Skip to content

Adoption Roadmap

The framework organises adoption into five phases. Most companies enter at Phase 0 or Phase 1; a few discover mid-Phase 2 that they skipped Phase 0 and have to retrofit it. That is normal. The phases are not a waterfall: each phase has exit criteria, and you move forward only when they are met.

flowchart LR
    P0[Phase 0<br/>Readiness] --> P1[Phase 1<br/>Pilot<br/>0–90 days]
    P1 --> P2[Phase 2<br/>Scale<br/>Months 3–12]
    P2 --> P3[Phase 3<br/>Optimize<br/>Ongoing]
    P3 -.-> P4[Phase 4<br/>Sunset & migrate<br/>When needed]
    P1 -.->|retrofit| P0

Phase summary

Phase Duration Goal Primary output Exit criteria
0: Readiness 2–6 weeks Decide whether, when, and under what conditions to adopt. Leadership charter + interim AUP + risk classification Charter signed; interim AUP published; at least one named executive sponsor.
1: Pilot 0–90 days Prove value on a bounded scope. Pilot retrospective report with go/no-go recommendation Success criteria met or not; either way, decision is recorded.
2: Scale Months 3–12 Expand from pilot team to whole organization. Center of Excellence, tool standards, training program 60%+ of eligible engineers onboarded; metrics trending within ±20% of targets.
3: Optimize Ongoing Continuously improve practices, tools, and metrics. Quarterly framework reviews; evidence-backed policy updates N/A; steady state.
4: Sunset As needed Safely retire a tool, practice, or vendor. Migration plan with data and evidence continuity Old system decommissioned; audit trail preserved; no orphaned dependencies.

Sizing the roadmap

Sizing guide

Company size Typical timeline to Phase 2 exit Key compressions
Small (< 50) 3–6 months Skip Phase 2 CoE formation; single DRI owns all roles.
Mid (50–500) 6–12 months Combine Phase 2 CoE with existing platform/DevEx team.
Enterprise (500+) 12–18 months Full phases; expect parallel pilots per business unit.
Regulated / public sector 18–24 months Phase 0 doubles to include DPIA, legal sign-off, FedRAMP/CJIS review. See Regulated adoption.

Why phased, not all-at-once?

Two pieces of public evidence:

  1. Klarna's AI rollout and partial reversal. Klarna announced in Feb 2024 that an OpenAI-powered assistant was doing the work of 700 FTE customer service agents1; by May 2025 it had reversed course and started rehiring2. Going fast in one direction without governance made the reversal just as expensive as the rollout.

  2. DORA 2024's finding that a 25% AI-adoption increase correlated with a 1.5% drop in delivery throughput3. Adoption without measurement looks like success because individuals feel faster; team outcomes tell a different story.

Phased adoption buys you reversibility. Every phase has an explicit exit criterion, and each phase produces evidence you can use if you need to pause or roll back.

How to use this roadmap

  1. Identify your current phase. Use the exit criteria tables in each phase page. If you haven't met them, you're still in that phase, no matter how long you've been working.
  2. Read the phase page. Each one lists moves, worksheets to run, and anti-patterns to avoid.
  3. Run the relevant worksheets. The phase pages link to the specific worksheets that produce that phase's outputs.
  4. Record decisions durably. Each worksheet ends with a Decision Record block. Keep those in a single index. Leaders rotate; rationale must not.

Anti-patterns to avoid

Common roadmap failures

  • Skipping Phase 0 because "devs are already using it." The fact that adoption is already happening is exactly why you need to formalize it: quickly, not eventually. Publish an interim AUP within two weeks even if the full policy takes six months.
  • Treating the pilot as a foregone conclusion. If the pilot can't fail, it isn't a pilot; it's a procurement activity with extra steps. Define success criteria that could produce a "no" result.
  • Rolling out to the whole company after one successful squad. A Google-funded squad delivering a greenfield service is not a representative sample of your payments platform team.
  • Measuring only the things the tool vendor measures. Copilot acceptance rate is a vanity metric. Pair it with DORA + SPACE.
  • Deferring governance until "we understand the tools better." Governance is how you safely learn. See Phase 0.

Start

  1. Klarna press release, "Klarna AI assistant handles two-thirds of customer service chats in its first month", Feb 2024. ↩

  2. Coverage of Klarna's course correction; see Evidence: enterprise rollouts. ↩

  3. Google Cloud / DORA, "2024 Accelerate State of DevOps Report", Oct 2024. ↩