Implementation

Implementation Principles

• Gate before scale: each phase proceeds only when the required owners, data, controls, evidence, and acceptance checks are in place.
• Co-build with AES: AES is embedded from day one so operating knowledge is built during implementation, not transferred only at the end.
• Business case per agent: every agent rollout is tied to a defined workflow problem, baseline, KPI, owner, control requirement, run-cost view, and continue / scale / stop rule.
• Governed foundation first: broad retrieval and agent action do not start until permissions, data readiness, labels, routing, baselines, and source ownership are confirmed.
• Foundation first, whatever the option: AES can stop at Gate G1 with a working governed Brain in place before any agent scale-up proceeds.
• Scale-Up Review at Week 8: measured early data volumes, retrieval results, Proposal Writer evidence, user evidence, and agent priorities confirm the next-stage agent specifications within the selected option — and give AES the evidence to extend from Option 1 to Option 2 or 3 if desired.
• Handover by launch: every agent launch includes documentation, monitoring guidance, evaluation evidence, and AES owner sign-off.

Implementation Phases

Phase 0 - Mobilise & Foundations

Timing: Weeks 1-2 Commitment status: Included in all options (1–3) Goal: De-risk the programme before broad retrieval or agent pilots begin.

Build focus

• Confirm AES owners, users, entity map, real headcount, M365 tenant geography, licensing, and system access.
• Prepare Azure workspace, controlled environments, CI/CD, secrets, observability, and audit logging.

AI / data requirements

• Confirm priority corpora, source-of-truth owners, permissions, Purview DSPM (Data Security Posture Management), sensitivity labels, and DLP (data loss prevention) baseline.
• Create contract and regulatory routing rules for GACA, government, NCA (National Cybersecurity Authority), SSI (Sensitive Security Information), NDA, and privacy obligations.
• Start oversharing remediation, DOA codification, and KPI baselines before any pilot claims value.

Milestones / outputs

• Mobilisation pack, owner map, and system-access plan.
• Tenant geography confirmation, entity map, Jisr module / API audit, QMS integration workshop, and Procore contract check.
• Contract-clause register, label taxonomy, data-readiness checklist, baseline dashboard, and approved pilot corpus candidate.

Gate G0 - Week 2: proceed only when tenant geography, contract register, label taxonomy, remediation plan, baseline measures, pilot corpus, and build environment are approved. The programme does not proceed to broad retrieval without this passing.

Phase 1 - Knowledge Platform MVP

Timing: Weeks 2-6 Commitment status: Included in all options (1–3) Goal: The Brain exists: governed search that works, with citations, on an approved pilot corpus.

Build focus

• Ingest the pilot corpus: proposals archive, two to three flagship project libraries, policies, QMS, and HR policy content.
• Build the document refinery: OCR, parsing, chunking, metadata, and citation handling.
• Build retrieval API, reranking, and Brain portal / Teams bot experience.

AI / data requirements

• Verify ACL-trimmed search, Purview label enforcement, and no permission bleed.
• Run golden-question evaluation for answer quality, citation quality, and source freshness.
• Model org chart, RACI, projects database, and digitised DOA as governance context.

Milestones / outputs

• Working Brain MVP with 20-30 governed pilot users.
• Retrieval-quality report and findability baseline comparison.
• Copilot enablement wave 1: prompt patterns, training, and starter declarative agents on approved scopes.

Gate G1 - Week 6: golden-set retrieval reaches at least 85% top-5 hit rate, citation checks pass, zero ACL-bleed findings, findability uplift is demonstrated, and steering signs off to build Agent 1. AES can stop here with a working, governed Brain in place.

Phase 2 Early Start - Proposal Writer Design Sprint

Timing: Weeks 4-6 Commitment status: Included in all options (1–3) Goal: Start Proposal & Bid Writer design and draft-pilot preparation before Gate G1, so momentum carries straight through the gate with no idle weeks.

This early-start sprint prepares the Proposal Writer backlog, RFP intake pattern, reviewer workflow, draft-pilot setup, and trust-fabric requirements. Build-to-production continues only after the next-stage scope is confirmed.

Phase 2 - Proposal Writer Agent + Trust Fabric v1

Timing: Weeks 6-8 Commitment status: Included in all options (1–3) · compressed track, finalised by Week 8 Goal: Build the Proposal & Bid Writer to production and harden the Trust Fabric that every later agent reuses.

Build focus

• Build RFP intake, requirement shredder, compliance matrix, precedent retrieval, sectioned drafting, and reviewer workflow in Teams.
• Implement agent identity blueprint, Conditional Access pattern, and standing-identity fallback where needed.

AI / data requirements

• Implement Trust Fabric v1: OPA (Open Policy Agent) DOA checks, approval cards, correlation-ID audit chain, WORM (write-once, read-many) export, disclosure footers, model-routing rules, and standing agent identity pattern.
• Run quality clinics on real bids with citation checks, human review, and prompt / retrieval tuning.
• Run the live pilot on real bids shadow-first, then assisted.
• Track usage, run cost, exceptions, and reviewer feedback from the first live pilot.

