The AES AI Brain
One intelligent platform connecting knowledge across the business.
Enterprise Knowledge
- Project Documentation
- Risk Registers
- Technical Standards
- Programmes & Schedules
- Templates & Procedures
- Contracts
- Commercial Knowledge
- Meeting Minutes
- Lessons Learned
- Business Development
- Design Reviews
- Industry Best Practice
Business Outcomes
- Instant answers to complex queries
- Reports & technical documents
- Presentations & proposals
- Risk reviews & assessments
- Technical guidance & standards
- Decision support & recommendations
- Knowledge sharing across teams
- Meeting summaries & actions
The AES AI Brain is a practical enterprise capability, not a chatbot. It combines governed knowledge, organisational reasoning, agent assistance, and trust controls built on AES's existing Microsoft estate.
In plain language, the AI Brain is a controlled knowledge, reasoning, and agent layer over AES's existing Microsoft estate, project systems, and governance rules. It gives staff cited answers and useful drafts while approvals, access, residency, and audit remain governed.
What The AI Brain Is
| Capability | What it means for AES | Control built in |
|---|---|---|
| Governed Knowledge Platform | AES's important documents become a searchable corporate memory: proposals, contracts, policies, reports, drawings metadata, QMS content, and project knowledge. | Retrieval remains permission-trimmed, version-aware, label-aware, and source-cited. |
| Organisational Reasoning Layer | The Brain understands AES's operating model: who owns a project, who approves a decision, which entity is responsible, and which rules apply. | The LLM does not decide authority. DOA and policy checks are deterministic rules that agents must consult. |
| Governed Agent Workforce | Named agents help AES teams draft, coordinate, retrieve, and prepare work inside Teams and Outlook. | Each agent has scoped permissions, a charter, human approval gates, auditable actions, and a defined lifecycle. |
| Trust Fabric | The system defines where AI augments people and where automation is allowed, using DOA checks, approval workflows, AI disclosure, audit trail, evaluation, observability, and model-routing rules. | AES keeps humans in control, protects trust, and implements AI in an ethical and fair way before scale-up. |
How A Request Moves Through The Brain
- User asks: AES users ask a question, start a draft, or request an agent task in Teams, Outlook, or the Brain portal.
- Permissions filter: the system checks user permissions, labels, project compartments, and data-residency routing before retrieval.
- Sources retrieved: the Brain searches approved corpora, reranks the best evidence, and keeps source citations attached.
- Policy checked: DOA, disclosure, approval, and always-human rules decide whether the agent can draft, flag, route, or act.
- Output delivered: the user receives a cited answer, draft, task pack, or approval card, with audit trail and feedback captured.
Expected Outcomes
The outcome is not simply a configured AI platform. The AI Brain is designed to help AES turn existing knowledge, systems, and AI appetite into measurable business value while keeping people in control. ROI is anchored in Phase 0: current-state costs and effort — proposal drafting time, findability, rework — are baselined before anything changes, and value is reported against those baselines at each gate.
Before any agent moves into rollout, SYNRTECHS will support AES in building an agent-by-agent business case. Each business case will define the workflow problem, baseline effort, expected value, measurable KPIs, data readiness, control requirements, run-cost assumptions, owner sign-off, and the success threshold required to continue, scale, or stop.
Business Value AES Can Unlock
- Faster proposal and tender response: cited first drafts, compliance matrices, and reusable bid material, with human review before anything goes to a client.
- Knowledge no longer trapped in folders or individuals: approved policies, project records, proposals, and technical documents become findable with citations and permissions intact.
- Less coordination friction: agents prepare follow-ups, reporting packs, and internal handoffs so AES teams spend less time chasing information.
- Safer enterprise AI adoption: staff get a governed route that addresses hallucination, leakage, copyright, residency, auditability, and DOA concerns.
- A scalable AES-owned capability: the first pilots create the reusable Brain, control pattern, handover materials, and operating model for future agents.
