How to Orchestrate Planning & Prioritization with AI (AI PM Part 3)
Hey, Nacho here. TL;DR: This is Part 3 of the PM Orchestrator series. In Part 1, we introduced the "Orchestrator" mindset: shifting from a "doer" to a "system designer". In Part 2, we built the first part of our engine by orchestrating Execution & Delivery, using AI to automate tasks like status updates and ticket creation to free up hours of our time. Now, we use that reclaimed time to tackle Planning & Prioritization.
Let’s be honest about product planning. For most, it's a painful, broken process.
It’s the endless cycle of opinion-driven debates during long meetings. It’s the constant fire hose of new requests—a feature idea from sales, a vague demand from marketing, a pet project from an executive—each arriving with little more than a title and an urgent demand for prioritization.
It ends up in a constantly changing roadmap, not because of new evidence, but because of a lack of alignment. We spend our time defending our plans instead of improving them.
This isn't high-impact work. But it’s the reality of most PMs.
Most try to fight this chaos, struggling to impose order on it with ever-more-complex spreadsheets and rigid processes. They always fail. The chaos will not go away.
The "Orchestrator PM" takes a different approach. They don't fight the chaos; they build a system to navigate it. They design a continuous, intelligent engine that sources the right information, structures the conversation, and allows them to make strategic decisions with confidence.
In our last post, we automated the execution engine. Now, let’s use that reclaimed time to build the most critical system of all: your AI-powered Evidence Engine for planning and prioritization.
Phase 1: Structuring the Opportunity Space with AI
The first rule of prioritization is that you can't prioritize what you can't see. Our work is buried under an avalanche of unstructured data: interview notes, support tickets, Slack messages, and survey responses. Yes, even a never-ending backlog.
The Orchestrator’s first job is to create a single, structured map of this messy reality.
The Orchestrator's Playbook:
Your role isn't to manually sift through every piece of feedback or request. It's to design an AI agent that does it for you, transforming raw input into a structured "Opportunity Map."
Example in Action:
You create a custom AI agent and feed it everything: hundreds of interview transcripts from Dovetail, a CSV of support tickets from Zendesk, notes from sales calls in Salesforce, and the one-line email from your CEO about "gamification."
The agent’s job is not to judge or score. Its job is to cluster.
It scans the corpus of text and identifies recurring themes, grouping them into distinct customer problems or needs.
"Users are confused about the pricing tiers."
"Customers are asking for a Xero integration."
"The onboarding flow feels too long."
Even the CEO’s idea gets processed. The AI researches "gamification for B2B SaaS," finds related articles on user engagement, and clusters it with an existing opportunity you've identified around "improving initial user activation."
You’ve just transformed a chaotic fire hose of information into a clean, visual map of your entire opportunity space. This structure is the foundation for everything that follows.
The best part: you can automate the process of putting this output into any management tool of your choice. Even switching from tool to tool became easy!
Phase 2: Connecting Needs to Strategic Outcomes
An opportunity map is useful, but it lacks direction. Why should you work on "improving activation" over "adding integrations"? To answer that, you need to connect customer needs to what the business is trying to achieve.
Many teams struggle here because they don't have a clear picture of their own business levers.
The Orchestrator's Playbook:
You use an AI agent first to help you model your business, then to connect your Opportunity Map to that model.
Example in Action:
You source a new AI space with read-only access to your company's top business intelligence dashboards from Looker or Tableau.
Your first prompt is: "Based on these dashboards, draft a KPI Tree for our top goal, 'Recurring Revenue'. Identify the input metrics that correlate with a change in revenue, and then the input for those second-level metrics, and continue recursively until the lowest-level metrics."
The AI analyzes the data and proposes a structure:
L1: Recurring Revenue
L2: Number of Active Subscriptions
L3: New Customer Acquisition
L4: Website-to-Trial Conversion Rate
L4: Trial-to-Paid Conversion Rate
L3: Customer Retention Rate (inverse of Churn)
L4: Monthly Active Users
L4: Adoption Rate of 'Sticky Feature X'
L2: Average Revenue Per User (ARPU)
L3: Upsell/Cross-sell Revenue
L4: Adoption Rate of 'Premium Add-on Y'
L4: Percentage of Users on 'Enterprise' Tier
You now have an AI-assisted, data-backed model of your business.
Your next step is to feed the agent the Opportunity Map from Phase 1. Your prompt: "For each opportunity cluster, identify which KPI Driver from our tree it would most directly influence. Explain the connection."
