The problem with most AI roadmaps
They are written by people who will not build them. The list is long, the estimates are guesses, and nothing on it connects to the operational bottleneck that actually costs money. And when a project does start, it stalls on foundations nobody checked: data that cannot be reached, systems that cannot change safely, no tests, no rules for what may go into a prompt.
We approach both as engineers who will be asked to deliver the first item.
Part one: find the opportunities
- Find the expensive bottleneck. We start with where time and money go today: onboarding, manual data mapping, support volume, release cycles.
- List the candidates. For each, we describe what AI would do, what data it needs, and what could go wrong.
- Size them honestly. Effort, risk, data readiness and payback, scored the same way so you can compare them.
- Choose a first project. Small enough to ship, valuable enough to matter, with success criteria agreed up front.
Part two: check the foundations
Data. Where it lives, how clean and connected it is, and what it would take to make it usable.
Systems and integration. How your platforms exchange data today, and whether AI features can be added without a rewrite.
Delivery. Test coverage, release process and tooling. AI speeds up teams that already ship reliably; it amplifies chaos in teams that do not.
Governance. Access control, documentation, data-handling rules and review steps, especially important in regulated industries.
What you get
- A ranked shortlist of AI use cases with effort and value estimates
- A readiness view across data, systems, delivery and governance, with the three to five fixes that unblock the most work
- A recommended first project with scope, team and success measures
- Guardrails: what data stays out, who reviews output, how results get measured
- A plain-language summary your leadership can decide on
Why us
We do the foundation work for a living: designing the data integration layer for an enterprise platform, bringing governance and documentation to a healthcare provider's IT estate, and taking test automation from zero to 65% coverage in four months for a digital health platform. The discovery draws on the same people.
Frequently asked questions
How long does it take?
Typically a few weeks, depending on how many teams we need to talk to. We agree the timeline before we start.
Do we have to hire you to build it afterwards?
No. The roadmap is yours. Many clients do ask us to build the first project, and the discovery is designed so we could start soon after if you do.