AI Product Quarterly Planning: How to Plan AI Roadmaps in 90-Day Increments
TL;DR
Annual AI roadmaps die by week six. The model changed, a competitor shipped, a customer pivoted. Quarterly planning is the right cadence for AI — long enough to ship meaningful work, short enough to absorb capability shifts. This guide gives you the planning format, the kickoff ritual, and the mid-quarter and end-quarter review patterns that turn 90 days of work into compounding momentum.
Why Quarterly Beats Annual for AI
Annual AI roadmaps are stale within a quarter. Frontier model capabilities shift, costs drop, competitors ship, customers re-baseline expectations. The quarterly horizon is where AI work actually compounds: long enough for real shipping, short enough for re-planning before reality drifts too far from plan.
Quarterly theme + 3 bets
One unifying theme per quarter (e.g., 'reliability') and 3 specific bets that ladder up to it.
Each bet has a kill criterion
Specific signal that triggers killing or pivoting the bet mid-quarter. Pre-decided, not improvised.
20% capacity reserved for response
Held back for new model capabilities, urgent customer asks, or competitor responses. Budgeted; not forgotten.
Weekly anchor reviews
30-minute team review every Monday: are bets on track, blocked, or at-risk? Course-correct early.
The Quarterly Planning Doc — One Page Max
A quarterly plan that takes more than one page won't be re-read. The discipline of fitting on one page forces the right level of abstraction.
Theme (1 sentence)
"Q3 theme: take our AI feature from delight to reliability." Sets the lens.
Three bets (3 sentences each)
Each bet: what it is, why it matters, kill criterion. No more.
Specific success metrics
End-of-quarter target for each bet, with a baseline. Concrete numbers.
Top 3 risks (1 sentence each)
Things that could break the plan. With one-line mitigations.
What we're explicitly NOT doing
Underrated section. Listing the rejected bets prevents scope creep mid-quarter.
Kickoff Ritual — Week 1
Quarterly planning is a ritual, not a meeting. Done well, it aligns 10-50 people around what matters and creates the energy to ship for 12 weeks. Done badly, it's a status update that nobody remembers by week 4.
Day 1: Pre-read circulated
1-page draft plan shared 48 hours before the kickoff. Read in advance; meeting is for debate, not introduction.
Day 3: Cross-functional kickoff (2 hrs)
Eng, design, PM, marketing, sales aligned. Each function's top concerns voiced. Plan revised live.
Day 5: Final plan published
1-pager finalized. Slack/email-broadcast to extended stakeholders. Becomes the anchor doc for 12 weeks.
Day 7: First weekly review
Team gets into the cadence early. Friction surfaces while it's still cheap to fix.
Plan AI Roadmaps That Survive Reality
The AI PM Masterclass walks through quarterly planning rituals with real plans and instructor feedback — taught by a Salesforce Sr. Director PM.
Mid-Quarter Review (Week 6)
What we said we'd ship vs. what we're shipping
Honest delta. If you're behind, name it and re-baseline. If you're ahead, ask whether to add scope or compound on quality.
What changed in the world
Model releases, vendor news, competitor launches, customer signals. Decide explicitly which to absorb into the plan and which to ignore.
Re-decide on at-risk bets
Apply kill criteria. If a bet is failing, kill or pivot now. The 6-week mark is the right time; week 11 is too late.
Communicate the update broadly
Stakeholders outside the core team need to know the quarter's direction. A short written update beats a meeting.
Quarterly Planning Anti-Patterns
30 bets in a quarter
"Top 30" means top zero. Three bets max. The discipline of cutting is what makes shipping possible.
No kill criteria
Bets without kill criteria run until quarter-end on momentum alone. Kill criteria force honest mid-quarter decisions.
Plan in private, share at end
Planning that doesn't involve eng and design produces plans that can't ship. Cross-functional from day 1.
100% capacity allocated
AI quarters always include surprises. 80% allocated, 20% reserved is the real ratio.
Treating the plan as immutable
Plans should bend with reality. Bend the plan; don't break the plan's discipline.