All 7 run on whichever AI agent you already use -

Microsoft Copilot - connected to the tools listed with each use case.

| # | Use case | Key integrations | Est. time saved About the estimate | | - | --------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | 1 | Analyze customer feedback |

Product feedback

CRM | **2** hr/wk | | 2 | Synthesize user research |

Meetings

Design

Docs

Docs

Product analytics

Tasks

Docs | **1** hr/wk |

Analyze customer feedback

Best first workflow - turn scattered feedback into an evidence-backed queue

2 hr/wkest. time saved How this estimate is calculated

How it's done today

You read support conversations, CRM notes, sales calls, surveys, and feedback tools, tag them inconsistently, merge duplicates, and manually turn recurring requests into a product summary.

How AI helps

On a weekly schedule, your agent gathers new feedback, groups repeated problems, keeps source links and customer context, and prepares themes, examples, confidence, and follow-up questions.

Weekly feedback cutoff new items since last review

Your AI agent

Deduplicates related feedback, separates problems from requested solutions, and supports every theme with source IDs and representative examples.

Theme brief

Evidence table

Follow-up queue

How to set it up

Step 1Connect your tools

Required

Product feedback

Reads new feedback, interview notes, tags, account IDs, and source links.

Recommended

CRM

Reads segment, lifecycle, deal, and account context needed to interpret the feedback.

Optional

Help desk

Reads support conversations and issue outcomes.

Docs

Writes the weekly synthesis and source-linked evidence table.

Export the week’s feedback with source IDs and account fields. The workflow works from a spreadsheet as long as each finding can still be traced to the original item.

Step 2Build the workflow

Have a clear review window, product area, customer segments, existing taxonomy, and 20 past examples already labeled as distinct, duplicate, actionable, or noise.

```text Setup prompt theme={null} Help me build a customer-feedback synthesis workflow. Each week, turn new feedback into a source-linked product brief. 1. Ask for the product area, review window, sources, segments, and the decision this synthesis should support. 2. Import only new items and preserve source ID, link, date, account, segment, verbatim problem, and any stated workaround. 3. Separate the underlying problem from the customer's proposed solution. 4. Group genuinely repeated problems, retain outliers, and count unique accounts rather than raw messages. 5. For each theme, provide evidence, affected segments, severity signals, confidence, representative examples, and unresolved questions. 6. Do not create roadmap items; prepare a review queue with source links. 7. Test on duplicates, conflicting feedback, and one important outlier. ```

What good looks like

Every theme should trace to real feedback, counts should use the correct unit, problems should not be collapsed merely because they share words, and the PM should see both repeated patterns and meaningful outliers.

Step 3Turn it on

Choose your trigger

Run weekly over feedback created since the last successful cutoff. Exclude spam, internal tests, and items without a source ID; retain low-frequency feedback rather than silently dropping it.

What runs without you

After four weekly cycles where the themes held up, let the brief generate and circulate on its own - the PM reads it instead of assembling it. Interpretation, and anything that touches the roadmap, stays human. Audit the theme taxonomy monthly; drifting labels quietly merge problems that deserve separate answers.

Pairs well with synthesize user research and prepare roadmap decisions - themes from the field become evidence in the decision pack.

Synthesize user research

Best for turning several interviews into findings without losing the evidence

2 hr/wkest. time saved How this estimate is calculated

How it's done today

After each interview you clean notes, tag observations, pull quotes, compare participants, and assemble findings while trying not to overgeneralize from the loudest conversation.

How AI helps

Your agent structures each transcript against the research plan, extracts evidence, compares participants, and drafts findings with participant counts, counterexamples, quotes, and open questions.

Interview set completes transcripts and plan available

Your AI agent

Codes observations against the research questions, compares patterns across participants, and preserves quotes and counterevidence.

Research synthesis

Evidence matrix

Open questions

How to set it up

Step 1Connect your tools

Required

Meetings

Reads complete transcripts, speakers, timestamps, and interview dates.

Product feedback

Reads the research plan, participant attributes, codes, and prior findings.

Writes coded observations and a draft synthesis.

Optional

Docs

Reads the discussion guide and synthesis template.

Writes the source-linked report.

Upload de-identified transcripts and a participant table manually. Keep stable participant IDs so quotes and patterns can be checked without exposing unnecessary personal data.

Step 2Build the workflow

Have the research question, discussion guide, participant criteria, consent rules, segment fields, and examples of the evidence standard your team uses for a finding.

