PRIME PRODUCTS · MISSION CONTROL
AI-first transformation · by TPL · vanos.tpl.one

Docs / 08-transformation/02-program/work-breakdown-structure

Work Breakdown Structure

Full task decomposition of the 12-month Prime Products AI transformation across 13 phases with owners, dependencies and deliverables.

type: plan updated: 2026-07-03 owner: kotsalidis

Work Breakdown Structure — Prime Products AI Transformation

Thirteen phases (0–12) over program months M1–M12 (M1 ≈ September 2026 — assumption, to confirm at kickoff). Phases deliberately overlap: knowledge base (Phase 6) and Mission Control (Phase 7) start in M1–M2 to give the program visible early wins; hardware procurement (Phase 4) is ordered in M3 to absorb the 6–10-week supplier lead time; training (Phase 10) runs alongside pilots and rollout rather than as a block at the end. Calendar projection: roadmap-12-month.md. Owner values are roles, per governance: TPL lead, Prime Products IT, dept champion, CEO, steering (steering committee).

Overlap summary: 0∥1 (M1) · 1∥2∥3∥6∥7 (M2–M3) · 4 spans M3–M6 under 5’s early work · 5∥6∥8 (M5–M7) · 8∥9∥10 (M6–M9) · 9∥10∥11 (M9–M11) · 11∥12 (M11–M12).


Phase 0 — Preparation & governance (M1)

Objective: stand up the program so that every later phase has a decision path, a budget envelope, and a communication channel.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
0.1Kickoff & charter sign-offConfirm M1 start date; CEO + TPL align on scope, out-of-scope, success criteriaCEOSigned project-charter.md
0.2Form steering committeeCEO, TPL lead, IT head, 2–3 department heads; monthly cadence booked for 12 monthsCEO0.1governance-model.md, calendar invites
0.3Appoint program rolesTPL lead, Prime Products internal coordinator, IT platform ownersteering0.2RACI in governance model
0.4Budget envelope approvalCapEx (hardware) + OpEx (TPL, licenses, training) envelopes; €[TBD] placeholder until Phase 3 sizingsteering0.1Approved budget memo
0.5Communication kickoffCEO all-hands announcement; Teams program channel; first Mission Control news postCEO0.1Announcement, #ai-transformation channel
0.6Confidentiality & data handling baselineNDA coverage check, data_sensitivity rules, no-plaintext-credentials policy applied to program artifactsTPL lead0.1Data-handling note in charter
0.7Tooling accessTPL access to M365 tenant (read), SoftOne test environment request, GitHub/vault accessPrime Products IT0.3Access checklist complete

Phase 1 — Discovery & requirement analysis (M1–M3)

Objective: capture how Prime Products actually works today — per department, per system — and register validated needs as REQ-NNN. Interviews only; near-zero load on daily operations (≤2 h per interviewee).

WBSTaskSubtasks / notesOwnerDepends onDeliverables
1.1Finalize discovery instrumentsQuestionnaires per audience (exec, dept head, operator, IT, compliance) — see requirement-analysis-framework.mdTPL lead0.15 questionnaire files in 03-discovery/
1.2Interview schedule15 departments × 1–2 sessions; 90 min each; spread across M1–M2 to keep operational load lightTPL lead1.1, 0.5Interview calendar
1.3Executive interviewsCEO + management: strategy, pain points, decision bottlenecksTPL lead1.2Interview notes (type: interview)
1.4Department interviews — commercialSales, procurement, customer service, marketingTPL lead1.2Interview notes per dept
1.5Department interviews — operationsLogistics, warehouse (incl. bonded), product/technical supportTPL lead1.2Interview notes per dept
1.6Department interviews — back officeFinance, accounting, HR, administration, reporting/BITPL lead1.2Interview notes per dept
1.7Department interviews — IT & complianceIT, compliance/quality (ISO 22000, bonded-warehouse controls)TPL lead1.2Interview notes per dept
1.8Systems inventorySoftOne modules in use, customizations, API availability; M365 tenant layout; shadow tooling (Excel, email workflows)Prime Products IT0.7Systems inventory doc in 05-technical/
1.9Data landscape assessmentSoftOne data quality sampling (items, customers, prices), document repositories, file sharesTPL lead1.8Data assessment note
1.10Requirements registrationOne REQ-NNN file per validated need; register table updated in same changeTPL lead1.3–1.702-Requirements/_index.md populated
1.11Discovery synthesis reportFindings, quick wins, department readiness scores; feeds Phase 2 and pilot selectionTPL lead1.9, 1.10Discovery report; steering presentation
1.12KPI baseline captureMeasure current values for company/department KPIs during interviews (quote turnaround, report latency, etc.)TPL lead1.4–1.7Baselines in kpi-framework.md

