SCALING 9 min read

From pilot to scale: the 5 stages of AI rollout

The pilot worked great. Scaling it is a disaster. Why this happens and how to avoid it.

From pilot to scale

According to McKinsey State of AI 2025: 88% of companies use AI, but only 7% have it fully deployed at organizational scale. Nearly two thirds are still stuck in the experiment-and-pilot phase. That gap isn't a technology problem - it's a problem of getting from pilot to scale. Here's a roadmap that works.

Stage 1: Discovery (Week 1-2)

Before you write a single line of code, answer these questions:

  • • What specific problem are we solving?
  • • Who is the business owner?
  • • How will we measure success?
  • • Do we have data? At what quality?

Output: a 1-pager defining the project and its success criteria.

Stage 2: Proof of Concept (Week 3-6)

A PoC is a test of technical feasibility. It doesn't have to be pretty, it just has to answer the question: "will this even work?"

WHAT YOU DON'T DO IN A POC

  • ✗ UI - Excel/a notebook is enough
  • ✗ Scalable infrastructure - a laptop is enough
  • ✗ Integrations - mocked data
  • ✗ Security review - that comes later

Output: a working prototype + an assessment of whether it's worth pushing on.

Stage 3: Pilot (Month 2-3)

A pilot is a test with real users, but at limited scale. 5-20 people, one department, one region.

What you're testing:

  • • Do users actually use it?
  • • Does the solution deliver value?
  • • What needs to change before scaling?
  • • What does the feedback loop look like?
"A pilot that 'works' isn't success. Success is a pilot that shows you what to fix before scaling."

Output: a go/no-go decision + a list of production requirements.

Stage 4: Production Ready (Month 4-6)

This is where most projects die. Because the move from pilot to production requires:

PRODUCTION READINESS CHECKLIST

  • Infrastructure: Scalable, with redundancy
  • Monitoring: Alerting, logging, dashboards
  • Security: Review, penetration testing
  • Integrations: Real APIs, not mocks
  • Documentation: For users and support
  • Training: Onboarding for users
  • Support: Who answers when something breaks?
  • Rollback: How do we revert to the old process?

Output: a production-ready system + a rollout plan.

Stage 5: Scale (Month 6+)

Scaling isn't "turn it on for everyone." It's a controlled rollout:

  • Wave 1: 10% of users, you watch the metrics
  • Wave 2: 25% of users, you gather feedback
  • Wave 3: 50% of users, you optimize
  • Wave 4: 100% - full rollout

Between each wave: review, fixes, a decision on whether to continue.

Why do projects die between stages?

COMMON CAUSES OF DEATH

  • PoC → Pilot: No business owner, "interesting but not now"
  • Pilot → Production: Underestimated production cost, no budget
  • Production → Scale: Legacy systems, organizational resistance, no change management

How to improve your odds of success

  1. 1. Have a C-level sponsor from day one
  2. 2. Budget the whole path, not just the PoC
  3. 3. Start with a simple problem with clear ROI
  4. 4. Involve end users from the pilot onward
  5. 5. Plan change management in parallel with the technology

Summary

From pilot to scale is a marathon, not a sprint. Every stage has its own goals and success criteria. Skipping stages is a recipe for failure. Patience and method - that's the road to the 7% of companies that have AI fully deployed.

SP

Szymon Paluch

ex-CTO · AI Strategy

Want to get from pilot to scale?

30 minutes of substance. No sales pitch.

Book a call