```text Setup prompt theme={null} Help me build a user-research synthesis workflow. When a research round closes, prepare an evidence-backed synthesis for review. 1. Ask for the research questions, participant criteria, segments, discussion guide, consent limits, and synthesis template. 2. For each transcript, extract observations relevant to the research questions with participant ID, timestamp, concise note, and supporting quote. 3. Keep behavior, stated opinion, researcher interpretation, and recommendation distinct. 4. Compare patterns across participants and segments; report counts as participants, not number of mentions, and preserve counterexamples. 5. Draft findings with evidence strength, quotes, implications, and open questions. 6. Do not generalize beyond the sample or turn findings directly into requirements. 7. Test on a clear pattern, a split pattern, and a contradictory interview. ```

What good looks like

Findings should answer the research questions, link to participant evidence, show sample and segment limits, retain counterexamples, and make the researcher’s interpretation distinguishable from what participants actually said or did.

Step 3Turn it on

Choose your trigger

Run after the planned interview set closes or at a defined interim checkpoint. Exclude unconsented recordings and transcripts without participant IDs or a research plan.

What runs without you

Synthesis stays researcher-reviewed permanently - the samples are too small and the stakes too high for autopilot. What can graduate is the coding: after researchers accept three rounds, let transcripts be coded automatically as they arrive, so synthesis starts from structured evidence. Reconfirm consent limits and the discussion guide at the start of every study.

Pairs well with analyze customer feedback and draft product requirements - interview evidence grounds both the themes and the requirements.

Build product prototypes

Best for testing a workflow before committing design and engineering time

2 hr/wkest. time saved How this estimate is calculated

How it's done today

You turn a concept into screens and states, write sample content, wire interactions, revise after feedback, and spend time polishing pieces that may be discarded after one test.

How AI helps

Your agent turns an approved problem and flow into a disposable interactive prototype with realistic states, sample data, and test scenarios that the team can critique quickly.

Concept ready to test workflow and learning goal clear

Your AI agent

Maps scenarios into screens and states, builds the interaction with realistic data, and prepares a test script and known limitations.

Interactive prototype

Test scenarios

Feedback log

How to set it up

Step 1Connect your tools

Required

Design

Reads design system, existing components, and relevant product flows.

Writes a separate prototype or branch, not production UI.

Docs

Reads the problem, scenarios, constraints, decisions, and learning goals.

Writes test instructions and limitations.

Optional

Product feedback

Writes structured observations from prototype tests.

A coding agent can create a standalone local prototype from screenshots and design tokens, while a design agent can create a clickable Figma flow. Keep it clearly separated from production code.