Phase 2 — Process mapping & AI opportunity design (M2–M4)

Objective: turn discovery raw material into mapped processes and a scored AI use-case portfolio. Overlaps Phase 1 — mapping starts as soon as the first interviews land.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
2.1Process mapping method & templateLightweight SIPOC-style one-pagers, not BPMN ceremonyTPL lead1.1Template in 06-knowledge-base library
2.2Map core commercial processesQuote-to-order (ship supply RFQs), order-to-delivery, tender/public-sector bidsTPL lead1.4, 2.1Process maps
2.3Map operations processesBonded-warehouse in/out, vessel delivery logistics, ISO 22000 traceability, chart/e-navigation fulfilmentTPL lead1.5, 2.1Process maps
2.4Map back-office processesInvoicing, month-end close, payroll interface, management reportingTPL lead1.6, 2.1Process maps
2.5AI opportunity identificationPer mapped process: candidate assist/automate points; harvested into ai-use-case-pipeline.mdTPL lead2.2–2.4Use-case long-list (~30–50)
2.6Use-case scoring & prioritizationValue × feasibility × data-readiness; classify quick-win / pilot / laterTPL lead2.5Scored pipeline, top-10 shortlist
2.7Pilot department selectionRecommend 2–3 pilot departments to steering (criteria: pain, data readiness, champion strength)steering2.6, 1.11Gate M4: pilots approved
2.8Department one-pagers15 files in 04-departments: as-is, opportunities, KPIsTPL lead2.515 department docs

Phase 3 — On-prem AI infrastructure architecture (M2–M3)

Objective: design the on-prem NVIDIA platform for ~50 daily users (Greek + English) precisely enough to order hardware in M3. Runs parallel to discovery.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
3.1Requirements → sizing inputsConcurrency (~50 DAU), model classes, context lengths, RAG corpus size estimateTPL lead1.8Sizing assumptions doc
3.2Reference architectureGPU server(s), storage, network, virtualization/containers, backup, DR stance — architecture-overview.mdTPL lead3.1Architecture doc
3.3Hardware bill of materialsGPU spec, RAM, NVMe, redundancy; options at 2–3 price points €[TBD] — hardware-and-sizing.mdTPL lead3.2BoM with options
3.4Model & serving stack selectionOpen-weight LLMs with strong Greek performance; serving engine; embedding models — llm-serving-stack.mdTPL lead3.1Stack decision (ADR)
3.5Security architectureNetwork segmentation, SSO via M365/Entra, secrets handling, audit logging — security-and-operations.mdTPL lead3.2Security design
3.6Site surveyServer-room power/cooling/rack at Piraeus HQ; UPS capacity; assumption — to validate in discovery: existing room is adequatePrime Products IT3.3Site readiness report
3.7Hardware budget approvalPresent BoM + TCO to steering/boardsteering3.3, 3.6Gate M2/M3: board approves hardware budget

Phase 4 — Hardware procurement & deployment (M3–M6)

Objective: order in M3, absorb the realistic 6–10-week supplier lead time, install and burn in by M5–M6. The lead time is the program’s critical path — everything GPU-dependent is sequenced behind it.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
4.1Vendor RFQ & selection≥2 quotes; delivery time weighted as heavily as price; warranty/support termsPrime Products IT3.7Vendor selection memo
4.2Purchase order & order trackingPO issued M3; weekly vendor check-ins; slippage flagged to steering immediatelyPrime Products IT4.1PO; delivery ETA log
4.3Site preparationRack, power circuits, cooling verification, network drops — done during lead time, not after deliveryPrime Products IT3.6, 4.2Site ready certificate
4.4Interim dev environmentSmall cloud/loaner GPU sandbox so Phase 5 software work starts before delivery; no Prime Products business data in cloud sandboxTPL lead3.4Dev environment note
4.5Delivery, installation & cablingPhysical install, firmware, out-of-band managementPrime Products IT4.2, 4.3Installed hardware
4.6Burn-in & acceptance testingGPU stress tests, thermals, storage throughput; acceptance criteria from BoMPrime Products IT4.5Acceptance report
4.7Base platform installOS, drivers, CUDA, container runtime, monitoring agentsPrime Products IT4.6Platform-ready checklist

Phase 5 — Local LLM stack implementation (M4–M7)

