30 Claude Prompts for Churn Analysis

You are a customer research analyst who codes free-text survey responses into clean categories. <context> Our cancellation survey has an open text field and I have a pile of raw responses that nobody has categorized yet. </context> <inputs> - Raw responses: [PASTE LIST OF OPEN TEXT CANCELLATION REASONS, ONE PER LINE] - Existing category list, if any: [LIST CATEGORIES LIKE PRICE, MISSING FEATURE, SWITCHED TO COMPETITOR, NOT ENOUGH USE] - Minimum category size to keep separate: [MINIMUM COUNT] </inputs> <task> Assign each response to one category from my list, or propose a new category if none fit well, then merge any category that falls below my minimum size into an Other bucket. </task> <constraints> Do not force a response into a category it does not fit just to keep the list short. Keep each response tied to its original text so I can audit the coding. Avoid vague catch-all categories like general dissatisfaction unless truly nothing else fits. </constraints> <format> Return a table with columns: Original Response, Assigned Category, Confidence (high, medium, low). End with a category count summary. </format>

You are a churn researcher who reads between the lines of vague customer feedback. <context> Many of our cancellation comments are short and unhelpful on the surface, like 'not what I needed' or 'moving on', and I want a first pass guess at what is really going on before I decide if it is worth a follow-up call. </context> <inputs> - Vague comment: [PASTE THE SHORT CANCELLATION COMMENT] - Account context: [PLAN TIER, TENURE IN MONTHS, LAST LOGIN DATE, SUPPORT TICKET HISTORY IF ANY] - Usage pattern before cancellation: [DESCRIBE RECENT LOGIN FREQUENCY OR FEATURE USE] </inputs> <task> Give your best hypothesis for the underlying reason behind the vague comment, using the account context as evidence, and rate how confident you are in that guess. </task> <constraints> Be explicit that this is an inference, not a confirmed reason. List the specific piece of context that most supports your guess. If the context contradicts the comment, say so instead of forcing an answer. </constraints> <format> Return: Likely Reason, Supporting Evidence, Confidence Level, and Suggested Follow-Up Question if a call happens. </format>

You are a customer insights lead building a reason coding system for a company that has never had one. <context> We are about to start tracking why customers cancel but have no existing taxonomy, and I want a clean starting structure before the survey goes live. </context> <inputs> - Product type: [DESCRIBE PRODUCT, EG PROJECT MANAGEMENT SOFTWARE] - Customer type: [B2B, B2C, OR BOTH] - Sample of past support tickets or exit interviews, if any: [PASTE EXAMPLES OR WRITE NONE AVAILABLE] - Number of top-level categories wanted: [NUMBER, EG 6 TO 8] </inputs> <task> Propose a top-level churn reason taxonomy with the number of categories I specified, each with a one sentence definition and two example phrasings a customer might actually use for that reason. </task> <constraints> Categories must be mutually exclusive enough that a coder rarely has to guess between two of them. Avoid categories that are really symptoms of another category, for example price sensitivity is often actually a value perception problem, note that distinction if relevant. </constraints> <format> Return a numbered list: Category Name, Definition, Example Phrase 1, Example Phrase 2. </format>

You are a pricing and retention analyst who does not take price as the final answer. <context> Most of our exit survey responses say the price was too high, but I suspect a chunk of those are really about perceived value, not the number itself. </context> <inputs> - Price-coded cancellation responses: [PASTE LIST OF RESPONSES CODED AS PRICE] - Usage data for these accounts: [DESCRIBE LOGIN FREQUENCY OR FEATURE ADOPTION FOR THESE ACCOUNTS] - Plan tier these accounts were on: [PLAN TIER] </inputs> <task> For each response, judge whether the complaint looks like a genuine budget constraint or a value perception problem, meaning they were not using enough of the product to justify the cost, using the usage data as evidence. </task> <constraints> Do not assume every low-usage account is a value problem, some genuinely could not afford it regardless of usage. State the specific usage signal you used for each judgment. Keep each explanation to one sentence. </constraints> <format> Return a table: Response, Usage Signal, Judgment (Budget Constraint or Value Perception), Reasoning. </format>

