Strategic AI Transformation Report — Prime Products Ltd

Prepared by TPL for the CEO and Board of Prime Products Ltd — July 2026.

This report sets out why Prime Products Ltd should become an AI-first company, what that means concretely for a ~109-person, €23.7M ship-supply and safety-solutions business, and how the transformation will run over 12 months. It is the anchor document of the strategy area; the executable program — charter, governance, roadmap, KPIs, risks — is detailed in 02-program. Company baseline facts are documented in the company profile; formal needs are tracked as REQ-001…REQ-010 in the requirements register.

---

1. Why PRIME PRODUCTS should become AI-first

PRIME PRODUCTS is not a distressed company looking for a rescue. It is a healthy, respected, nearly-century-old business (brand since 1929, S.A. since 1987) with a strong balance sheet (~50.6% equity ratio) and rebounding pre-tax profit. That is exactly why the timing is right: the transformation can be funded and absorbed from a position of strength, before competitive pressure forces it from a position of weakness. Five company-specific dynamics make the AI-first case unusually strong here:

1. Thin margins turn productivity directly into profit. On ~€23.7M revenue, net profit is ~€0.36M — a ~1.5% net margin. At that margin, every €100k of operating cost removed or capacity freed is roughly equivalent to winning €6–7M of additional revenue at current profitability. No sales initiative can plausibly deliver that; productivity can. AI-first operations attack exactly the cost base that dominates a wholesale trading business: manual document handling, quoting, order processing, coordination, and reporting.

2. The core commercial work is knowledge work in disguise. Quoting a vessel provisioning order, responding to a shipyard tender, or bidding into a military/civil-defense or public-sector procurement requires assembling product data, certificates, prior quotes, IMPA/ISSA codes, delivery constraints, and compliance evidence — fast, across thousands of SKUs, often under a port-call deadline. Today this lives in SoftOne, mailboxes, spreadsheets, and experienced heads. It is precisely the workload where retrieval-augmented LLMs deliver step-change speed and consistency.

3. The multilingual documentation burden is structural. PRIME PRODUCTS operates between Greek internal reality and an English-speaking maritime world: customer correspondence, tender documents, safety datasheets, training material (ISO 21001/29993 certified OHS training), ISO 22000 food-safety documentation, Admiralty/e-navigation product content. Bilingual drafting and translation currently consumes skilled staff time on every transaction. An on-prem LLM platform with strong Greek and English makes bilingual output near-free.

4. Tribal knowledge is an aging, unhedged asset. Decades of know-how — which supplier delivers to Perama on a Saturday, how a particular flag state inspects bonded stores, which PPE certificate a Navy tender actually requires — sits with long-tenured employees and is not written down (assumption on workforce tenure profile — to validate in discovery). The Obsidian knowledge base plus AI-assisted capture converts this into a durable company asset before retirements make it a crisis.

5. Revenue is drifting and sales needs leverage, not headcount. Revenue declined ~2.2% y/y. Across seven sales segments (maritime, yachting, e-navigation, industry, military & civil defense, public sector, retail), the constraint is salesperson throughput: quotes per day, tenders answered, follow-ups done. AI assistance raises throughput per salesperson without adding cost — the only growth lever consistent with a 1.5% margin.

The alternative — waiting — has a real cost. Larger international ship chandlers and marketplace platforms are already deploying AI in quoting and catalog operations (assumption on competitor maturity — to validate in discovery). In a business where the winner of a provisioning order is often whoever quotes accurately first, a competitor with a 10× faster quote cycle is an existential threat, not an inconvenience.

2. Strategic objectives

Numbered, measurable, 12-month horizon. Each maps to KPIs in the KPI framework.

1. Deploy a production on-prem AI platform (NVIDIA GPU LLM infrastructure) serving ~50 daily active users in Greek and English, with ≥99% business-hours availability, by end of Q2 of the program.
2. Reduce median quote/tender turnaround time by ≥50% in the pilot sales segments, measured against a discovery-phase baseline.
3. Capture the company’s operating knowledge: every one of the 15 in-scope departments has its core processes documented in the Obsidian knowledge base — target ≥80% of processes identified in discovery documented by month 12.
4. Put an AI assistant into every role’s daily flow: ≥70% of the ~50 target users active on the platform at least weekly by month 12, ≥50% daily.
5. Deploy ≥10 production AI use cases and ≥5 operating agents across departments (registered in the AI agent registry), each with a measured before/after baseline.
6. Free ≥15% of working time in the top-3 most impacted back-office functions (target: reporting/BI, accounting document handling, customer-service correspondence) — redeployed to higher-value work, not headcount reduction.
7. Zero confidential-data incidents: no customer, defense, or commercial data leaves company-controlled infrastructure; full audit trail on AI access to business data.
8. Establish measurable transformation governance: Mission Control (vanos.tpl.one) live with program KPIs, and a monthly steering rhythm running without slippage from month 1.

