Choose one decision.
Supply the bracketed inputs, run the prompt, review the evidence, and choose a real next action. Every copy button includes the session instructions. These prompts do not publish, contact customers, or execute campaigns.
Independently authored. HubSpot’s public staged prompt-resource concept informed the request; its gated prompts were not accessed or copied.
40 of 40 prompts
MC-U01 · Understand
Customer-language evidence map
Bring: [anonymized interviews], [offer], [research question]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [anonymized interviews], [offer], [research question]. Task: Group repeated customer situations, desired outcomes, and objections. Preserve contradictions and tie each observation to its supplied source. Count only observed mentions; do not imply population prevalence. Return: A table of theme, supporting evidence, conflicting evidence, question to investigate, and messaging implication. Before finalizing: Check that each attributed statement exists in the input and that interpretation is labeled.
MC-U02 · Understand
Interview guide for a buying decision
Bring: [target situation], [decision], [what we already know]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [target situation], [decision], [what we already know]. Task: Design a short interview about the last time the problem occurred, the workaround, alternatives, purchase process, and consequences. Avoid leading questions and hypothetical enthusiasm as the main evidence. Return: Eight neutral questions, follow-up probes, recruiting criteria, and a note-taking structure. Before finalizing: Remove questions that reveal the desired answer or presume the problem is important.
MC-U03 · Understand
Alternative-solution map
Bring: [customer problem], [known alternatives], [source excerpts with dates]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [customer problem], [known alternatives], [source excerpts with dates]. Task: Compare doing nothing, manual work, direct competitors, and adjacent approaches. Distinguish what each source claims from independently observed behavior. Do not infer missing functionality from absent marketing copy. Return: A sourced alternatives table with tradeoffs, unknowns, and questions for buyers. Before finalizing: Check source dates, comparable scope, and whether a gap is evidence or speculation.
MC-U04 · Understand
Differentiation claim check
Bring: [our offer], [approved proof], [competitor material]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [our offer], [approved proof], [competitor material]. Task: Identify claims that are specific, relevant, and supportable. Separate real differences from category requirements and unsupported superlatives. Suggest a way to validate a potentially useful difference. Return: Three positioning options with evidence, limitations, and a buyer question for each. Before finalizing: Delete any claim of being best, unique, or faster unless evidence supports the comparison.
MC-U05 · Understand
Journey friction investigation
Bring: [journey steps], [observed behavior], [support questions]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [journey steps], [observed behavior], [support questions]. Task: Map the customer goal, action, uncertainty, and handoff at each step. Identify where evidence points to friction and where instrumentation is missing. Separate acquisition from first-use and ongoing value. Return: A journey table and three prioritized investigation tasks. Before finalizing: Ensure each proposed fix addresses an observed problem or an explicitly labeled hypothesis.
MC-U06 · Understand
Audience evidence brief
Bring: [research notes], [CRM field summary], [business objective]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [research notes], [CRM field summary], [business objective]. Task: Describe candidate audience situations without inventing demographics. Separate declared needs, observed actions, and inferred intent. Identify what could make each audience a poor fit. Return: Two audience briefs with evidence strength, exclusions, missing data, and next research step. Before finalizing: Check that a behavioral signal has not silently become a psychological certainty.
MC-U07 · Understand
Brand voice operating guide
Bring: [approved writing samples], [positioning], [words to avoid]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [approved writing samples], [positioning], [words to avoid]. Task: Identify repeatable voice choices and turn them into usable instructions. Keep factual claims separate from tone. Write examples showing the same message in an acceptable and unacceptable form. Return: A one-page guide covering tone, vocabulary, structure, claim boundaries, and three examples. Before finalizing: Have the brand owner approve the choices; do not treat stylistic analysis as authority.
