IT

Pilot wave 2 (M7). Impact: high. See the hub. IT is unique in this program: it is both a department being transformed and the operator of the transformation’s infrastructure. This plan covers both, with the platform role first — nothing else in the program works if IT cannot run the platform.

Current state (assumptions to validate)

Assumptions — to validate in discovery: a small team (~2–4) covering everything: SoftOne administration (user management, customizations via SoftOne’s tooling, report tweaks), M365 tenant administration, networking across three sites (HQ Piraeus, Akti Miaouli, Perama), endpoints (~109 users), the PPE e-commerce platform (likely externally hosted, vendor-managed), printers/scanners, and phone systems. Support is reactive via email/calls; no formal ticketing; documentation thin and in heads; vendor dependency for SoftOne changes; no prior experience operating GPU/LLM infrastructure. Pain points: interrupt-driven workload, single-person dependencies, no runbooks, integration requests to SoftOne slow through the vendor channel.

Dual role in the transformation

1. Platform operator — run the on-prem LLM platform (~50 daily users, Greek+English models, RAG services) described in the architecture overview: hardware, model serving, updates, monitoring, backup, access control, incident response.
2. Integration owner — own the M365 + SoftOne integration layer: API credentials and scopes, n8n instance, data pipelines feeding RAG (with BI as data steward), and the security boundary around the most sensitive scopes (finance, HR).

Target AI-first operating model

By M12 IT operates the platform as a routine service with runbooks, monitoring, and an on-call habit — and uses the platform itself for its own work: an internal IT helpdesk assistant answers employee how-to questions (password flows, M365 usage, printer setup) from a curated IT knowledge base; support requests are triaged into a lightweight queue; every recurring operation has an AI-drafted, human-verified runbook in this vault. Configuration changes, access grants, and integration deployments remain human-executed with peer/lead review.

AI use cases

Use case	Pain addressed	Data needed	Complexity	Impact	Pilot
IT helpdesk assistant for employees (how-to Q&A, EN/GR)	Interrupt-driven support load	IT knowledge base, M365 docs, internal guides	M	H	Y
Support-request triage into a queue (category, urgency)	No ticketing, lost requests	IT support mailbox	L	M	Y
Runbook drafting from incident notes & shell history	No documentation, key-person risk	Incident notes, operational procedures	L	H	N
Platform-log summarization & anomaly narrative (LLM platform + infra)	New, unfamiliar ops surface	Platform/monitoring logs	M	M	N
Integration-spec and n8n-flow documentation generation	Undocumented integrations	Flow exports, API configs (no secrets)	L	M	N
Access-review briefs (who has which assistant/data scope, quarterly)	Security-review burden	Access-control config, user directory	M	H	N

Process transformation opportunities

• Support intake: from inbox/shoulder-taps to a triaged queue with the helpdesk assistant deflecting the repetitive half.
• Operations documentation: runbook-first culture — every incident closes with a drafted runbook entry; the vault becomes IT’s memory.
• Change management: lightweight change log for platform/integration changes (what, why, rollback), assistant-drafted from the change itself.
• Access governance: quarterly scoped access reviews as a standing, evidence-backed routine rather than best-effort.

Required data sources

• IT support mailbox; internal how-to guides (to be written/collected).
• Platform and infrastructure logs/monitoring; n8n flow definitions.
• M365 admin/audit data; asset and license inventories (likely Excel).
• No credentials anywhere in corpora — secrets stay in the designated secret store, referenced only.

Potential AI agents

• Helpdesk Assistant Agent — answers employee IT questions from the knowledge base; escalates to IT when unsure; no system actions.
• Runbook Drafting Agent — turns incident/change notes into runbook drafts; IT lead approves before the runbook is trusted.
• Access Review Agent — compiles quarterly access-review briefs; IT lead and each data owner approve findings.

Automation opportunities

• n8n: support-mailbox triage → queue; platform health checks → Teams alerts; backup verification reports.
• Scheduled license/asset reconciliation report.
• New-starter/leaver checklist automation (with HR): account creation/deactivation task lists (execution stays human).

Required integrations

IT owns the integration layer rather than merely consuming it — see integrations-m365-softone.md for the full inventory. For its own use cases: M365 (support mailbox, Teams), monitoring stack, n8n.

KPIs

KPI	Baseline	M12 target
LLM platform availability (business hours)	n/a	≥99.5%
Employee IT questions deflected by helpdesk assistant	0	≥40%
Recurring operations with a current runbook	TBD (~0)	100%
Median support-request response time	TBD	−50%
Quarterly access reviews completed on evidence	0	4/4

Risks

• Platform ops overwhelm a small team → managed-service support contract for the platform in year 1 (recommendation); strict runbook discipline; TPL handover plan with shadowing.
• Integration credentials/scopes mismanaged → least-privilege scopes, secret store, access reviews; no secrets in any corpus or vault file.
• IT becomes the transformation bottleneck (every dept needs integrations) → integration backlog prioritized at program level, not by loudest requester; n8n self-service for read-only reports where safe.
• Key-person dependency worsens (platform knowledge in one head) → runbooks + cross-training are deliverables, not options.

Training needs

• IT team: platform operations (model serving, monitoring, backup/restore, incident response) — deep, hands-on, M3–M5 with TPL shadowing through M8.
• IT team: n8n development, API scope management (M4–M6).
• IT lead: security/access governance routine (M6).

Deliverables

• Operated platform with monitoring, backup, and runbook set (from M4, hardened by M8).
• Integration layer (n8n + SoftOne/M365 connectors) with documentation.
• IT knowledge base + helpdesk assistant (pilot M7); support queue.
• Access-governance routine and quarterly review evidence.

12-month execution milestones

Month	Milestone
M1–M2	Discovery: infrastructure audit, skills assessment, support-load baseline
M3	Platform hardware installed; IT ops training begins (TPL-led)
M4	Platform go-live for all users; backup/monitoring in place; first runbooks
M5	SoftOne read integration live (serves wave-1 pilots); n8n in production
M6	Sensitive-scope integrations (finance) security-reviewed and live
M7	Wave-2 pilot (own dept): helpdesk assistant + support triage
M8	TPL shadowing ends; IT operates platform independently; first access review
M9–M10	Write integrations (PO/quote drafts) staged in; runbook coverage completed
M12	Ops handover complete; KPI review; year-2 capacity plan