Step 2Build the workflow

Have an approved problem, target scenario, design system, existing flow, required states, sample data, and a specific learning question.

```text Setup prompt theme={null} Help me build a rapid product-prototype workflow. For an approved concept, create a disposable prototype that tests one product question. Do not modify production code. 1. Ask for the user, scenario, learning goal, entry and exit points, required states, constraints, and what fidelity is actually needed. 2. Read the design system and existing flow; reuse its patterns and terminology. 3. Map the happy path, empty, loading, error, permission, and recovery states. 4. Build a separate interactive prototype with realistic but synthetic data. 5. Label shortcuts, unsupported behavior, and decisions that are intentionally unresolved. 6. Create three test scenarios and a short observation guide tied to the learning goal. 7. Test the workflow on a new flow, a changed interaction, and an edge state. ```

What good looks like

The prototype should answer the named learning question, reflect the real product’s language and states, use safe sample data, expose its shortcuts, and remain easy to discard rather than masquerading as production-ready code.

Step 3Turn it on

Choose your trigger

Run manually after the problem and learning goal are approved. Exclude concepts that still need foundational research or prototypes that would require real customer data.

What runs without you

Prototypes never graduate - they stay isolated from production and reviewed before anyone outside the team sees them. The discipline that matters comes at the end: archive or delete each one once its question is answered, because yesterday's prototype circulating as the plan is this workflow's only real failure mode.

Pairs well with draft product requirements - the prototype answers the questions the PRD still has open.

Draft product requirements

Best for turning approved evidence and decisions into a complete first draft

1.5 hr/wkest. time saved How this estimate is calculated

How it's done today

You move evidence, decisions, constraints, scenarios, dependencies, and unresolved questions into a PRD, then chase stakeholders for details hidden across meetings and tools.

How AI helps

Your agent assembles the approved sources into your template, keeps decisions separate from assumptions, drafts requirements and acceptance criteria, and visibly marks every unresolved question.

Problem is approved evidence and owner named

Your AI agent

Maps source evidence into the PRD structure, traces each requirement to a decision, and marks conflicts or missing inputs instead of inventing scope.

PRD draft

Open decisions

Source appendix

How to set it up

Step 1Connect your tools

Required

Docs

Reads the PRD template, research, decisions, constraints, and product principles.

Writes the draft and source appendix.

Recommended

Tasks

Reads initiative context, dependencies, owners, and linked work.

Writes review tasks for unresolved decisions.

Optional

Meetings

Reads decision transcripts and stakeholder discussions.

Product analytics

Reads baseline behavior and success-measure definitions.

Place the approved source pack in one folder and upload it with the template. Source completeness matters more than live integrations for the first draft.

Step 2Build the workflow

Have the approved problem, evidence, decision log, product and technical constraints, metric baseline, template, and two strong past PRDs.

```text Setup prompt theme={null} Help me build a product-requirements drafting workflow. For an approved problem, create a reviewable PRD from the supplied sources. 1. Ask for the problem, target users, evidence, desired outcome, constraints, decisions already made, non-goals, owner, and approvers. 2. Read the PRD template and source pack; create a source appendix with stable links. 3. Draft context, problem, users, scenarios, functional and non-functional requirements, success measures, dependencies, rollout, and open questions. 4. Trace each material requirement to evidence or a recorded decision. 5. Mark conflicts and [DECISION NEEDED] items; never turn a hypothesis into an approved requirement or invent dates and owners. 6. Prepare acceptance criteria and review tasks, but do not mark the PRD approved. 7. Test on a small enhancement, new workflow, and cross-team dependency. ```

What good looks like

A team should understand the problem and intended result, every requirement should have a source or owner, non-goals and dependencies should be explicit, and open decisions should remain visible rather than polished away.

Step 3Turn it on

Choose your trigger

Run when a discovery item changes to Ready for requirements and has an owner, evidence pack, and approved problem statement. Return incomplete requests with a gap checklist.

What runs without you

Approval is always human - the PRD is a decision record, not a form to fill. After five drafts accepted without structural rework, let the workflow create the document and its review tasks the moment an item reaches Ready. Review the template quarterly so the structure keeps matching how your team actually decides.

Pairs well with synthesize user research and build product prototypes - both supply the evidence a requirement has to trace back to.

Research competitors

Best for a current, source-linked comparison tied to a real decision

1.5 hr/wkest. time saved How this estimate is calculated

How it's done today

You search product pages, documentation, release notes, pricing, reviews, and internal notes, then reconcile different dates and plans into a comparison that quickly becomes stale.

How AI helps

Your agent researches a defined question, captures dated primary evidence, compares the same dimensions, and produces a brief with source links, unknowns, and implications for the decision.