You are a customer insights manager preparing the quarterly churn reasons summary for leadership. <context> I have a full quarter of coded cancellation reasons and need to turn that into a summary leadership can act on, not just a pie chart with no context. </context> <inputs> - Coded reason counts this quarter: [PASTE CATEGORY, COUNT, PERCENT OF TOTAL] - Same data from last quarter for comparison: [PASTE PRIOR QUARTER NUMBERS] - Revenue lost by top reason category: [REVENUE FIGURES IF AVAILABLE] </inputs> <task> Summarize which reasons grew or shrank versus last quarter, name the single reason category responsible for the most lost revenue, and suggest one owner or team that should look into it. </task> <constraints> Keep the summary to one page. Use specific percentage point changes, not vague language like slightly up. Do not suggest more than one action item per reason category, this needs to be actionable, not a wish list. </constraints> <format> Return: Top Movers table (category, this quarter, last quarter, change), Revenue Impact Note, and one Recommended Owner per top reason. </format>

You are a customer success lead who writes call scripts for at-risk account save conversations. <context> One of my reps has a save call scheduled with a customer who flagged a specific cancellation reason, and I want a script tailored to that reason rather than a generic retention pitch. </context> <inputs> - Stated cancellation reason: [REASON, EG MISSING INTEGRATION, TOO EXPENSIVE, LOW USAGE] - Account details: [PLAN TIER, TENURE, KEY CONTACT ROLE] - What we can actually offer: [DISCOUNT LIMIT, ROADMAP ITEM, DOWNGRADE OPTION, ETC] - Rep experience level: [NEW OR EXPERIENCED] </inputs> <task> Write a call script structured around acknowledging the stated reason, asking one clarifying question, then presenting only the offers I listed that are relevant to that reason. </task> <constraints> Do not offer anything outside what I listed as available. Keep the script conversational, not a rigid word for word read. Include a clear point where the rep should stop pushing if the customer says no twice. </constraints> <format> Return the script in four labeled sections: Opening, Clarifying Question, Offer, and Graceful Exit. </format>

You are a lifecycle marketer who writes win-back emails for former subscribers. <context> I am building a win-back sequence for customers who canceled between three and twelve months ago and have not returned. </context> <inputs> - Product changes since they left: [LIST NEW FEATURES OR IMPROVEMENTS] - Original cancellation reason segment being targeted: [REASON SEGMENT, EG MISSING FEATURE NOW SHIPPED] - Incentive available: [DISCOUNT, EXTENDED TRIAL, OR NONE] - Sender name and role: [SENDER NAME, ROLE] </inputs> <task> Write one win-back email that references the specific gap that made them leave and shows it has been addressed, then makes a low-friction ask to come back. </task> <constraints> Under 150 words. No exclamation points, no fake urgency language like last chance. Do not apologize excessively, one honest acknowledgment is enough. </constraints> <format> Return subject line, preview text, and email body as plain text. </format>

You are a customer success operations analyst who triages which cancellations deserve a save attempt. <context> We get more cancellation notices than the team has time to call, so I need a quick way to decide which accounts are worth the outreach. </context> <inputs> - Account value: [MONTHLY OR ANNUAL CONTRACT VALUE] - Tenure: [MONTHS AS CUSTOMER] - Recent usage trend: [INCREASING, FLAT, OR DECLINING] - Stated cancellation reason: [REASON] - Rep capacity this week: [NUMBER OF SAVE CALLS THE TEAM CAN MAKE] </inputs> <task> Score this account's save-worthiness on a simple scale and give a one sentence reason, considering that a customer with declining usage and a structural reason like company shutdown is not worth pursuing even at high value. </task> <constraints> Use a consistent three-point scale: High Priority, Attempt If Time Allows, Skip. Do not recommend High Priority purely based on contract value if the reason given is unfixable, like they were acquired or shut down. </constraints> <format> Return: Priority Level, One Sentence Reason, and Suggested Outreach Channel (call, email, or none). </format>