Milestones / outputs

• Proposal Writer pilot live on selected proposal work.
• Compliance matrix and cited first-draft workflow tested with AES reviewers.
• Trust Fabric evidence pack: approvals, audit trace, disclosure, and evaluation results.

Scale-Up Review

Timing: Week 8 Status: Scope-confirmation checkpoint, not a gate

The Scale-Up Review sits at the end of compressed Phase 2, once the Brain MVP, Proposal Writer pilot evidence, and Trust Fabric evidence have landed. Measured data volumes, retrieval performance, adoption signals, and agent priorities confirm the specifications and sequencing of the next-stage blocks already covered by the selected option. It is also the natural point for AES to extend commitment from Option 1 to Option 2 or 3. AES confirms, defers, or stops each subsequent block; pricing is already fixed by the selected option.

Gate G2 - Week 8: first-draft time is at or below two hours on at least five real proposals, citation validity reaches the agreed target, bids-team adoption is positive, Trust Fabric v1 is evidenced, and the MD demo is complete. A kill / continue decision is recorded before further agent build.

Phase 3 - Agent Workforce

Timing: Weeks 8-20 Commitment status: Included in Options 2 and 3 · agent specifications confirmed at the Week 8 Scale-Up Review Goal: Extend the platform to the HR Assistant and Project Coordination & Liaison agents with measured value and controlled actions. Delivery is progressive: the first agent pilot is live in controlled workflows by Week 14, the second by Gate G3, each with its own scorecard — not a single release at Week 20. The 12-month structure and gate sequence are fixed; start dates and resourcing of individual blocks are confirmed with the AES team during mobilisation, depending on resource availability and requirements.

Build focus

• Build Project Coordination & Liaison: Procore integration, reporting packs, action chasing, and controlled external draft flows.
• Build HR Assistant: Jisr API integration, policy Q&A, template drafting, and HR service guidance.
• Reuse the Phase 2 trust fabric rather than re-hardening it for each agent.
• Expand approved knowledge scope through scoped Exchange / Teams ingestion and a CAD / BIM metadata pilot decision.

AI / data requirements

• Confirm tool permissions, action boundaries, approval routing, and external-communication guardrails for each agent, as specified at the Scale-Up Review.
• Validate PDPL (Personal Data Protection Law) handling for HR content and privacy-sensitive flows.
• Prepare ISO 42001 AIMS materials, internal audit dry-run, and per-agent monitoring model.

Milestones / outputs

• Approved priority agents live in controlled workflows.
• Per-agent scorecards covering adoption, quality, run cost, exceptions, and human-review evidence.
• Utility-agent backlog sorted into thin agents, integration-heavy agents, and future work orders.

Gate G3 - Week 20: Project Coordination has zero unauthorised external actions, HR Assistant is adopted by its named owner, per-agent KPIs are green, run cost is within model, and a kill / continue decision is recorded before Phase 4 scale-out.

Phase 4 - Scale & Institutionalise

Timing: Months 5-12, starting at Gate G3 (Week 20) Commitment status: Included in Option 3 · scope confirmed at the Week 8 Scale-Up Review Goal: Harden successful pilots into an AES-owned operating capability.

Build focus

• Harden successful pilots for broader production use with monitoring, support model, release discipline, and run-cost controls.
• Activate agreed office waves and expand corpus coverage where readiness, permissions, and economics support it.
• Score future agent backlog quarterly, with market-intelligence agents held until licensed-data approval.

AI / data requirements

• Finalise operating runbooks, admin training, incident process, quality-monitoring routine, and model / prompt change-control process.
• Review the KSA enclave migration once Azure Saudi Arabia East reaches general availability and required services are verified.
• Prepare ISO 42001 certification-readiness and EU AI Act evidence pack where relevant.
• Confirm ownership, data refresh rules, support responsibilities, and quarterly value-review rhythm.

Milestones / outputs

• Production-ready releases for approved agents.
• AES handover pack: runbooks, monitoring guidance, admin training, evaluation sets, and backlog governance.
• Year-1 value report and year-2 implementation plan for MD review.

Completion checkpoint: production acceptance requires owner sign-off, handover pack, monitoring in place, support model confirmed, security / compliance evidence complete, and run cost within approved thresholds.

Delivery Team & Ways Of Working

AES's named team is embedded from day one, so the operating knowledge is built during implementation rather than bolted on at the end.

Delivery Model

Delivery Team

Delivery is led by SYNRTECHS's founding partners with named specialist advisors, embedded alongside AES's own workstream, engineering, adoption and innovation leads. Every role below is named and accountable. Programme contact: Khaled Hourani — khaled@synrtechs.com · +966 50 430 1529.

The model stays lean and co-owned: SYNRTECHS leads the architecture and build, AES provides the embedded workstream and operating knowledge, and every launch includes review, evaluation, documentation, and handover discipline.

Co-Development & Full Handover

This is not an outsourcing model. SYNRTECHS builds with AES, transfers operating knowledge as each agent goes live, and leaves AES with the assets, documentation, and control needed to own the platform.