Objective: a working, secured, monitored LLM platform — chat + RAG + API — at GA quality by end of M6/early M7. Software work starts in M4 on the interim environment (4.4), then migrates onto delivered hardware.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
5.1Serving engine deploymentInference server, model registry, quantization strategy per llm-serving-stack.mdTPL lead4.4 (dev) / 4.7 (prod)Serving stack running
5.2Greek/English model evaluationBenchmark shortlisted open-weight models on Prime Products-realistic Greek + English tasks (emails, specs, ISO docs)TPL lead5.1Eval report; model choice confirmed
5.3Chat UI deploymentWeb chat front-end, SSO via Entra ID, conversation history, feedback thumbsTPL lead5.1, 3.5Chat UI live for pilot group
5.4RAG pipelineIngestion from Obsidian KB + selected SharePoint libraries; chunking, embeddings, retrieval eval — rag-and-data-pipeline.mdTPL lead5.1, 6.4RAG v1 with retrieval eval ≥ target
5.5Integrations v1M365 (SharePoint read, Teams bot skeleton) and SoftOne read-only queries — integrations-m365-softone.mdTPL lead5.1, 1.8Integration endpoints
5.6Telemetry & usage analyticsPer-user/query metrics, latency, GPU utilization; feeds kpi-framework.md and Mission ControlTPL lead5.1Telemetry dashboard
5.7Security hardening & pen checkAccess review, prompt-injection surface review, audit-log verificationPrime Products IT5.3, 5.5Hardening report
5.8Platform GA readiness reviewLoad test at 50-user profile; runbook handover to ITsteering5.2–5.7Gate M6: go/no-go platform GA

Phase 6 — Obsidian knowledge base design & rollout (M2–M9)

Objective: the company knowledge base — the substrate for RAG and the durable asset of the program. Starts early (M2) because content accumulation is the long pole, and pilots need real content.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
6.1Vault architecture & conventionsFolder taxonomy, frontmatter schema, linking rules — obsidian-vault-blueprint.mdTPL lead0.1Blueprint doc
6.2Template librarySOP, process map, product note, meeting note, decision templates — obsidian-template-library.mdTPL lead6.1Template files
6.3Vault infrastructureGit-backed vault, sync for ~50 users, access model per sensitivityPrime Products IT6.1Working shared vault
6.4Seed content migrationHighest-value existing docs (ISO 22000 procedures, key SOPs, product/tender boilerplate) converted first — priority set by 2.6TPL lead6.2, 2.5~50 seed docs (RAG-ready)
6.5Department content sprints2-week sprints per pilot department with champion; then remaining departments in rollout orderdept champion6.4, 2.7Dept sections ≥ SOP-coverage target
6.6KB quality mechanicsFreshness dates, orphan detection, review-cycle assignments; metrics into KPI frameworkTPL lead6.5KB health dashboard
6.7KB ↔ RAG feedback loopRetrieval-miss reports drive content gaps list; monthly triage with championsTPL lead5.4, 6.5Gap-fix log

Phase 7 — Mission Control website (vanos.tpl.one) (M1–M12, continuous)

Objective: the program’s public face inside the company — status, KPIs, news, training. Live by end of M1 in v0 form; evolves all year.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
7.1Blueprint & page specsIA, audiences, update model — mission-control-blueprint.md, page specsTPL lead0.5Blueprint approved
7.2v0 launchStatic site at vanos.tpl.one: program overview, roadmap, news feed, FAQTPL lead7.1Site live end M1
7.3KPI dashboard pagesWire telemetry (5.6) and KPI framework into live dashboard panelsTPL lead5.6, 7.2Dashboards live ~M6
7.4Training & help sectionCourse calendar, self-serve learning paths, prompting guide linksTPL lead7.2, 10.2Training hub
7.5Monthly content operationsNews post per steering cycle; milestone announcements; champion spotlightsTPL lead7.212 monthly updates

Phase 8 — Department pilots (M6–M8)

Objective: prove value in 2–3 departments with real work, real data, honest measurement. Gated on platform GA (5.8).

WBSTaskSubtasks / notesOwnerDepends onDeliverables
8.1Pilot design per department2–3 use cases each from top-10 shortlist; success criteria = KPI deltas vs 1.12 baselinesTPL lead2.7, 5.8Pilot plans
8.2Pilot onboarding & trainingChampions + pilot users trained first (Phase 10 curriculum); accounts, access, KB sections readydept champion8.1, 10.3, 6.5Trained pilot cohort
8.3Pilot execution — dept ACandidate (assumption — to validate at gate 2.7): sales/quoting — RFQ response drafting from KB + SoftOne datadept champion8.2Running pilot, weekly metrics
8.4Pilot execution — dept BCandidate: customer service — inquiry triage & drafted repliesdept champion8.2Running pilot, weekly metrics
8.5Pilot execution — dept CCandidate: compliance/quality — ISO 22000 document Q&A and audit prepdept champion8.2Running pilot, weekly metrics
8.6Weekly pilot reviews30-min per pilot: usage, friction, wins; issues fed to platform backlogTPL lead8.3–8.5Review notes
8.7Pilot evaluation & rollout decisionKPI deltas, CSAT, lessons; rollout wave order proposedsteering8.6Gate M8: pilot results accepted; rollout approved

