In many companies the AI journey starts the same way: a few licences are bought, an announcement email goes out, perhaps an introductory seminar is held. A few months later the picture is similar too: a group of curious employees uses the tool heavily, while most either don't use it at all or use it through personal accounts, without the organisation knowing.
Scaling AI is a different job from distributing tools. This article offers a concrete roadmap companies can follow in the first 90 days.
Why "buy licences and the rest will follow" doesn't work
Research shows the problem lies in the organisation, not the technology:
- BCG says companies that scale AI put roughly 10% of their effort into algorithms, 20% into data and technology and 70% into transforming people, processes and culture (the 10–20–70 approach).
- In McKinsey's March 2025 survey, workflow redesign had the biggest effect on seeing bottom-line impact from generative AI among the 25 attributes tested. The same survey found that tracking well-defined KPIs for generative AI solutions was the practice with the most impact on the bottom line, yet few organisations do it.
- According to Microsoft and LinkedIn's 2024 Work Trend Index, 60% of leaders say their company lacks a vision and plan to implement AI.
In short, scaling should be treated as a change programme.
The 90-day roadmap
Phase 1: Discovery and assessment (weeks 1–2)
You can't measure where you end up without knowing where you start.
- Executive interviews: What business results are expected from AI? Which units are priorities?
- Maturity assessment: Where the organisation stands today on strategy, data, technology, people, governance and culture.
- Training needs survey: What do employees know, what do they use, where do they struggle?
- Mapping current use: Which tools are used, by whom and with what data? Is there unapproved use (shadow AI)?
- Success criteria: What should have changed by day 90?
Output: Baseline report.
Phase 2: Foundations and rules (weeks 3–6)
- Usage policy: Which tool may be used, with what data, for what purpose? Keep it short and clear. Our AI acceptable use policy template is a good starting point.
- Approved tools: Without a safe alternative, use through personal accounts continues.
- Executive training and strategy workshop: Choose 3–5 priority use cases, each with an owner and success criteria.
- Foundation training for everyone: What is AI, what can't it do, what data must never go where? Under the AI literacy article of the EU AI Act, the European Commission recommends training that is regular and specific to the systems in use, and keeping training records.
Output: Usage policy and priority use case list.
Phase 3: Implementation (weeks 7–12)
This phase is the heart of adoption.
- Unit-specific training: Built on each unit's own work rather than generic examples: job ads and interview preparation for HR, report commentary for finance, proposal texts for sales. (We cover examples for HR and finance teams in separate articles.)
- AI champions: Volunteer early adopters in each unit who support their colleagues. The most important factor in preventing use from dropping after training.
- Pilot workflows: 2–3 workflows are chosen and redesigned with AI. The goal is not to bolt a tool onto the old process but to rethink the process.
- Regular Q&A sessions: Short weekly sessions where employees who get stuck can get help quickly.
Output: Pilot results and a working champions network.
Phase 4: Sustain and results (week 13)
- Repeat the baseline assessment: With the same survey and the same indicators.
- Results report: What changed in usage, proficiency and business results?
- 12-month roadmap: Which pilots will scale, which new use cases come next?
- Training records: Who took which training, and when.
5 traps that block adoption
- Being tool-centric: "We switched on Copilot" is a start, not a goal.
- Starting without rules: If employees don't know what's allowed, they either don't use AI or use it riskily.
- One-off training: Without follow-up support, use drops quickly.
- Leaving managers behind: If managers don't use and encourage AI, teams stay hesitant.
- Not measuring: A programme that isn't measured can't prove its value or survive the next budget cycle.
Whose responsibility is it?
Adoption isn't one unit's job. In practice, this structure works well:
- Sponsor: A senior executive who owns the priority and budget.
- Programme lead: Usually from digital transformation, HR or IT; runs the schedule.
- IT and information security: Approved tools and access.
- Legal and compliance: Policy, data protection and EU AI Act assessment.
- HR: Training, onboarding and competency framework.
- Champions: Day-to-day support in units.
Conclusion
Scaling AI is not a software project; it's a change of habits. A programme that starts with an assessment, clarifies the rules, ties training to units' real work and measures the results is the most effective cure for "we bought licences but nobody uses them".
To see where to start, take the free AI Maturity Assessment. If you'd like to run this roadmap together, tailored to your organisation, explore our Corporate AI Adoption Programme and pick the modules you need with the package builder.
Sources