Success Criteria - finalised at Gate 0
Success criteria will be finalised at Gate 0. At that point, AES and SYNRTECHS will confirm the priority agents, business case, baseline effort, available data, owners, control requirements, run-cost assumptions, and KPI thresholds for rollout.
| Success area | Gate 0 definition |
|---|---|
| Business relevance | The agent solves a clearly defined AES workflow problem with an identified business owner. |
| Measurable KPI | The agent has agreed KPIs, baseline values, target thresholds, and a data source for measurement. |
| Data readiness | The required corpus has an owner, approved source location, access boundary, quality threshold, and review cadence. |
| Control readiness | Required approvals, DOA boundaries, audit trail, permission trimming, and residency routing are defined before rollout. |
| Adoption readiness | Target users, champions, training needs, and feedback channels are confirmed. |
| Cost readiness | Run-cost assumptions, caps, and reporting cadence are agreed before usage scales. |
| Ownership readiness | AES has a named owner, runbook, handover pack, monitoring routine, and support model. |
How Value Is Measured
Value measurement is the evidence layer behind the success criteria. SYNRTECHS will support AES in measuring each agent against its approved business case before rollout and again before scale-up.
| Measurement area | What AES will measure |
|---|---|
| Workflow value | Time saved, cycle-time reduction, bid capacity, document retrieval speed, or reduced coordination effort. |
| Quality and trust | Citation validity, answer accuracy on golden questions, user review feedback, and reduction in unsupported outputs. |
| Financial value | Estimated monthly value from time saved or capacity gained, using AES-approved staff cost assumptions. |
| Adoption | Weekly active use, repeat use, department participation, and champion feedback. |
| Control performance | DOA routing, approval evidence, audit-chain completeness, permission trimming, and residency routing. |
| Cost discipline | Model usage, infrastructure consumption, support effort, and run-cost cap performance. |
| Decision evidence | Continue, scale, change, or stop recommendation for each agent based on measured results. |
Architecture Overview
The architecture is layered so AES can see what users experience, what agents do, what controls them, and which data foundations make the system reliable.
Design Principles
The solution uses what AES already owns, puts foundation work before agents, compartmentalizes sensitive content by client and regime, routes data by classification, uses the right model for the right task, keeps governance deterministic, requires citations for important outputs, and scales only through gates.
Five-Layer Architecture
L1 - Data Foundation
Connects AES's priority systems and content sources, including M365, Procore, Jisr, QMS, CAD/BIM metadata, regulatory references, licensed corpora, OCR, labels, and DLP.
AES meaning: approved knowledge becomes searchable, permissioned, version-aware, and ready for agent use.
L2 - Knowledge Platform
Turns prepared content into cited answers using search, reranking, permission trimming, source citations, a governance knowledge graph, and golden-question evaluations.
AES meaning: answers are grounded in current AES sources, not generic model memory.
L3 - Trust Fabric
Defines the right balance between augmentation and automation, with DOA policy, approvals, audit chain, disclosure, evaluations, and model routing keeping AI assistance inside AES governance rules.
AES meaning: humans stay in control of judgement, approvals, and sensitive decisions while AI is implemented in a fair, ethical, and trusted way.
L4 - Governed Agent Workforce
Launches agents as bounded digital workers with owners, allowed tools, data boundaries, autonomy levels, KPIs, run-cost caps, and handover packs.
AES meaning: agents are controlled work assistants, not open-ended bots.
L5 - Experience Layer
Brings the Brain into Teams, Outlook, M365 Copilot, and a thin Brain portal.
AES meaning: users search, draft, review, approve, and monitor agent work where AES already operates.
Model Strategy And Residency Routing
Model choice and data routing are governed by content class, contract obligations, geography, and sensitivity. Restricted content is hard-blocked from non-approved paths, while general corporate content can use the best approved model path for the task. KSA-sensitive routing is handled through classification and a future migration gate when required regional services are available and verified.
