI/CD, registry, eval harness), iterative build, demo every second Tuesday, RAID every second Thursday, monthly steering. Mid-build production-readiness review at G2.
G2 production-readiness mid-Build
Phase 3
UAT, evaluation report, bias and fairness pack, pen-test, load test, regulator dry-run, hand-over rehearsal. Closes at G3 — go-live approval.
G3 — regulator dry-run cleared
Phase 4
Hyper-care SRE rota, drift alerting, monthly knowledge-transfer milestones, quarterly model-review board. Exit criteria committed at G3.
90-day hyper-care, then named on-call
Phase deep-dives
4 weeks (fixed)
Week 1: stakeholder map and use-case scoring against the Brocode value-risk matrix. Week 2: data-readiness audit and model-feasibility spike. Week 3: reference architecture and SoW drafting. Week 4: stage-gate review with the client steering committee.
Deliverables
12–26 weeks, 2-week sprints
Product-owner pair (Brocode delivery lead + client product owner), sprint demo every second Tuesday, RAID review every second Thursday, monthly steering. Sprint-zero stands up dev environment, CI/CD, MLflow registry, eval harness and regulator-evidence template.
Deliverables
4–6 weeks
Dedicated UAT, model-evaluation report, bias and fairness pack, security and penetration test, performance and load test, regulator dry-run, hand-over rehearsal.
Deliverables
Continuous, 90-day hyper-care
Managed-service ramp with a fixed 90-day hyper-care, named SRE on-call rota, model-drift alerts wired to the client ticketing system, quarterly model-review board.
Deliverables
The four stage gates
Every gate review is minuted, signed, and attached to the regulator-evidence pack version.
Engagement charter signed, sponsor confirmed, PMO-Pack handed over (22 templates), data-access provisional approvals lodged with the second line of defence.
Feasibility memo + target architecture + SoW + RACI signed off by steering. Regulator-evidence pack v1 — initial control mapping. Go / No-Go committed in writing.
Eval harness baseline + first regression suite + initial bias report + observability plane live. Regulator-evidence pack v2 — interim audit-ready evidence. Internal-audit dry review optional at the client's request.
UAT sign-off + final model-evaluation + bias + pen-test + regulator dry-run minutes + hand-over rehearsal complete. Regulator-evidence pack v3 — final, attached to the regulator submission.
96%
On-time at go-live since 2023
4
Named gates: G0, G1, G2, G3
22
PMO-Pack templates handed over at G0
4w
Fixed Discover phase
Sample sprint plan
Sprint demo cadence, RAID rhythm, monthly steering, and the G2 mid-Build production-readiness review.
| Sprint | Demo | RAID | Steering |
|---|---|---|---|
| Sprint-zero | Dev environment + CI/CD + registry + eval harness | Initial risk log; data-access blockers flagged | Sprint-zero exit review |
| Sprint 1 | First end-to-end vertical slice on a hold-out sample | Model-feasibility risks updated | — |
| Sprint 2 | Eval suite + baseline metrics on the use-case KPI | Bias-class hypotheses logged | Monthly — sponsor present |
| Sprint 3 | Inference path + observability + drift baseline | Production-readiness checklist opened | — |
| Sprint 4 | Integration with client system-of-record + UAT-readiness | Pen-test scope confirmed | G2 mid-Build production-readiness review |
Five delivery patterns
Each starter is editable on day one and pre-mapped to the G0–G3 gate evidence.
Pattern
Net-new use case with no existing ML in scope; product owner + delivery lead pair from day one.
Pattern
Replacing an incumbent model; parallel-run RACI with safe-cut-over playbook.
Pattern
Retrieval-augmented platform with a corpus owner, a vector-store owner and a guardrails owner explicitly named.
Pattern
Multi-tool agentic workflow with a tool-owner matrix and a separation-of-duties model for action authorisation.
Pattern
MLOps control plane (registry, eval, drift, lineage) deployed across multiple model teams.
Versus the alternatives
| Capability | Brocode | Big-4 AI practice | Offshore consultancy | Boutique Agile shop |
|---|---|---|---|---|
| Published, named stage gates (G0–G3) with deliverables | No | Internal only | In bespoke decks | |
| Fixed-price option out of fixed 4-week Discover | Rare | T&M only | T&M by default | |
| PMO-ready RACI starters (5 patterns) | On request, weeks later | |||
| Regulator-evidence pack at every gate | TDRA / CBUAE / FSRA / NCA pre-mapped | Built per engagement | Built per engagement | TDRA only |
| On-time at go-live since 2023 | 96% | ~38% industry baseline | ~45% industry baseline | ~50% industry baseline |
| Named Head of Delivery on every engagement | Pool of project managers | Partner-led, rare in delivery | Senior on kick-off only |
Tooling backbone
Sprint board
Jira or Linear
ADRs
Confluence or Notion
Engineering work
GitHub Projects
Model lineage
MLflow 2.16
Architecture
Lucid or Mermaid
PMO-side view
Asana or MS Project
Free download
A downloadable ZIP with the SoW template, change-control template, RAID register, stage-gate checklist (G0–G3), regulator-evidence template, hand-over checklist and 16 more — all in Microsoft Office and Confluence-compatible formats.
PMO questions
Real Agile inside a phase, with phase-level gates that procurement and internal audit can recognise. Inside Build, sprints are 2 weeks with a demo every second Tuesday, a RAID review every second Thursday and a monthly steering. Scope change is handled through a unit-priced change-control framework agreed at G1, not by re-baselining the whole programme. Hardening is a separate phase, not a tail of Build — which is the most common mistake we see in failed AI programmes.
Book the walk-through
We open the templates on screen, walk a real sample evidence pack (redacted), and answer the questions internal audit is going to ask you.
Prefer chat? Message us on WhatsApp.
Continue exploring
The discipline behind the Harden phase.
Read moreThe pinned stack the Build phase delivers on.
Read moreThe operating backbone for the Run phase.
Read moreThe named Head of Delivery and senior leads.
Read morePricing and timeline objections answered.
Read more