A practical guide to creating an AI implementation roadmap

Why a roadmap, not a wishlist

Most organisations we speak to have an AI wishlist: a scattered set of ideas from vendors, board members, and enthusiastic engineers. Someone saw a demo. Someone read an article. Someone's competitor announced a chatbot. None of it is sequenced, costed, or tied to a business outcome, and the list grows every quarter because nothing on it ever gets finished.

A roadmap is different. It sequences those ideas against capability, risk, and value so the business can actually deliver. It tells you what happens first, what depends on what, and what you are deliberately choosing not to do this year. If you cannot say why an initiative sits in quarter two rather than quarter four, you have a wishlist with dates on it, not a roadmap.

A good AI implementation roadmap answers four questions:

1. What are we trying to change about the business in the next 12 to 24 months?
2. Which AI initiatives materially move those outcomes?
3. What foundations (data, skills, governance, tooling) must be in place first?
4. Who owns each step, and how will we know it worked?

The five phases we recommend

Phase 1: Anchor to strategy. Start with the business model, not the technology. If you cannot point to the customer segments, revenue streams, or cost structures a piece of AI will affect, park it. "We should do something with AI" is not a starting point; "our claims processing cost per unit is too high and the volume is growing" is.

Phase 2: Assess readiness honestly. Data quality, platform maturity, security posture, and people skills all constrain the pace you can move. A candid readiness assessment saves you from committing to a delivery date you cannot hit. This is the phase organisations most often skip, and it is the one that causes the most damage later.

Phase 3: Prioritise a portfolio. Balance quick wins that build confidence with foundational investments that unlock later work. Aim for a mix of internal productivity, customer experience, and revenue-generating initiatives, rather than betting everything on one flashy customer-facing project.

Phase 4: Sequence and resource. Convert the portfolio into quarters. Attach owners, budget, and dependencies. Be explicit about what you are not doing this year — an unstated "no" is where scope creep comes from.

Phase 5: Govern and adapt. Set a cadence to review progress, retire initiatives that are not working, and re-plan every quarter. Roadmaps that do not change quickly become fiction.

A worked example

Take a mid-sized insurer wanting to cut claims handling time. The wishlist version: "build an AI claims assistant." The roadmap version looks more like this.

Quarter one: readiness assessment finds claims data is split across three systems with inconsistent status codes — a data engineering task, not an AI task. Quarter two: standardise the data model and stand up a single claims event log; in parallel, run a low-risk pilot using AI to summarise adjuster notes for internal review only, no customer-facing output. Quarter three: expand the summarisation tool to all adjusters and measure time saved; begin the governance work (sign-off process, audit trail, human review threshold) needed before anything touches a customer directly. Quarter four: pilot an AI-assisted first-notification-of-loss triage tool, with the governance framework from quarter three already in place, and a named product owner accountable for its accuracy.

Notice what is not in quarter one: the customer-facing AI. It is the most visible idea and the one leadership usually wants first, but it depends on data and governance work that has to happen earlier. Sequencing this correctly is the entire point of a roadmap.

Common mistakes

• Treating the roadmap as a one-off document rather than a living artefact that gets revisited every quarter.
• Skipping the readiness assessment because leadership finds it uncomfortable to hear that the data isn't ready.
• Loading every quarter with new initiatives instead of protecting time to finish existing ones — a roadmap with no slack is a roadmap that slips.
• Forgetting change management. The model going live is the middle of the project, not the end; adoption, training, and monitoring take as long as the build.
• Letting the roadmap outrun governance. If phase five review meetings are the first time anyone asks "should we still be doing this," you have gone too long without checking.

What to do next

If you are starting from a wishlist, do not try to sequence all of it at once. Pick the two or three initiatives with the clearest link to a business outcome, run them through a readiness check, and sequence just those. Set a date, four to six weeks out, to review what you learned and add the next tranche. A roadmap built in one sitting and never revisited is worse than no roadmap at all.

Where Florence helps

Florence walks you through each of these phases in order, keeping your organisational context in one place so it doesn't have to be reconstructed at every stage. The Business Model Canvas anchors initiatives to actual revenue streams and cost structures rather than vague ambition. The AI Readiness Assessment gives you the honest, phase-two view of your data, skills, and governance gaps before you commit to dates. Insights and Strategy turn that picture into a prioritised, sequenced portfolio, and Projects and Briefs produce the concrete artefacts — owners, scope, dependencies — that a roadmap needs to survive contact with delivery.