MC-U08 · Understand
Audience preference brief
Bring: [explicit preferences], [content feedback], [sample constraints]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [explicit preferences], [content feedback], [sample constraints]. Task: Summarize evidence about preferred depth, format, examples, and objections. Mark uncertain patterns and possible sampling bias. Suggest a small test rather than claiming predictive accuracy. Return: A preference brief with confidence labels and two testable creative hypotheses. Before finalizing: Confirm preferences are sourced and that unknown readers still receive a useful default.
MC-U09 · Understand
Message comprehension research
Bring: [page or ad copy], [intended audience], [research access]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [page or ad copy], [intended audience], [research access]. Task: Design a test that asks readers what they think the offer does, whom it serves, what happens next, and what remains unclear. Separate comprehension from purchase intent. Return: A short test script, observation sheet, and revision criteria. Before finalizing: Do not fabricate participant responses or score copy as tested before sessions occur.
MC-U10 · Understand
Research-to-campaign brief
Bring: [reviewed evidence], [offer], [constraints], [goal]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [reviewed evidence], [offer], [constraints], [goal]. Task: Turn the evidence into one focused campaign hypothesis. Identify the audience, promise, proof, destination, and action. List assumptions that could invalidate the plan. Return: A campaign brief with evidence links, owner decisions, and a smallest-useful test. Before finalizing: Verify the selected objective is a customer or business outcome rather than asset volume.
MC-S01 · Shape
Actionable segment specification
Bring: [field dictionary], [goal], [anonymized examples]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [field dictionary], [goal], [anonymized examples]. Task: Propose at most three groups that should receive different help. Define membership, exclusions, time windows, unknown-value behavior, update cadence, and exit conditions using available fields only. Return: A segment specification per group with purpose, treatment, success measure, and unsupported requirements. Before finalizing: Reject any segment that cannot be explained, populated, or connected to a different action.
MC-S02 · Shape
Segment logic and edge-case review
Bring: [proposed rule], [field meanings], [sample records]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [proposed rule], [field meanings], [sample records]. Task: Translate the rule into parenthesized AND/OR logic. Keep permission and suppression outside all qualifying branches. Test qualifying, nonqualifying, missing, overlapping, and stale records. Return: Readable logic, an edge-case table, and unresolved semantics. Before finalizing: Ensure current customer or opt-out changes cannot leave someone eligible for an inappropriate send.
MC-S03 · Shape
Audience overlap and message priority
Bring: [segments], [campaigns], [frequency policy]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [segments], [campaigns], [frequency policy]. Task: Identify where people could qualify for multiple treatments. Propose priority and deduplication rules that protect customer onboarding and current preferences. Distinguish audience membership from send eligibility. Return: An overlap matrix, precedence rules, and final eligibility checks. Before finalizing: Check that one person cannot receive duplicate versions through retries or separate journeys.
MC-S04 · Shape
Segment-specific message workshop
Bring: [approved offer], [segment needs], [brand guide]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [approved offer], [segment needs], [brand guide]. Task: Write different explanations of the same truthful offer for each audience. Change examples, objections, and next steps where relevant; preserve price, scope, evidence, and policy facts. Return: A message matrix with headline, explanation, proof, CTA, and rationale. Before finalizing: Check factual consistency and remove language that reveals intrusive behavioral monitoring.
MC-S05 · Shape
Landing-page treatment brief
Bring: [base page], [known context signals], [approved variations]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [base page], [known context signals], [approved variations]. Task: Specify which page elements should vary and why. Keep essential terms consistent, avoid unsupported claims, and provide a complete default when context is absent or tracking is unavailable. Return: A treatment table with condition, content change, fallback, and measurement. Before finalizing: Verify the default path is useful and that a visitor can understand the offer without personalization.
MC-S06 · Shape
Welcome journey with exit rules
Bring: [resource promise], [onboarding milestone], [email permissions]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [resource promise], [onboarding milestone], [email permissions]. Task: Plan a helpful welcome sequence that delivers the promise and supports the first useful action. Define timing, entry, exit, suppression, and ownership. Separate resource delivery from optional promotion. Return: A journey table with trigger, message purpose, CTA, stop rule, and observable success. Before finalizing: Check that completing the milestone stops redundant reminders and that delivery failure is visible.