Phase 9 — AI agent & automation development (M7–M10)

Objective: move beyond chat — build the agents and automations from the scored pipeline into a governed registry.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
9.1Agent platform foundationsOrchestration framework, tool-calling into SoftOne/M365, human-approval steps, sandboxingTPL lead5.5, 5.8Agent runtime
9.2Agent registry & lifecycleEvery agent registered with owner, scope, permissions, review date — ai-agent-registry.mdTPL lead9.1Registry live
9.3Automation wave 13–5 quick-win automations from pilots (e.g. quote assembly, inquiry routing, report drafting)TPL lead9.1, 8.6Wave-1 agents in production
9.4Automation wave 2Next 5–8 from pipeline, incl. non-pilot departments (finance close support, warehouse doc handling)TPL lead9.3, 8.7Wave-2 agents
9.5Straight-through monitoringPer-agent run metrics: runs, success %, hours saved, human-override rate → KPI frameworkTPL lead9.2, 5.6Agent metrics on Mission Control
9.6Power-user enablementSelected power users build/modify simple automations under IT guardrailsTPL lead9.2, 10.4≥5 user-built automations

Phase 10 — Training & change management (M4–M11)

Objective: execute change-management-and-training-plan.md. Runs as a continuous track; keyed to pilot and rollout waves.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
10.1Champions program launch1 champion per department (15), selection with dept heads, weekly guild from M4TPL lead2.8Champion roster, guild cadence
10.2Curriculum development5 audience tracks (all-staff, dept heads, power users, IT, management); labs + self-serve KB pathsTPL lead6.2Curriculum in KB + Mission Control
10.3Champion & pilot-user trainingFirst cohort, ahead of pilots (M5–M6)TPL lead10.2~25 people trained
10.4All-staff training waves2-h hands-on labs per department, sequenced with rollout waves (M8–M11); ≤4 h/person/monthdept champion10.2, 8.7~109 staff trained
10.5IT platform-ops trainingRunbooks, on-call model, model-update procedure handoverTPL lead5.8IT certified on ops
10.6Management trackAI-assisted decision-making, reading the dashboards, use-case sponsorshipTPL lead10.2, 7.3Management sessions done
10.7Skills assessment & certificationPrompt-skill assessment; internal certification badge; results → KPI frameworkTPL lead10.4Certification records
10.8Comms & feedback operationsMonthly all-hands segment, pulse surveys, #ai-help channel opsTPL lead0.5Monthly comms artifacts

Phase 11 — Full rollout (M9–M11)

Objective: every department onboarded, in waves, with no big-bang cutover and parallel-run on anything process-critical.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
11.1Rollout wave plan12 remaining departments in 3 waves (M9/M10/M11), ordered by readiness score from 1.11TPL lead8.7Wave plan approved
11.2Wave 1 onboarding~4 departments: accounts, training (10.4), KB sections live, 2 use cases eachdept champion11.1Wave 1 active
11.3Wave 2 onboarding~4 departmentsdept champion11.2Wave 2 active
11.4Wave 3 onboardingRemaining departments incl. branches (Akti Miaouli, Perama) and warehouse floor access modeldept champion11.3Wave 3 active; all 15 depts live
11.5Capacity & performance watchMonitor GPU utilization/latency as DAU climbs to ~50; scale plan triggered if p95 breaches targetPrime Products IT5.6, 11.2Capacity reports
11.6Hypercare per wave2-week elevated support after each wave; champions + #ai-help + TPL escalationTPL lead11.2–11.4Hypercare logs
11.7Adoption checkpointWAU ≥ target across departments; laggard-department action planssteering11.4Gate M11: rollout complete declared

Phase 12 — Optimization, governance & continuous improvement (M11–M12)

Objective: shift from project to operating model — Prime Products runs the platform, TPL steps back to advisory.

WBSTaskSubtasks / notesOwnerDepends onDeliverables
12.1KPI review vs M12 targetsFull read of kpi-framework.md; ROI narrative for boardTPL lead11.7KPI attainment report
12.2Model & stack refresh cycleProcedure for evaluating/adopting new open-weight models quarterlyTPL lead5.2, 10.5Refresh runbook
12.3AI governance board (permanent)Steering committee transitions to quarterly AI governance board; use-case intake process owned by Prime ProductsCEO12.1Governance charter v2
12.4Use-case pipeline handoverBacklog, scoring method, and agent registry ownership transferred to Prime Products IT + champions guildTPL lead9.2Handover sign-off
12.5Year-2 roadmapNext-wave opportunities (deeper SoftOne write-back, customer-facing AI, advanced analytics) with budget sketch €[TBD]steering12.1Year-2 proposal
12.6Program closureFinal board presentation, lessons-learned, archive of program artifacts in vaultsteering12.1–12.5Gate M12: program closed; ops handed over