3. Business case

All figures are placeholder ranges for board framing; the binding numbers arrive with the stage-offer structure and hardware sizing (hardware and sizing).

Cost structure (12 months):

Cost block	Range (placeholder)	Nature
On-prem GPU hardware (server, networking, UPS)	€XX–XXk	One-off capex, ~4–5 yr useful life
LLM platform build & integration (M365, SoftOne)	€XX–XXXk	One-off, TPL services
Discovery, department pilots, agents, knowledge base	€XX–XXXk	Staged services across 12 months
Training & change management	€X–XXk	Services + internal time
Run costs (power, support, licenses)	€X–XXk / yr	Recurring, modest — no per-token cloud fees
Internal time (champions ~10–15%, staff interviews/training)	opportunity cost	The real hidden cost — governed via steering

Benefit mechanisms (how money is actually made or saved — each measured against a discovery baseline):

• Quoting/tender throughput — more quotes and tenders answered per salesperson per day; higher win volume at unchanged headcount. Mechanism: RAG over product catalog, prior quotes, certificates; bilingual drafting.
• Back-office cycle compression — order entry, invoice matching, customs/bonded-warehouse documentation, ISO compliance evidence assembly. Mechanism: agents for repetitive document work with human approval.
• Error and rework reduction — fewer mis-picked provisioning orders, fewer document rejections at customs/port authorities. At a 1.5% margin, one avoided botched vessel delivery can exceed a month of platform run cost.
• Knowledge retention — reduced key-person risk and faster onboarding (relevant with 109 employees where single individuals own whole domains).
• Management visibility — reporting/BI that currently consumes days per month produced continuously; better working-capital and inventory decisions across bonded warehouses.

Payback logic for a company this size. The conservative test: the program pays back if it frees the equivalent of 3–5 FTEs of capacity across 109 employees (≈3–4% of workforce time) or lifts sales throughput enough to recover ~1–2% of revenue (≈ the recent decline). Either alone plausibly covers the placeholder cost range within 18–24 months of go-live; the levers are independent and additive. Because the infrastructure is owned capex with ~4–5-year life and no per-usage cloud fees, marginal AI usage after go-live costs approximately electricity — the economics improve with adoption, the opposite of cloud-API models.

4. Expected operational benefits per value lever

Value lever	What changes	Where it lands	Indicative effect (to baseline in discovery)
Assisted quoting & tendering	Draft quotes/tender responses generated from catalog + history; human review before send	Sales (all 7 segments), technical support	50%+ faster turnaround; more tenders answered
Bilingual document automation	GR↔EN correspondence, datasheets, training material, tender annexes drafted by AI	Sales, marketing, compliance, customer service, HR	Hours per document → minutes
Knowledge retrieval (RAG)	“Ask the company” over knowledge base, product data, procedures, past cases	All departments	Minutes to answers that today take calls and searching
Agentic back-office work	Repetitive multi-step tasks (order confirmation drafting, invoice matching prep, report assembly) run by agents, human-approved	Accounting, finance, logistics, warehouse, administration	≥15% time freed in top-3 functions
Compliance & quality documentation	ISO 22000/21001/29993 evidence, audit prep, certificate tracking assisted by AI	Compliance/quality, warehouse, training business	Audit prep days reduced; fewer findings
Continuous reporting/BI	Standard management reports generated on demand from SoftOne + M365 data	Reporting/BI, executive management, finance	Monthly reporting cycle → continuous
Knowledge capture & onboarding	Tribal knowledge documented; new hires productive faster	HR, all departments	Onboarding time reduced; key-person risk hedged

5. Target operating model — what “AI-first” means at PRIME PRODUCTS

AI-first does not mean AI replaces people. It means the company is organized so that AI does the repetitive cognitive work and people do judgment, relationships, and exceptions. Five principles define the target state:

1. Every process is documented in the knowledge base. The Obsidian vault (see knowledge-base blueprint) is the single source of company know-how — if a process isn’t written down, it can’t be assisted, automated, or safely handed over.
2. An AI assistant sits in every role’s daily flow. Not a separate tool people must remember to visit: assistance where the work happens — drafting the email, preparing the quote, summarizing the tender — in Greek or English as the situation demands.
3. Agents handle repetitive work. Recurring multi-step tasks are delegated to registered, monitored agents (see AI agent registry) with defined scope and owners.
4. Humans approve everything external-facing. No quote, invoice, tender response, or customer message leaves the company without a named human approving it. AI drafts; people decide. This is non-negotiable, particularly for defense and public-sector business.
5. Measure everything. Every use case ships with a baseline and a metric; the program reports through Mission Control (see mission-control blueprint) and the KPI framework. What isn’t measured isn’t claimed.

Organizationally, the model adds a light layer — department AI champions, a knowledge manager, a steering committee — detailed in the governance model. It changes how existing roles work; it does not restructure the company.

