Published
Updated
What an AI roadmap contains
An AI roadmap sequences use cases, resources, owners and decision points. It may look twelve to twenty-four months ahead while specifying near-term commitments more closely.
A use-case list identifies possibilities. A strategy explains the intended direction and limits. The roadmap connects them to delivery: what happens first, what waits, what stops and who is responsible.
Why pilots do not always become routine practice
A technically successful pilot still needs integration into daily work, ongoing funding, an accountable owner and agreement about what it replaces. Without those decisions, the demonstration can remain separate from normal operations.
Deloitte’s 2026 global findings illustrate the gap between pilots and production. A roadmap provides the practical mechanism for crossing it. Consider less visible administrative and support tasks alongside applications that are easy to demonstrate.
Five workstreams to develop together
The use-case portfolio. Maintain a ranked queue with estimated value, effort and ownership. Decide what enters production, what waits and what leaves the portfolio.
Skills and adoption. Train people for their changed tasks. Include checking results, handling exceptions and understanding when human judgement is required.
Data and governance. Define what data can be used, who reviews sensitive cases and how the inventory of uses is maintained. Use the inventory to assess applicable duties, including the EU AI Act where relevant.
Budget by horizon. Include implementation, training, support and ongoing operation. A budget should have an owner and a decision date.
Infrastructure and integration. Plan connections to existing systems, access to data and maintenance. A useful application needs a workable place in the process.
Three planning horizons
Zero to three months: establish the basis and test value. Choose a bounded area, a useful first case and proportionate governance. Measure the result against a baseline.
Three to twelve months: extend and build capability. Add a small number of cases, develop skills by role and make explicit decisions to retire old practices or stop unsuitable pilots.
Beyond twelve months: sustain and scale. Confirm recurring resources, established ownership and revised processes. These horizons are planning guides, not a promise that every case should wait a year to enter production.
Organisation and assessments
Make organisational decisions with a clear assessment
Our assessments identify practical changes to help your teams and processes work together.
Build the roadmap from actual work
Workshops generate possibilities, but test them against the tasks and constraints described by teams. Map the work, consider whether to automate, assist, reorganise or protect each activity, and take the resulting choices to leaders who can commit resources.
Unofficial AI use can provide useful evidence. Ask what employees already use, where it helps and which safeguards are missing. Assess these practices before deciding whether to adopt, change or stop them.
Keep the roadmap active
Use monthly reviews with clear choices: continue, change, expand or stop. Assign a named owner and resources to each active case. Put unowned cases back into the portfolio until someone can take responsibility.
Stopping a pilot should preserve what was learned and release resources deliberately. Review the roadmap when evidence, technology or business priorities change.
A practical test
A manager should be able to explain what changes for their team next quarter, who owns it and what work will stop or change to make room. If the roadmap offers only ambitions and technology names, the delivery decisions are still missing.
Further reading
Frequently asked questions
How does an AI roadmap differ from an AI strategy?
How does an AI roadmap differ from an AI strategy?
The strategy sets direction, purpose and limits. The roadmap specifies cases, sequence, budgets, owners and decision points. Together they connect ambition to delivery.
Where should we start introducing AI?
Where should we start introducing AI?
Start with tasks and a clear decision question. Understand where work is difficult or repetitive, then choose a first case for its practical value. Check the data, risks and integration before committing to a tool.
How long does it take to build an AI roadmap?
How long does it take to build an AI roadmap?
Spentia develops the roadmap over four weeks, followed by six months of follow-up. The scope, participation and decisions required are agreed at the start. The roadmap then evolves as implementation produces evidence.
Should a pilot come before or after the roadmap?
Should a pilot come before or after the roadmap?
Begin with a short scoping exercise and a case grounded in actual work. Develop the roadmap alongside the pilot so that a successful test has an owner and a path forward, and an unsuccessful one has clear stopping criteria.
Sources
Deloitte, The State of AI in the Enterprise: The Untapped Edge, 2026. Report overview.
