The AI roadmap
The AI roadmap is the output of the strategy process, not a separate exercise. The steering committee’s decisions become a defined plan: which use cases go first, what each one delivers and who owns it, which shared capabilities are built and when, and which gates must be passed before more money is committed. It is a funding and dependency plan as much as a delivery plan, and the committee revisits it as results arrive.
What a defined roadmap contains
- Use cases in sequence: the approved initiatives, in the order the committee funded them, each traced to a stated outcome.
- Deliverables per use case: what goes live, for which users, and the measurable target it is judged against.
- Owners: a business owner for each outcome and a delivery owner for each deliverable.
- Shared capabilities: the platform pieces (retrieval layer, evaluation tooling, model gateway) and the initiative that first builds each.
- Gates and dependencies: what must be true before the next tranche of money, and what each use case waits on.
Phases over 12 to 24 months
- 1
First opportunities
A few use cases with owners and data.
Baselines measured
- 2
Proof of concept
Reach acceptable quality on real data.
Quality met, cost known
- 3
Pilot
Real users, measured outcome.
Benefit proven
- 4
Production
Full scale, with operational ownership.
Benefit held a budget cycle
- 5
Shared platform
Turn what was built into reusable services.
A second use case runs on it
Investment gates
Each gate releases the next tranche of funding only when its criteria are met. Gates are where initiatives are stopped as well as advanced. A proof of concept that cannot reach quality on real data is closed and the lesson recorded; it does not drift into an extended pilot because the team is attached to it.
Dependencies
- Shared knowledge base before agents. An agent that acts on behalf of customers or staff needs accurate, owned content to reason from. Fix and govern the knowledge base first.
- Data platform before cross-domain use cases. Churn or fault-correlation initiatives that join CRM, billing and network data need that data reconciled and accessible in one governed place.
- APIs before actions. An assistant that only reads can tolerate replicas; one that raises a trouble ticket or changes an order needs a proper API in front of the system of record.
- Security policy before broad rollout. Rules for customer data in prompts and model access must exist before the first production launch, not after it.
Workforce enablement and benefit tracking
The roadmap carries a workforce line alongside the technology line: training for staff who use AI outputs, skills development for those who build and run them, and consultation where roles change. Benefit tracking is also on the roadmap, with review points after each production launch, so the portfolio is judged on outcomes rather than on the number of initiatives started.
Sequences investment so shared capabilities are built once and each initiative has its dependencies in place.
It does not guarantee that pilots will reach quality, or that benefits will be realised by the business.
Gates and dependencies slow individual initiatives down, and the roadmap must be re-planned as lessons arrive.