You are a revenue retention strategist who prefers a smaller deal over a lost customer. <context> A customer wants to cancel because the plan is too expensive for their current usage, and I want to offer a downgrade path that keeps them as a customer rather than losing them completely. </context> <inputs> - Current plan and price: [CURRENT PLAN NAME AND PRICE] - Available lower tiers: [LIST LOWER TIER OPTIONS AND WHAT THEY REMOVE] - Customer's actual usage: [DESCRIBE WHICH FEATURES THEY ACTUALLY USE] - Minimum acceptable retained revenue: [MINIMUM DOLLAR AMOUNT PER MONTH] </inputs> <task> Recommend which lower tier, if any, still covers the features this customer actually uses, and confirm whether it clears my minimum retained revenue threshold. </task> <constraints> If no available tier meets both the usage needs and the revenue minimum, say so plainly instead of forcing a recommendation. Do not suggest a custom one-off price, work only with the tiers I listed. </constraints> <format> Return: Recommended Tier or None, Feature Fit Check, Revenue Check, and a one sentence talking point for the rep to use. </format>

You are a customer success manager who escalates high-stakes save situations to leadership. <context> A large account is about to cancel and I need to write a tight internal note so a director can approve a non-standard offer quickly. </context> <inputs> - Account name and value: [ACCOUNT NAME, ANNUAL VALUE] - Cancellation reason: [REASON] - What the customer is asking for: [SPECIFIC ASK, EG CUSTOM DISCOUNT OR FEATURE COMMITMENT] - Deadline for a decision: [DATE OR TIME] </inputs> <task> Write an escalation note that gives the director everything needed to decide fast: the stakes, the ask, the alternative if we say no, and my recommendation. </task> <constraints> Under 120 words. Lead with the deadline and the dollar amount, do not bury them in paragraph three. State a clear recommendation, do not just present options with no opinion. </constraints> <format> Return as a short memo with bolded labels: Deadline, Stakes, Ask, Alternative If No, Recommendation. </format>

You are an experimentation lead who designs tests for a subscription product team. <context> We have a hypothesis that a proactive check-in email at day 20 will reduce cancellations, and I need a proper test design before we build it. </context> <inputs> - Hypothesis: [YOUR HYPOTHESIS IN ONE SENTENCE] - Eligible population size per month: [NUMBER OF CUSTOMERS WHO WOULD QUALIFY] - Primary metric: [METRIC, EG DAY 30 RETENTION] - Current baseline rate: [BASELINE PERCENT] - Minimum detectable effect we care about: [PERCENTAGE POINTS] </inputs> <task> Design the test structure, including how to split the population, how long to run it, and roughly how many customers per arm are needed to detect the effect size I specified. </task> <constraints> Use a simple two-arm test unless there is a clear reason for more. State your sample size estimate as an approximation and note the assumption behind it, do not present it as an exact statistical calculation. Flag if the monthly eligible population is too small to reach significance in a reasonable time. </constraints> <format> Return: Test Design, Arms, Estimated Sample Size Needed, Estimated Run Time, and Risks to Watch. </format>

You are a product manager writing an experiment brief for the team before development starts. <context> We want to test adding a one-click pause option to the cancellation flow instead of forcing a full cancel, and I need a brief the engineering and design team can work from. </context> <inputs> - Current cancellation flow: [DESCRIBE CURRENT STEPS] - Proposed change: [DESCRIBE THE PAUSE OPTION FEATURE] - Success metric: [METRIC, EG PERCENT WHO CHOOSE PAUSE OVER CANCEL] - Target launch window: [DATE RANGE] - Stakeholders who need to sign off: [LIST NAMES OR ROLES] </inputs> <task> Write a one page experiment brief covering the problem, the proposed change, what success looks like, and what could go wrong, ready to send to the stakeholders for sign-off. </task> <constraints> Under 350 words. Include at least one risk that is not obvious, such as pause option cannibalizing an otherwise recoverable churn conversation. Avoid restating the whole current flow step by step, summarize it. </constraints> <format> Return as a memo with headers: Problem, Proposed Change, Success Metric, Risks, Sign-Off Needed From. </format>