Build vs Configure Split
AES-owned Microsoft capabilities are configured first where they are sufficient: Copilot, Purview, SharePoint controls, Power Automate, Dataverse, and low-code utility agents. SYNRTECHS builds where AES needs capability beyond configuration: proposal-grade retrieval, the document refinery, evaluation harness, Retrieval API, flagship agents, DOA engine, audit chain, model router, and CAD/BIM findability pipeline.
Agent Roadmap
AES gets a practical agent roadmap: three priority pilots first, then a governed backlog that turns the wider agent vision into a controlled, incremental build-out.
Operating Model And Autonomy
Each agent ships with a one-page Agent Charter covering mission, owner, sponsor, scope of tools and data, least-privilege permissions, autonomy level, DOA interaction, disclosure behaviour, KPIs, run-cost cap, kill / continue criteria, and handover pack. Autonomy starts with draft-for-approval, can move to act-with-recorded-authorisation for approved task classes, and only reaches autonomous operation for internal, reversible, low-risk work after gate review. Client submissions, executive decisions, and DOA-reserved decisions remain human-owned.
| Autonomy level | How AES uses it |
|---|---|
| L1 - Draft-for-approval | The agent prepares work; AES reviews, approves, and sends. |
| L2 - Act-with-recorded-authorisation | The agent performs pre-approved task classes; every action is logged and exceptions are escalated. |
| L3 - Limited autonomous operation | Used only for internal, reversible, low-risk work after gate review. |
| Never delegated | Client submissions, executive decisions, employment decisions, and DOA-reserved decisions remain human-owned. |
Priority Agents - additional detail
| Priority agent | Basic role | Phase / autonomy |
|---|---|---|
| Proposal & Bid Writer | Turns tender intake into a structured bid workspace and prepares cited first drafts for human review. | Phase 2 / L1 draft-for-approval |
| Project Coordination & Liaison | Prepares coordination packs, reporting inputs, follow-ups, and draft correspondence from project signals. | Phase 3 / L2 act-with-recorded-authorisation |
| HR Assistant | Answers HR policy questions, drafts routine materials, and supports approved HR service workflows. | Phase 3 / L1 draft-for-approval |
Proposal & Bid Writer - additional detail
The Proposal & Bid Writer supports RFP intake, requirements and compliance matrices, precedent retrieval, CV and past-section reuse, cost benchmark lookup where licensed, cited drafting in AES's voice, reviewer workflow, and proposal knowledge-base reuse. The boundary is clear: the agent drafts and prepares; AES reviews, approves, and submits. Rollout evidence will focus on first-draft time, citation validity, compliance-matrix completeness, user satisfaction, and human review.
Project Coordination & Liaison - additional detail
The Project Coordination & Liaison agent monitors coordination signals across Procore, Teams, and mailbox flows; prepares action summaries, reporting packs, reminder queues, and draft correspondence; and flags escalations. External messages, commitments, and DOA-sensitive actions remain recorded, reviewable, and human-approved where required. Rollout evidence will focus on reporting-pack assembly time, open-action cycle time, coordinator hours saved, and zero unauthorised external sends.
HR Assistant - additional detail
The HR Assistant answers policy questions with citations, drafts routine letters and onboarding materials, supports approved Jisr lookups, and prepares onboarding or offboarding checklists. It does not screen candidates, rank employees, make employment decisions, approve sensitive matters, or decide exceptions. Rollout evidence will focus on HR owner sign-off, adoption by Shared Resources, citation quality, and PDPL (Personal Data Protection Law)-scoped data handling.
Future Agent Backlog
All remaining roles are held in a prioritised backlog and scored by value evidence, data readiness, DOA or regulatory exposure, sponsor pull, implementation complexity, expected KPI, and run-cost view. Many future roles can be thin declarative or Copilot Studio agents over the same Brain rather than bespoke builds.