MC-S07 · Shape
Customer activation assistance
Bring: [first-value definition], [product events], [support patterns]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [first-value definition], [product events], [support patterns]. Task: Identify an observable point where new customers stall. Propose a helpful intervention with a fallback for unknown event data. Do not infer dissatisfaction from inactivity alone. Return: An activation hypothesis, audience rule, intervention draft, and evaluation plan. Before finalizing: Confirm the first-value event represents benefit, not merely a login or page view.
MC-S08 · Shape
Renewal and expansion relevance
Bring: [renewal dates], [documented outcomes], [service scope]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [renewal dates], [documented outcomes], [service scope]. Task: Prepare a relevant customer conversation based on realized value and unmet needs. Separate renewal risk from an expansion opportunity. Exclude unsupported financial projections and pressure tactics. Return: A value-review outline, evidence gaps, customer questions, and next-step options. Before finalizing: Check that projected value is not reported as realized revenue or an agreed commitment.
MC-S09 · Shape
Data usefulness and freshness audit
Bring: [field inventory], [source systems], [intended segments]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [field inventory], [source systems], [intended segments]. Task: Assess completeness, freshness, ownership, allowed use, and business relevance for the fields needed. Identify redundant or unnecessary data and the consequence of unknown values. Return: A field audit with critical fixes and a minimum viable dataset. Before finalizing: Do not recommend collecting sensitive information merely to improve a model score.
MC-S10 · Shape
Personalization quality review
Bring: [treatments], [brand guide], [facts], [default experience]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [treatments], [brand guide], [facts], [default experience]. Task: Review relevance, accuracy, accessibility, consistency, and customer control. Identify stereotypes, overconfident inferences, and missing fallback behavior. Suggest precise edits. Return: An issue list ordered by customer impact, revised passages, and release checks. Before finalizing: Confirm every treatment is independently understandable and truthful.
MC-R01 · Reach
Channel decision brief
Bring: [audience access], [offer], [budget and time], [prior results]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [audience access], [offer], [budget and time], [prior results]. Task: Compare feasible distribution channels based on the buying situation and evidence. Select one initial channel and one complementary route. Make assumptions about reach and costs explicit. Return: A channel comparison, selection rationale, first test, and stop conditions. Before finalizing: Do not invent platform benchmarks or spread a small budget across too many tests.
MC-R02 · Reach
Search intent content plan
Bring: [real customer questions], [site inventory], [search evidence]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [real customer questions], [site inventory], [search evidence]. Task: Group questions by intent and propose pages that answer them distinctly. Identify existing pages to improve before creating duplicates. Connect each page to a useful next step. Return: A topic map with intent, unique value, existing-page relationship, and internal links. Before finalizing: Treat keyword volume as unknown unless a verified source was supplied.
MC-R03 · Reach
AI-answer visibility investigation
Bring: [approved queries], [dated answer observations], [site pages]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [approved queries], [dated answer observations], [site pages]. Task: Separate mentions, citations, and referrals. Compare observed answers with what the site actually explains. Propose improvements to clarity, useful coverage, and source support without promising visibility. Return: An observation log, content gaps, technical questions, and a repeatable review method. Before finalizing: Preserve engine, query, date, and limitations; never invent an answer-engine scan.
MC-R04 · Reach
Partner collaboration fit
Bring: [candidate partners], [verified audience information], [shared topic]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [candidate partners], [verified audience information], [shared topic]. Task: Assess relevance and complementary expertise. Design a useful collaboration that benefits both audiences. List unknown access, cost, rights, and logistics. Draft outreach only for human review. Return: A shortlist with evidence, collaboration concept, and draft invitation. Before finalizing: Do not invent follower counts, response likelihood, endorsement, or permission to send.