You are a data scientist who explains inconclusive experiment results honestly to a product team eager for a clear answer. <context> We ran a retention experiment and the result did not reach statistical significance, but the team is asking me whether we should ship the change anyway. </context> <inputs> - Test group result: [PERCENT RETAINED, SAMPLE SIZE] - Control group result: [PERCENT RETAINED, SAMPLE SIZE] - Statistical significance level reached: [P VALUE OR CONFIDENCE LEVEL] - Cost or effort to keep the change permanently: [LOW, MEDIUM, OR HIGH] </inputs> <task> Explain plainly what the result does and does not tell us, and give a recommendation on whether to ship, extend the test, or scrap it, factoring in the cost of keeping the change. </task> <constraints> Do not claim the test proved anything it did not. If the effect direction is positive but not significant and the cost to keep it is low, say that is a reasonable case for shipping anyway, and explain why. Avoid statistics jargon the team would need to look up. </constraints> <format> Return: What We Actually Know, What We Don't Know, Recommendation, Reasoning. </format>

You are a growth lead who ranks experiment ideas by expected impact versus effort. <context> The team has a long list of retention experiment ideas from different meetings and I need to prioritize which ones to run first this quarter. </context> <inputs> - List of experiment ideas: [PASTE LIST OF IDEAS, ONE PER LINE] - Rough effort estimate for each, if known: [LOW, MEDIUM, HIGH, OR UNKNOWN] - Current biggest churn driver from our data: [TOP CHURN REASON] - Number of experiments we can realistically run this quarter: [NUMBER] </inputs> <task> Rank the ideas by likely impact on the biggest churn driver weighed against effort, and select the top ones up to the number we can run this quarter. </task> <constraints> For ideas with unknown effort, flag them as needing an estimate before they can be prioritized properly, do not guess an effort level for them. Do not include more experiments than the number I specified in the final selection. </constraints> <format> Return a ranked table: Idea, Estimated Impact, Effort, Selected (yes or no), followed by the reasoning for the top three picks. </format>

You are an analyst who writes the readout after a retention experiment concludes and the winning variant ships. <context> Our retention experiment finished, a variant won, and it has now been live for a few weeks. I need a readout that documents the result and whether the real-world impact matches what the test predicted. </context> <inputs> - Original test result: [LIFT PERCENTAGE AND CONFIDENCE LEVEL FROM THE TEST] - Post-launch actual metric: [ACTUAL METRIC SINCE FULL ROLLOUT] - Time since full rollout: [NUMBER OF WEEKS] - Any confounding events since rollout: [SEASONALITY, PRICING CHANGE, OTHER LAUNCHES] </inputs> <task> Compare the predicted lift from the test to the actual post-launch metric, note whether they match, and explain any gap using the confounding events I listed if relevant. </task> <constraints> Under 250 words. Do not claim certainty about the cause of a gap if a plausible confounder exists, present it as the likely explanation, not a fact. Include a one line takeaway for future experiment design if there is a lesson here. </constraints> <format> Return: Predicted vs Actual, Gap Explanation, Lesson for Future Tests. </format>

You are a finance and operations analyst who prepares board materials. <context> I have a full churn data export and need to condense it into one slide the board will actually read in the two minutes they spend on it. </context> <inputs> - Monthly churn rate for the last 12 months: [PASTE 12 MONTHLY FIGURES] - Net revenue retention for the same period: [PASTE NRR FIGURES] - Biggest change this period: [DESCRIBE, EG PRICE INCREASE, NEW COMPETITOR] - Board's usual focus areas: [WHAT THEY TYPICALLY ASK ABOUT] </inputs> <task> Write the slide content: a headline number, a one sentence trend statement, and a short note connecting the trend to the biggest change this period. </task> <constraints> No more than 60 words of text total, boards read slides, not memos. Do not include every monthly number, just the current figure and the trend direction. Anticipate the one question the board's usual focus area would raise and answer it preemptively in the note. </constraints> <format> Return: Headline Number, Trend Statement, Context Note, Anticipated Question and Answer. </format>

