Why annual stopped working
Annual roadmaps for AI haven’t worked since 2023. The reasons are mechanical: the underlying capability shifts faster than a year, the vendor landscape shifts faster, and the company’s understanding of what’s possible shifts faster.
A roadmap committed to in January is obsolete by April for at least one of these reasons. The roadmap is then either ignored (the team builds what they actually believe in, and the document becomes a fiction) or defended (the team builds against an out-of-date plan, missing the better opportunity that emerged).
Quarterly cadence is the floor that holds. Twelve weeks is long enough to ship meaningful work and short enough to re-plan with current information. Below is the template that has held up across our practice.
The template structure
A working quarterly AI roadmap fits on one page and has four sections. Each section is short by design.
Section 1: Shipping this quarter (3-5 items)
Initiatives that are actively in build right now. Each line has:
- Initiative name (4-6 words).
- Owner (a person, not a team).
- One-sentence success definition.
- Kill criterion (what triggers killing this if it goes badly).
- Cost band (5-figure / 6-figure / 7-figure).
That’s the full row. If you can’t fit an initiative into that shape, scope it tighter or move it to the next section.
Section 2: Next quarter (2-4 items)
Initiatives committed to scoping during the current quarter so they’re ready to start at the quarter break. Each line has:
- Initiative name.
- Owner.
- Status of scoping (just started, mid-scope, ready to start).
These don’t have budgets yet. They might not survive scoping. The point of putting them in Section 2 is to commit to the work that lets them either advance or die deliberately.
Section 3: Backlog / under evaluation
A simple list. One line per item. No commitment, no owner required. Ideas, vendor pitches, internal proposals. Items get promoted when they earn it.
This section is allowed to be long. It’s the public memory of everything the team has considered. Items can stay here for quarters without prejudice.
Section 4: Foundation work
Specifically separate from the initiative columns. Foundation work is the platform investment that makes all subsequent initiatives possible — gateway, observability, eval harness, security tooling.
This section is short (2-3 items) and managed differently. Foundation work is not optional and doesn’t compete with initiatives for prioritization. It gets the engineering capacity it needs.
What goes where: the discipline
The temptation in any roadmap is to put more things in the “shipping” column than you can actually ship. The discipline:
Section 1 is bounded by capacity. If your team has 4 pairs of engineers and each pair can ship 1 substantial initiative per quarter, Section 1 has at most 4 items. Period. Wishful staffing produces wishful roadmaps.
Section 2 is bounded by scoping capacity. Scoping an initiative properly takes meaningful work — interviews, architecture sketches, cost estimates. Section 2 should have at most 4 items because that’s how many you can scope to depth in a quarter.
Section 3 has no bounds. Be generous. Items in Section 3 cost nothing.
Section 4 has no negotiation. Foundation work happens or the rest of the roadmap collapses on itself.
The meeting cadence
The roadmap is reviewed in three meetings:
Monthly working review (45 minutes)
Open the roadmap. Update Section 1 status. Note anything that moved between sections. Flag any kill candidates. Update Section 4 status.
This meeting is operational. It’s the engineering and product leads working the document. No leadership theater.
Quarterly planning (2 hours)
End of each quarter. Section 1 items get retired (shipped, killed, or rolled into Section 2 with revised scope). Section 2 items get promoted to Section 1 with finalized scope. Section 3 items are reviewed and some promoted to Section 2. Section 4 is reviewed and updated.
This meeting includes wider leadership — engineering, product, ops, finance. The product is the new quarter’s roadmap.
Bi-weekly initiative reviews (30 minutes each)
For each Section 1 initiative, the owner runs a 30-minute review every two weeks. Status, blockers, kill criterion check. This is initiative-level operations, not the roadmap-level meeting.
That’s the entire cadence. Once you’ve run this for a quarter, the rhythm sustains itself.
What this template prevents
The shape of this template specifically prevents common roadmap pathologies:
Over-commitment. Hard-bounded Section 1 forces honesty about what’s actually shipping.
Zombie initiatives. Section 2 items that haven’t been scoped after a quarter must either be killed or moved back to Section 3. They cannot linger.
Foundation neglect. Section 4 is structurally protected.
Drift between strategy and execution. Initiatives have to be specific enough to fit the template. Vague strategic intentions get pushed to Section 3 where they belong.
Hidden cost of failed initiatives. Killed initiatives are documented. The cost of killing is visible, which makes the team kill earlier next time.
What it looks like populated
A snapshot of a real quarterly roadmap from a $40M-revenue client, mid-second-year of structured AI work:
Shipping this quarter (Q2):
- Outbound prospect research agent. Owner: VP Sales. Success: SDRs accept 60%+ of AI-drafted research. Kill: if acceptance is below 30% by week 6. Cost: 6-figure annual.
- Customer-facing chat with handoff. Owner: Head of CX. Success: 70%+ of routine inquiries resolved without human touch. Kill: if customer NPS on AI-handled inquiries drops below baseline. Cost: 7-figure annual.
- Internal knowledge agent (rollout to all of engineering). Owner: CTO. Success: 50+ weekly active engineers. Kill: if engagement is below 20 weekly users by week 8. Cost: 5-figure.
Next quarter (Q3):
- Multi-language document translation pipeline (currently scoping with ops).
- AI-driven QA for product (architecture sketch in progress).
- Internal forecasting model (data prerequisite work scoping).
Backlog:
- Voice-based meeting transcription (vendor evaluations).
- Customer-facing recommendation agent (capability not mature).
- Sales pricing assistant (data quality blockers).
- Automated contract review (legal hesitation).
- AI for accounts payable (low CoD).
- …several more lines
Foundation work this quarter:
- Per-tenant cost meter (in build).
- Eval harness expansion (planning).
- Gateway hardening (just started).
The whole document is one page. The team can read it in two minutes. The board reviews it without a binder.
What this changes about leadership conversations
Two changes worth noting:
The CEO conversation is different. Instead of “what’s the AI strategy” — vague, hard to answer — the conversation becomes “look at this page, here’s what’s shipping, here’s what’s next, here’s what we’re considering.” Specific, defensible, actionable.
The CFO conversation is different. The cost bands in Section 1 produce a concrete annual cost projection. The CFO can budget. Surprise spend goes down.
The board conversation is different. Instead of a 40-slide AI deck, you show the roadmap and the metrics. Board members understand it because it’s the same shape as every other operational document they see.
When the template doesn’t fit
Two cases where the template needs adjustment:
Very early-stage companies. When you have one engineer and one AI initiative, you don’t need a quarterly roadmap — you need a focus list. The template is overkill.
Very large enterprises with regulated portfolios. Hundreds of AI initiatives need more structure. Each business unit or program might have its own quarterly roadmap, with a central program leadership layer above them.
For most mid-market and growth-stage companies — the bulk of where AI work is happening in 2026 — the template above is the right shape.
The take
Annual AI roadmaps don’t survive contact with reality. Quarterly roadmaps do. The template — Shipping / Next / Backlog / Foundation — fits on one page, runs on a monthly cadence, and survives the inherent volatility of AI work. Adopting the cadence is a structural change with structural benefits. Companies that run it produce better outcomes than companies that don’t, every year.
Roadmap shape is one of the first deliverables in Fractional CAIO engagements. If your AI roadmap isn’t surviving its own quarter, schedule a call.