Now, "improving initial user activation" isn't just a user need; it's explicitly linked to "Driver 1" of your North Star Metric. You have established a clear connection between customer issues and business value.
Phase 3: Building Your Automated ICE Scoring Engine
With opportunities structured and linked to strategy, you can now evaluate them. The ICE framework (Impact, Confidence, Ease) is a great tool, but manual scoring sessions often devolve into opinion-based debates.
The Orchestrator's Playbook:
You automate the initial ICE score by creating an AI agent that pulls data from source systems. This gives you an unbiased first pass, turning you from a "scorer" into an "editor" who refines the output with your unique context.
Example in Action:
You build an AI workflow that scores each opportunity:
Impact: The AI looks at the opportunity's connected KPI Driver from Phase 2. It then pulls from JIRA historical data on past initiatives that affected that same driver to generate a baseline score. For an opportunity linked to "User Retention," it analyzes past features and their effect on churn to estimate potential impact.
You can also feed it with daily KPI reports to analyze the latest values of the levers and estimate the impact of changes to them.Confidence: The AI assigns a confidence score based on the evidence available. An opportunity sourced from 50 user interviews and a supporting fake door test gets a high score. One sourced from a single comment gets a low score. It quantifies your certainty based on facts, not feelings. Not to mention the CEO’s “Gamification” email 🙂
Ease: The AI connects to Jira, analyzes past epics with similar technical components, and provides a rough, t-shirt size estimate for effort ("S," "M," "L"). We explored in the previous article how connections to GitHub can help LLMs understand the potential affected modules and compare to prior changes to get an estimation better than any PM can ever do.
Moreover, the AI doesn't just give you a score. It gives you a detailed justification. Stakeholders don't care if a score is a '2' or a '3'; stakeholders care about the evidence backing that reason. The AI provides the narrative: "Impact is a '3' because it directly influences our 'Retention' KPI, and it is comparable with X, Y, Z past features that had $$ impact. Confidence is a '2' because while we have strong qualitative feedback, we haven't yet validated demand with a smoke test."
This arms you with the logic needed for any stakeholder conversation.
Phase 4: The Continuous Planning Agent
A roadmap is not a static document created once a quarter. It's a living system that must adapt to new information.
The ultimate goal of the Orchestrator is not to create a perfect plan, but to create a system that can continuously update the plan as reality changes.
The Orchestrator's Playbook:
You design a "Planning Agent" that monitors for new inputs, runs them through your Evidence Engine, and proactively suggests changes to the roadmap. You are no longer reacting to chaos; your system is anticipating it.
Example in Action:
Imagine this workflow:
A new idea is added to your Jira Product Discovery board.
A Zapier automation detects this and triggers your Planning Agent.
The agent runs the new idea through the entire system you've just built. It clusters the opportunity (Phase 1), connects it to a KPI driver (Phase 2), and generates an evidence-based ICE score (Phase 3).
It then compares this new opportunity to the initiatives already planned for the next quarter.
Finally, it sends you a message in Slack:
New Opportunity Analysis: The idea "Add Xero Integration" has been processed.
Score: Impact: 4, Confidence: 3, Ease: 3. Strategic Fit: Directly supports OKR "Reduce churn in SMB segment."
Recommendation: This opportunity has a higher evidence score than "Improve Reporting Exports," which is currently planned for Q4. To accommodate this, I recommend postponing the reporting feature.
Do you want to proceed with this change and have me generate a stakeholder update report explaining the reasoning?
[Yes, update the roadmap] [No, keep as is]
You just received a proactive, data-driven recommendation. You are no longer chasing information; it is being synthesized and delivered to you for a final, strategic decision.
You have become the architect of your own intelligent planning system.
This is the future of product management. It’s not about having all the answers. It’s about building the engine that helps you find them, faster and more reliably than ever before.
Conclusion: The End of the Comfort Zone
Look back at the system we've just designed. For decades, the manual, political, and painful process of planning has consumed our days. Now, we can architect it. We can delegate it.
This reclaimed cognitive space allows the Orchestrator PM to operate at a higher strategic plane. This isn't the automation of product management. It's its elevation. We are being freed from the role of tactical project managers and the busywork that came with it.
The question is: Are we up to the task?
Isn't it a bit scary that the comfort zone where we spent so many hours is gone? Now, we need to truly show our value-add in the strategic layer, without the shield of tactical execution to hide behind. We will be held up to new standards.
Exciting times ahead. :)
In our next article, we’ll dive into Orchestrating Discovery & Validation, and show how to use AI to get to the core of customer problems at lightning speed. Stay tuned.