6. Transformation principles

1. On-prem first — customer and defense data never leaves the building. All confidential inference runs on PRIME PRODUCTS-owned GPU infrastructure in Piraeus. No cloud LLM dependency for business data. This is a hard requirement given military, civil-defense, and public-sector customers.
2. Human approval on all external output. AI drafts, humans sign. Always.
3. CEO-led, department-owned. The CEO sponsors and unblocks; department heads own their transformation. TPL enables; it does not own PRIME PRODUCTS processes.
4. Augment, don’t replace. This is explicitly not a headcount-reduction program; freed capacity is redeployed. Stated up front, repeated often — adoption dies without this credibility.
5. Minimal disruption to daily operations. Vessels get supplied every day of the program. Pilots run alongside existing processes until proven; no big-bang cutover; SoftOne ERP stays.
6. Bilingual by design. Every user-facing capability works in Greek and English from day one — not Greek as an afterthought.
7. Knowledge before automation. A process gets documented before it gets an agent. Undocumented automation is unmanageable risk.
8. Baseline before benefit. Every use case measures the “before” during discovery; no claimed improvement without a measured baseline.
9. Small, shippable, monthly. Value delivered in monthly increments visible on Mission Control — no six-month silent build phases.
10. Everything in the open (internally). Plans, status, decisions, and KPIs are visible to all staff via Mission Control and the knowledge base; secrecy breeds resistance.

7. Department impact summary

Full per-department analyses live in 04-departments. Impact level = expected magnitude of change to daily work within 12 months.

Department	Impact	Primary AI lever
Sales (7 segments)	High	Assisted quoting & tender response; bilingual customer correspondence
Procurement	High	Supplier comparison, RFQ drafting, price-list ingestion into structured data
Customer service	High	Draft replies with order context; bilingual correspondence; case summarization
Accounting	High	Document processing agents (invoice matching prep, reconciliation support)
Reporting/BI	High	Continuous report generation from SoftOne/M365; management Q&A over data
Compliance/quality	High	ISO evidence assembly, audit prep, certificate & document control
Logistics	Medium–High	Shipping/customs document drafting; delivery coordination summaries
Warehouse	Medium	Bonded-stores documentation; picking-list intelligence; stock queries in natural language
Marketing	Medium	Bilingual content for Protect & Defend e-commerce and product lines
Technical support	Medium	Product/spec Q&A over datasheets and certificates; PPE selection assistance
Finance	Medium	Cash-flow and working-capital reporting; scenario drafting
HR	Medium	Onboarding material, policy Q&A, OHS training-content production
Executive management	Medium	Briefings, decision memos, “ask the company” retrieval
IT	Medium (as user) / High (as operator)	Runs the platform; internal support assistant
Administration	Medium	Correspondence, filing, meeting summaries, document templates

8. Risk and mitigation strategy — summary

The full, owned, scored register is the risk register. The strategic posture: the dominant risks are human (adoption, fear of job loss, champion overload) and data (SoftOne extraction quality, tribal knowledge capture), not technological — on-prem LLM serving for 50 users is well-trodden engineering. Accordingly, mitigation weight goes to change management (plan), the explicit no-headcount-reduction commitment, CEO visibility, and an early, unglamorous investment in data access and knowledge capture during discovery. Confidentiality risk is mitigated structurally (on-prem only, security and operations) rather than procedurally.

9. Success metrics

The binding metric set, with baselines, targets, owners, and measurement methods, is the KPI framework. Headline board-level indicators: platform adoption (weekly/daily active users of ~50), quote/tender turnaround vs. baseline, hours freed in top-3 back-office functions, processes documented in the knowledge base, production use cases and agents live, availability, and confidential-data incidents (target: zero). All reported monthly via Mission Control.

10. 12-month strategic roadmap — summary

Quarter-level view; the full month-by-month plan is the 12-month roadmap, decomposed in the work breakdown structure.

Quarter	Theme	Headline outcomes
Q1 — Foundations & Discovery	Learn the company, build the base	Governance running; company-wide discovery (questionnaires + interviews, 03-discovery); requirements & baselines captured; hardware procured; knowledge-base skeleton and Mission Control live
Q2 — Platform & First Pilots	Prove it works here	On-prem LLM platform in production; M365 + SoftOne integrations; first pilots live in 2–3 high-impact departments (sales quoting first); first agents in back office; training wave 1
Q3 — Scale Across Departments	Make it everyone’s tool	Rollout to all 15 departments; ≥10 use cases in production; knowledge base ≥60% process coverage; champions running department backlogs; training waves 2–3
Q4 — Optimize & Institutionalize	Make it stick	Adoption and KPI targets hit; agent portfolio hardened; operating model handed to PRIME PRODUCTS roles (knowledge manager, champions, IT operations); year-2 portfolio and board review

---

Next document: board presentation outline. Program mechanics: project charter and governance model.