MC-R05 · Reach
One source, useful formats
Bring: [approved article], [channel], [audience], [brand guide]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [approved article], [channel], [audience], [brand guide]. Task: Adapt the core insight into an explanation, demonstration, comparison, and email. Give each piece standalone value and a relevant next step. Preserve factual boundaries. Return: Four channel-suitable drafts with a shared claim check. Before finalizing: Ensure the versions are useful adaptations, not identical snippets with different line breaks.
MC-R06 · Reach
Community contribution plan
Bring: [community rules], [observed questions], [expertise]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [community rules], [observed questions], [expertise]. Task: Propose contributions that answer real questions without requiring a promotional click. Identify when a link would be permitted and when asking an organizer first is appropriate. Return: Three contribution outlines, context requirements, and an optional resource invitation. Before finalizing: Do not fabricate membership, community permission, or conversations.
MC-R07 · Reach
Workshop as a discovery channel
Bring: [audience problem], [teaching skill], [capacity]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [audience problem], [teaching skill], [capacity]. Task: Outline a session that creates a useful participant result. Connect the exercise to an optional next step while keeping the lesson complete. Plan accessible materials and follow-up choices. Return: A session agenda, exercise, invitation copy, and outcome measures. Before finalizing: Check that attendance, satisfaction, and qualified interest are measured separately.
MC-R08 · Reach
Paid-search intent review
Bring: [offer], [reported search terms], [outcomes], [locations]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [offer], [reported search terms], [outcomes], [locations]. Task: Classify queries as relevant, investigate, or exclude. Explain context, suggest narrowly scoped negatives where justified, and flag conflicts with other campaigns. Return: A query decision table with reasons and data limitations. Before finalizing: Never treat free, reviews, or jobs as universally bad; do not confuse raw leads with qualified leads.
MC-R09 · Reach
Campaign activation readiness
Bring: [approved assets], [channels], [integrations], [owners]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [approved assets], [channels], [integrations], [owners]. Task: Check destination accuracy, consent eligibility, scheduling, segment freshness, deduplication, rollback, and error handling. Distinguish a drafted asset from a working connection. Return: A release checklist with pass, fail, unknown, owner, and evidence needed. Before finalizing: Do not mark a connection tested without an observed result; do not send or publish.
MC-R10 · Reach
Distribution calendar with capacity
Bring: [approved assets], [team availability], [events], [review times]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [approved assets], [team availability], [events], [review times]. Task: Create a manageable calendar around dependencies and audience needs. Include review, reuse, and support time. Keep time zones explicit and dates proposed until scheduled in a real system. Return: A calendar with owner, asset, channel, objective, dependency, and status. Before finalizing: Check that the calendar does not promise more execution than the team can support.
MC-L01 · Learn
Experiment definition before launch
Bring: [hypothesis], [eligible population], [baseline], [constraints]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [hypothesis], [eligible population], [baseline], [constraints]. Task: Define the primary outcome, guardrails, comparison, observation window, and decision criteria. Identify confounders and whether the available sample can support a causal conclusion. Return: A prerelease experiment card with measurement and interpretation limits. Before finalizing: Avoid inventing sample sufficiency or statistical confidence.
MC-L02 · Learn
Campaign bottleneck diagnosis
Bring: [actual funnel data], [definitions], [time window], [changes]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [actual funnel data], [definitions], [time window], [changes]. Task: Locate the observed drop-off and suggest competing explanations. Distinguish tracking failure, poor fit, weak offer, and conversion delay. Prioritize checks before remedies. Return: A diagnostic table with evidence, alternative explanation, next check, and owner. Before finalizing: Do not claim causation from a correlation or compare incompatible periods.
