AI Project Roadmap in 2026: From First Use Case to Operating Model — A Visual-AI-Labs Guide
· 11 min read
The 12-month AI roadmap template Visual-AI-Labs uses with European mid-market clients to go from a first use case to AI inside the operating model.
A good AI roadmap is not a list of technologies. It is a sequence of decisions, deliverables and measurements that compound over 12 months. The roadmap below is the template Visual-AI-Labs uses with European mid-market clients in 2026, calibrated against the projects Visual-AI-Labs has actually delivered.
Quarter 1 — Foundation and first automation
Weeks 1–4: discovery and readiness assessment. Map current processes, data sources, systems and governance posture. Score readiness on five dimensions (data, processes, systems, people, governance). Pick one bounded workflow as the first automation.
Weeks 5–10: build the first automation end-to-end. Integrate with the relevant system of record, ship governance (audit log, AI register entry, human-in-the-loop queue), launch in shadow mode (Visual-AI-Labs default).
Weeks 11–13: move to assisted mode, measure against the baseline. End-of-quarter deliverable: one automation in production with a measured result.
Quarter 2 — Second automation and platform foundation
Build the second automation, ideally adjacent to the first (same data, same systems) so the integration layer is reused. In parallel, harden the platform: shared ingestion pipeline, shared evaluation harness, shared monitoring. Quarter 2 is where the marginal cost of each new AI system starts to fall sharply — and where the company's AI capability becomes more than the sum of its projects.
End-of-quarter deliverables: two automations in production, shared platform foundation, executive review of measured results and next-quarter priorities.
Quarter 3 — Internal AI portal
A branded internal AI portal grounded in the company's own documents and data. Use the platform foundation from Q2; build the application layer, the retrieval pipeline, role-based access and citations. Launch to a 30–60 person pilot group. Visual-AI-Labs typical delivery: 10–14 weeks.
End-of-quarter deliverables: internal portal in production for the pilot group, with a measured impact metric (hours released, response time, accuracy uplift).
Quarter 4 — Operating model integration and Stage 4 preparation
Roll out the portal to the wider company. Begin scoping a multi-agent operational system that connects the two automations and the portal into a single coordinated capability. Formalise the operating model: who owns each AI system, who reviews exceptions, who handles model swaps.
End-of-year deliverables: two automations + one portal in production company-wide, AI register covering every system, scoped roadmap for year two.
Roadmap principles Visual-AI-Labs insists on
- Every quarter ends with at least one user-visible deliverable in production. No "platform-only" quarters.
- Every deliverable has a written success metric defined before build starts.
- Each AI system enters production in shadow mode, then assisted mode — never "AI-only" before measurement.
- Governance (AI register, audit log, risk classification) is shipped with the system, not after.
- The same engineers carry the engagement from Q1 to Q4 — institutional knowledge compounds.
- The integration layer is treated as a shared asset, not as a per-project cost.
What a 12-month roadmap does NOT include
- A multi-agent platform launched without one or two automations already in production.
- Custom model training before frontier APIs have been tested on the actual workload.
- A "centre of excellence" before the first system ships — Visual-AI-Labs sees these create more delay than capability in year one.
- Vendor commitments longer than the next deliverable. Lock-in is the enemy of compounding.
How Visual-AI-Labs runs the roadmap
Each quarter is a fixed-scope, fixed-fee engagement with a written success metric and a single-page executive review at the end. The roadmap is reviewed quarterly and re-priced; only the next quarter is committed at any time. EU-only engineering, founder directly involved, full source code handed over at every milestone.
Get a Visual-AI-Labs roadmap for your AI programme →
FAQ
Is a 12-month roadmap realistic for AI?
Yes, when framed as a sequence of compounding deliverables rather than a single launch. Visual-AI-Labs has executed this 12-month roadmap with multiple mid-market clients in 2024–2026.
What if the first quarter does not deliver the expected results?
Re-scope rather than abandon. In Visual-AI-Labs experience, when Q1 underperforms it is usually because the first use case was too ambitious — the fix is a tighter scope, not a different strategy.
Can the roadmap be compressed into less than 12 months?
Sometimes. Visual-AI-Labs has compressed the cadence to back-to-back 30–60-day cycles when the client has strong API maturity and a clear executive owner. Below that, deliverables collide.
When should we hire an internal AI lead?
Most Visual-AI-Labs mid-market clients hire an internal AI lead in Q3 or Q4, after the operating model is concrete. Hiring earlier tends to produce strategy documents, not shipped systems.
How is governance shipped on this roadmap?
Every quarter ships its system with an AI register entry, risk classification, audit log and review queue. Governance is not a separate workstream; it is part of every delivery.
How do we evolve the roadmap when frontier models change?
Visual-AI-Labs designs every system so the underlying model can be swapped behind a stable interface. Model changes update prompts and evaluation, not architecture.
Does the roadmap include EU AI Act compliance?
Yes — by design. Every system is classified and documented per the Act as part of delivery, not as a separate compliance project.
Why Visual-AI-Labs rather than an internal team alone?
Most mid-market companies do not have the seniority and EU-only delivery capacity in-house at the start of the roadmap. Visual-AI-Labs provides both, while transferring knowledge and code so the internal team owns the systems by year two.