Decision question set competitors and dimensions named

Your AI agent

Collects current evidence from comparable sources, normalizes it into one frame, and separates observed facts from product interpretation.

Competitive brief

Evidence table

Change watchlist

How to set it up

Step 1Connect your tools

Required

Docs

Reads the decision question, comparison template, and prior brief.

Writes the cited comparison and implications.

Recommended

Files

Reads saved product docs, release notes, pricing captures, and internal research.

Optional

CRM

Reads win-loss notes and customer comparisons that add market context.

Public web research is enough for a first version. Give the agent the exact question and require dated source links; add internal win-loss evidence manually when relevant.

Step 2Build the workflow

Have the decision, competitor set, comparison dimensions, geography, plan or persona, cutoff date, and a list of preferred primary sources.

```text Setup prompt theme={null} Help me build a competitor-research workflow. For a defined product decision, create a current, source-linked comparison. 1. Ask for the decision, competitors, audience, geography, product plans, comparison dimensions, cutoff date, and known internal evidence. 2. Research official product pages, documentation, pricing, release notes, and other primary sources first; record URL and publication or capture date. 3. Compare like with like and mark plan, region, beta, or availability differences. 4. Separate observed capability, vendor claim, customer evidence, and our inference. 5. Highlight meaningful differences, unknowns, recent changes, and implications for the stated decision - not a generic feature checklist. 6. Deliver the brief, evidence table, and a short monitoring list. 7. Test on a direct, adjacent, and apparently similar competitor. ```

What good looks like

Every material comparison should be current and source-linked, plan and regional differences should be explicit, unknowns should remain unknown, and implications should connect directly to the product decision.

Step 3Turn it on

Choose your trigger

Run manually for a roadmap or positioning decision and schedule a targeted refresh before major planning cycles. Exclude broad “track everything” requests.

What runs without you

Interpretation stays with the PM - competitive conclusions drive real bets. After three accurate briefs, automate only the source refresh, so the evidence table stays current between decisions. Review the watchlist monthly and prune it; tracking everything is how competitor research becomes noise.

Pairs well with prepare roadmap decisions - the competitive brief is written for exactly that decision meeting.

Analyze product usage

Best for answering a bounded behavior question with governed events

1.5 hr/wkest. time saved How this estimate is calculated

How it's done today

You translate a product question into events and cohorts, inspect tracking definitions, build funnels or paths, reconcile totals, and explain where users diverge.

How AI helps

Your agent maps the question to governed events, runs the funnel, retention, cohort, or path analysis, validates the population, and returns the result with definitions and queries.

Usage question approved population and behavior named

Your AI agent

Resolves the question against event definitions, builds the analysis, checks tracking quality, and explains the largest differences with source links.

Usage analysis

Query and checks

Product brief

How to set it up

Step 1Connect your tools

Required

Product analytics

Reads events, properties, cohorts, funnels, paths, and experiment context.

Recommended

Warehouse

Reads governed user and account models for reconciliation.

Optional

Data catalog

Reads event definitions, owners, lineage, and known tracking limitations.

Docs

Writes the analysis brief and linked evidence.

Export the relevant event rows or dashboard table with definitions. Keep stable event and cohort names so another analyst can reproduce the result.

Step 2Build the workflow

Have the decision question, population, event dictionary, identity rules, observation window, expected baseline, and one known control result.

```text Setup prompt theme={null} Help me build a product-usage analysis workflow. For each approved behavior question, return a reproducible analysis using governed events and cohorts. 1. Ask for the decision, user or account population, behavior, events, sequence, observation window, comparison cohort, and expected baseline. 2. Map each concept to the event dictionary and state identity, grain, and exclusions. 3. Check tracking completeness, duplicates, late events, instrumentation changes, and whether warehouse and product-analytics counts reconcile. 4. Build the appropriate funnel, retention, path, or cohort analysis and show definitions, filters, and query or chart links. 5. Identify the segments driving the result and label hypotheses as hypotheses. 6. Return findings, controls, limitations, and the next decision or test. 7. Test on a funnel, retention, and feature-adoption question. ```

What good looks like

The population, events, identity, and time window should be explicit; results should reconcile with a control; and the PM should be able to distinguish a real behavior pattern from a tracking problem.

