Roadmapping AI Initiatives

Core Concepts

The Three-Horizon Framework

The three-horizon model organizes AI initiatives by time and maturity, not just priority. Each horizon has a different goal:

H1 is not the "easy stuff" that you do while waiting to get to the real work. H1 is the real work. Without clean data pipelines, clear feedback loops, and at least one successful AI workflow under your belt, H2 initiatives will fail or stall.

At X-company, H1 covers embedding AI into internal workflows: PRD drafting, onboarding document generation, and support ticket triage. H2 moves to a customer-facing capability: an AI-assisted churn prediction model. H3 explores AI-native feature differentiation inside the product itself.

Sequencing Principles

The order of initiatives on a roadmap is not a preference: it is a logic. Three principles govern sequencing:

Dependency-first. An initiative that depends on infrastructure, data quality, or team capability built by another initiative must come after it. Not in parallel. After.

Confidence before complexity. Teams new to AI adoption should not begin with their highest-ambition initiative. Start where the feedback loop is short and the blast radius of failure is small. Build organizational confidence alongside technical capability.

Value visibility. Early initiatives should produce outcomes that leadership and the broader team can see. This is not about showing off: it is about sustaining investment and trust through the longer build phases.

Dependency Mapping

A dependency is any condition that must be true before an initiative can succeed. Dependencies are not always technical. They include:

• Data dependencies: the downstream initiative requires clean, structured data that the upstream initiative produces or prepares
• Capability dependencies: the team needs a skill or workflow pattern that an earlier initiative installs
• Trust dependencies: stakeholder or customer acceptance of AI in a lower-stakes context before a higher-stakes one

At X-company, the H2 churn prediction model has a hard data dependency on H1. The model needs structured, consistent records of customer behavior, support interactions, and product usage signals. If H1 support triage runs on messy, inconsistently tagged tickets with no feedback loop for correction, the training data for churn prediction is unreliable. H2 breaks.

This dependency is not obvious from a feature list. It only becomes visible when you map what each initiative needs to succeed and trace those needs back through the roadmap.

What Belongs on an AI Roadmap

Not every idea that mentions AI belongs on an AI roadmap. Include an initiative if it:

• Uses AI as a meaningful part of the solution, not a label applied after the fact
• Has a clear owner and a success metric
• Has its dependencies identified and scheduled

Exclude an initiative if it:

• Is a feature idea without a defined problem to solve
• Depends on a capability, dataset, or infrastructure that has no scheduled delivery
• Is on the list because it sounds impressive, not because it solves a real problem

A roadmap with ten items where four are unexecutable because their dependencies are missing is worse than a roadmap with six items where all six are ready to move.

Key Points

• The three-horizon model sequences AI initiatives by dependency and maturity, not just ambition
• H1 activates and builds foundations; H2 scales on those foundations; H3 differentiates
• Every initiative must have its dependencies made explicit before it is placed on the roadmap
• Roadmaps must be filtered: not every AI idea belongs on them

Discussion Prompt Which of your current AI ideas has an undeclared dependency that would block it from succeeding if you started today?

---

Actionable Takeaways

1. List every AI initiative your organization is considering. Before assigning any to a horizon, write down what each one needs to succeed: data, capability, infrastructure, team readiness.

2. Check each initiative's dependencies against your current state. If a dependency is not already in place or scheduled, the initiative cannot go into H1 or H2 without adjustment.

3. Assign initiatives to horizons based on dependency order, not on which ideas leadership finds most exciting. If the exciting idea is in H3, put it in H3.

4. For every H1 initiative, write one sentence describing what it produces that H2 or H3 will depend on. If you cannot write that sentence, the H1 initiative may not be pulling enough weight.

5. Remove any initiative from your roadmap that does not have a defined problem, a success metric, and a named owner. Park it in a backlog, not on the roadmap.

---

Practical Examples

X-company's Three-Horizon Roadmap

X-company's leadership team starts by listing their AI ideas without filtering. They have seven. After applying dependency analysis and the three-horizon framework, their roadmap takes this shape:

H1: Activate (Q1–Q2)

All three are internal or low-risk customer-facing. Feedback loops are short. If a PRD draft is poor, the PM catches it before it affects anyone. If a support triage misroutes a ticket, a human agent corrects it. The team builds confidence and installs the data infrastructure the next horizon depends on.

H2: Scale (Q3–Q4)

The churn prediction model does not appear out of nowhere. It consumes the structured support ticket data that H1 triage produces. If H1 triage is not running by end of Q2 with clean output, H2 slides. This is the roadmap making that dependency explicit and traceable.

H3: Differentiate (Next Year)

X-company's H3 is deliberately undefined at this stage. They know the direction: AI capabilities inside the product that professional services firms cannot get elsewhere. They do not know the specific features because they need the learning from H1 and H2 before they can design them responsibly.

What X-company Removed from the Roadmap

Two of X-company's original seven ideas did not make it onto the roadmap:

• An AI-powered contract review feature for law firm customers. The idea had no defined problem statement, no customer research, and would require training data X-company did not have. Moved to the product discovery backlog.
• An AI meeting summarizer integrated into the project management tool. This duplicated existing third-party tools the customers already use. No clear differentiation. Parked pending further research.

Neither removal was a rejection of the idea permanently. Both were a recognition that the ideas were not ready for a roadmap.

---

Implementation Workflow

Use this workflow to build a three-horizon AI roadmap for your organization.

1. List every AI initiative under consideration. Include ideas from leadership, product, engineering, and customer-facing teams. Do not filter yet. Write each one as a one-sentence problem statement: "We want to [do X] so that [Y outcome]."

2. Write the dependency profile for each initiative. For each item on your list, answer three questions:

   • What data does this initiative need, and do we have it in a usable form?
   • What team capability does this initiative require, and do we have it?
   • What earlier AI success does this initiative assume we have already demonstrated?

3. Filter the list. Remove any initiative that does not have a defined problem, a success metric, and a named owner. Move those to a discovery backlog. You are not discarding them: you are not scheduling work that is not ready to be scheduled.

4. Identify hard dependencies between initiatives. Draw a line from any initiative that needs something produced by another. These lines define your sequencing constraints. An initiative with an upstream dependency cannot be in an earlier horizon than the initiative it depends on.

5. Assign initiatives to horizons. Use this as your guide:

   • H1: No unresolved dependencies. Short feedback loop. Builds internal capability or data infrastructure.
   • H2: Depends on H1 outputs. Higher complexity or higher customer impact.
   • H3: Depends on H2 learning or capability. Highest ambition or product differentiation potential.

6. Validate H1 for weight. For each H1 initiative, write one sentence: "This produces [X], which H2 initiative [Y] depends on." If an H1 initiative produces nothing that a later horizon needs, reconsider whether it belongs on the AI roadmap or is simply a product improvement without an AI dependency.

7. Write the roadmap document. Format it as a table with columns for initiative, horizon, owner, success metric, and key dependency. One row per initiative. This is the document you bring to leadership alignment and quarterly planning.

8. Schedule a dependency review at the end of each horizon. Before moving from H1 to H2, confirm that the H1 outputs that H2 depends on are actually in place and in good shape. Churn prediction built on six months of inconsistently tagged support data will underperform. A thirty-minute review before committing to H2 prevents that.