MC-L03 · Learn
Segment outcome review
Bring: [membership snapshot], [treatments], [outcomes], [exclusions]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [membership snapshot], [treatments], [outcomes], [exclusions]. Task: Compare outcomes using consistent denominators and windows. Account for overlaps and changing membership. Distinguish a valuable audience from a treatment that improved its results. Return: A segment report with rates, counts, limitations, and one next test. Before finalizing: Do not calculate historical rates from today’s dynamic membership.
MC-L04 · Learn
Attribution reconciliation
Bring: [CRM outcomes], [analytics], [ad reports], [attribution windows]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [CRM outcomes], [analytics], [ad reports], [attribution windows]. Task: Explain discrepancies without adding platform-reported conversions together. Separate attribution rules, unknown source, delayed conversion, and duplicate counting from incremental impact. Return: A reconciliation plan, discrepancy table, and qualified conclusions. Before finalizing: Do not allocate missing revenue to a channel simply to make totals match.
MC-L05 · Learn
Retention and revenue movement
Bring: [opening cohort], [MRR movements], [definitions], [period]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [opening cohort], [MRR movements], [definitions], [period]. Task: Check opening recurring revenue, expansion, contraction, churn, and any reactivation convention. Calculate GRR and NRR only with compatible cohort data; keep new customers outside the opening-cohort ratios. Return: A calculation trail, missing-input list, and explanation of movements. Before finalizing: Use unknown rather than zero for missing values; state the treatment of reactivation.
MC-L06 · Learn
Offer economics review
Bring: [net sales], [variable costs], [spend], [refund assumptions]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [net sales], [variable costs], [spend], [refund assumptions]. Task: Calculate contribution before and after acquisition. Separate observed costs from allowances and fixed overhead. Show how changes in price or returns affect affordable acquisition. Return: A transparent scenario table with break-even assumptions and decision limits. Before finalizing: Do not label contribution as final profit or forecasts as earned revenue.
MC-L07 · Learn
Customer feedback after a campaign
Bring: [anonymized replies], [support issues], [campaign promise]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [anonymized replies], [support issues], [campaign promise]. Task: Group confusion, unmet expectations, useful outcomes, and requests. Preserve minority concerns and distinguish severity from frequency. Tie changes to actual statements. Return: A feedback summary with example evidence, priority, and proposed response. Before finalizing: Do not fabricate quotes or assume silent customers were satisfied.
MC-L08 · Learn
Measurement quality check
Bring: [event dictionary], [instrumentation results], [source systems]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [event dictionary], [instrumentation results], [source systems]. Task: Review event meaning, duplicate firing, missing records, time zones, identity boundaries, and reconciliation. Identify which business conclusions are unsupported until tracking is fixed. Return: A measurement audit and prioritized repair checklist. Before finalizing: Distinguish clicks, submissions, accepted leads, bookings, and paid outcomes.
MC-L09 · Learn
Content refresh decision
Bring: [content inventory], [dated performance], [customer questions]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [content inventory], [dated performance], [customer questions]. Task: Decide which assets to retain, improve, consolidate, or retire. Consider usefulness, changed facts, overlap, and business relevance rather than traffic alone. Return: A prioritized refresh plan with reason, dependency, and expected evidence. Before finalizing: Do not change publication dates merely to make content appear recent.
MC-L10 · Learn
Learning memo and next move
Bring: [experiment plan], [actual results], [limitations], [constraints]
Use only supplied evidence or sources you actually retrieve. Separate observations, interpretations, and hypotheses. If a required input is missing, ask for it or state the limitation. Do not invent customer quotes, data, results, source access, or completed actions. Use minimized, appropriate data; do not infer sensitive traits. Produce a reviewable draft, not an external action. Context and inputs: [experiment plan], [actual results], [limitations], [constraints]. Task: Write a brief decision memo separating what happened, what it might mean, what remains unknown, and what to do next. Identify evidence that would reverse the decision. Return: A one-page memo with continue, revise, pause, or investigate recommendation. Before finalizing: Keep unsupported certainty and retrospective changes to success criteria out of the memo.