Step 3Turn it on

Choose your trigger

Run manually from a structured question or at an experiment checkpoint. Reject requests without a decision, population, and governed event mapping.

What runs without you

Interpretation stays reviewed - a funnel chart without context makes decisions faster and worse. Recurring reports can refresh automatically after three cycles that reconciled with the warehouse. Watch instrumentation changes weekly; most behavior changes this workflow finds will turn out to be tracking changes wearing a costume.

Pairs well with prepare roadmap decisions - behavioral evidence keeps the decision pack honest.

Prepare roadmap decisions

Best for comparing candidate work against the same evidence and criteria

1 hr/wkest. time saved How this estimate is calculated

How it's done today

You gather initiative status, customer evidence, usage data, dependencies, effort ranges, and strategic context into a deck or table before the actual tradeoff discussion can begin.

How AI helps

Your agent builds a consistent decision pack for each candidate, applies the agreed rubric without hiding missing evidence, and surfaces dependencies, conflicts, and options for the roadmap owner.

Planning cutoff reached candidate set is locked

Your AI agent

Normalizes candidate evidence, applies the same decision frame, and shows what changes under different constraints without selecting the roadmap.

Decision pack

Comparison table

Roadmap options

How to set it up

Step 1Connect your tools

Required

Tasks

Reads candidate initiatives, status, dependencies, owners, effort ranges, and target windows.

Docs

Reads strategy, decision rubric, research, requirements, and prior decisions.

Writes the decision pack and recorded outcome.

Optional

Product analytics

Reads baseline usage, reach, and outcome measures.

CRM

Reads source-linked customer and revenue evidence.

Export the candidate list and attach the strategy, rubric, and evidence links. The agent should mark missing evidence rather than assigning confident scores from thin summaries.

Step 2Build the workflow

Have a locked candidate set, decision owner, strategy, scoring or comparison rubric, capacity constraints, dependencies, evidence links, and the last planning decision.

```text Setup prompt theme={null} Help me build a roadmap-decision preparation workflow. Before each planning review, create a consistent evidence pack for the locked candidate set. Do not choose or reorder the roadmap. 1. Ask for the decision owner, planning horizon, strategy, candidate set, capacity constraints, rubric, and required evidence. 2. For each candidate, gather problem, target user, evidence, expected outcome, reach, effort range, dependencies, risks, confidence, and owner. 3. Cite every input and mark missing or stale evidence; do not fill gaps with generic scores. 4. Apply the same rubric and units across candidates and show how the result changes under alternative capacity or strategic constraints. 5. Surface conflicts, sequencing dependencies, irreversible decisions, and candidates that are not ready for comparison. 6. Produce the comparison, options, open decisions, and a record-ready summary. 7. Test against three past planning decisions and compare with the actual outcome. ```

What good looks like

Candidates should be comparable on the same evidence and units, missing inputs should remain visible, dependencies should affect the options, and the decision owner - not the agent - should make and record the final tradeoff.

Step 3Turn it on

Choose your trigger

Run at the planning evidence cutoff, after the candidate list is locked. Exclude unowned ideas and items without a problem statement; list them separately as not ready.

What runs without you

After three planning cycles where the packs held up, let preparation run automatically at each evidence cutoff - planning starts from evidence instead of assembly. The ranking, and every tradeoff behind it, remains the decision owner's. Revisit the rubric each period; a stale rubric quietly decides your roadmap for you.

Pairs well with analyze customer feedback and analyze product usage - they feed the customer and behavioral evidence the pack is built from.