Operating Infrastructure Co.Client Engagement
v1.0
OPS-ENG-2026-01
May · 2026
Engagement & System Architecture
From audit to operating system, in 90 days — for knowledge-work SMBs.
ProcessArchitectureGovernanceCadence
What we install for knowledge-work SMBs ready to make AI usable across the company — in what order, with whom, against which deliverables. A single page the client, the operator, and the engineering team can read together.
Engagement process, six phases, audit to managed ops
PHASE 01
Intake & Scope
Week 0, 3 days
OutputsDiscovery brief, executive alignment, signed audit SOW
PHASE 02
Infrastructure Audit
Weeks 1, 2 to 3
OutputsData & tool inventory, workflow shortlist, governance read, ROI sizing
PHASE 03
Roadmap & Buildout SOW
Week 4, 1 week
OutputsPrioritized roadmap, three buildout options, signed buildout SOW
PHASE 04
Buildout Sprint
Weeks 5, 6 to 12
OutputsConnectors live, SOPs encoded, retrieval & agents wired, governance scaffold
PHASE 05
Launch & Adoption
Weeks 13, 14
OutputsRunbook, training, internal launch comms, 30-day usage targets
PHASE 06
Managed Operations
Ongoing, monthly
OutputsTuning, monthly report, incident triage, QBR every 90 days
System architecture, what we install end-to-end
Layer 01
Data Layer
Connect, clean, classify, and permission the client's existing knowledge so AI can find and use it safely.
Source connectors
M365, Workspace, CRM, drives, email, tickets
Document processing
Parse, chunk, classify — brand books, specs, methodology, prior work
Knowledge index
Vector + hybrid search, re-rank
Permission mirror
Source ACLs mapped to metadata, per-department
Master records
Customers, products, brand voice, methodology, approved messaging
Source-of-truth registry
Authoritative system per data class, with named owner
Buyer promise
"AI can find our information — and every department draws from the same reference."
Example stackM365 / Workspace, HubSpot / Salesforce, Notion / Confluence / Figma, QuickBooks / Xero, vector DB
↓↓↓
Layer 02
Brain Layer
Encode how the company actually works — SOPs, decision rights, brand voice, approval flows, and rules that turn retrieval into reliable judgment.
SOP & decision rules
Encoded operating logic, who owns what, escalation
Brand voice & messaging
Approved tone, claims, do-not-say list
Approval flows
Cross-functional review routing, sign-offs
Prompt & agent library
Templated prompts, tools, roles
Guardrails
Tone, scope, safety, refusal rules
Memory & context
Session, account, conversation state
Buyer promise
"AI knows our rules — who owns what, and what counts as authoritative."
Example stackOpenAI / Anthropic / Google, orchestration framework, rule engine, prompt versioning
↓↓↓
Layer 03
Output Layer
Deploy AI inside the tools the team already uses, so adoption happens without a new app or login.
Email integration
Inbox triage, drafts, send approval
CRM workflows
Activity creation, draft proposals, gates
Chat surface
Teams / Slack bot, ask-the-business
Document drafting
Word / Docs / PDF generators
Ticketing & ops
Auto-classify, route, propose response
Reporting dashboard
Usage, ROI, incidents, drift
Buyer promise
"My team uses it without learning another system."
Example stackOutlook / Gmail, HubSpot / Salesforce, Teams / Slack, Notion / Linear / Jira, Zapier / Make / n8n
Governance scaffold
- Role-based access mirrors source ACLs
- Source-of-truth rules per data class
- Human approval gates on external actions
- Audit log of every AI invocation
- Sampling & review policy in first 90 days
Standard operating rhythm
- Weekly engagement standup & status note
- Bi-weekly sponsor sync & in-progress demo
- Monthly ops report & incident review
- Quarterly business review, roadmap & expansion
Deliverables on handover
- Live workflows in production tools
- Encoded SOP & decision-rule library
- Governance scaffold & audit log
- Operator runbook & training assets
- 30-day usage report & baseline metrics