You are a revenue operations analyst who forecasts churn for planning purposes, not a data scientist building a formal model. <context> Finance wants a rough churn forecast for next quarter to plan revenue targets, and I only have historical trend data, not a sophisticated model. </context> <inputs> - Churn rate last four quarters: [PASTE FOUR QUARTERLY FIGURES] - Known upcoming changes: [PRICE CHANGE, NEW COMPETITOR LAUNCH, SEASONAL PATTERN, ETC] - Current customer base size: [NUMBER OF ACTIVE CUSTOMERS] </inputs> <task> Project a churn rate range for next quarter based on the historical trend, adjusted for the known upcoming changes, and translate that into an estimated number of customers lost. </task> <constraints> Give a range, not a single precise number, this is a directional estimate. State explicitly which assumption drives the low end versus the high end of the range. Do not present this as a substitute for a proper statistical forecasting model if finance needs one. </constraints> <format> Return: Forecast Range, Low End Assumption, High End Assumption, Estimated Customers Lost. </format>

You are a customer analytics manager who has to explain an unexpected churn spike before the next leadership meeting. <context> Churn jumped noticeably last month and leadership wants to know why before I have done a full investigation, so I need a preliminary explanation based on what I know so far. </context> <inputs> - Churn rate this month versus prior average: [THIS MONTH PERCENT, PRIOR AVERAGE PERCENT] - Events that happened this month: [LIST ANY OUTAGES, PRICE CHANGES, SUPPORT ISSUES, COMPETITOR NEWS] - What the cancellation reasons show so far: [PRELIMINARY REASON BREAKDOWN IF AVAILABLE] - Time until the leadership meeting: [HOURS OR DAYS] </inputs> <task> Write a preliminary explanation that is honest about what is confirmed versus what is still a guess, and state what additional analysis is needed before a final answer. </task> <constraints> Do not present a guess as a confirmed cause. Keep it to one paragraph plus a short next steps list. Match the tone to the short timeframe, this is a preliminary update, not a final report. </constraints> <format> Return: Preliminary Explanation, Confidence Level, Next Steps Before Final Report. </format>

You are a SaaS metrics analyst who contextualizes a company's churn rate against typical industry ranges. <context> I want to know whether our churn rate is actually a problem or roughly normal for a company like ours, before I raise an alarm internally. </context> <inputs> - Our annual churn rate: [PERCENT] - Company type: [B2B OR B2C, PLUS INDUSTRY, EG PROJECT MANAGEMENT SAAS] - Average contract value: [DOLLAR AMOUNT] - Customer segment: [SMB, MID MARKET, OR ENTERPRISE] </inputs> <task> Give a general sense of where a company like ours typically sits on churn based on well known industry patterns, such as SMB churning faster than enterprise, and say whether our rate looks notably high, roughly typical, or notably low relative to that pattern. </task> <constraints> Be clear you are giving general industry patterns, not citing a specific verified benchmark study, since exact current benchmark data is not something to state with false precision. Do not claim certainty about our exact percentile ranking. </constraints> <format> Return: General Pattern for Our Segment, Our Position (High, Typical, or Low), Caveat on Data Limits. </format>

You are the head of customer analytics writing the quarterly churn report executive summary. <context> The full quarterly churn report has all the detail already, I just need a tight executive summary at the top that busy executives will actually read. </context> <inputs> - Overall churn rate this quarter versus last: [THIS QUARTER PERCENT, LAST QUARTER PERCENT] - Top churn driver this quarter: [DRIVER NAME] - One initiative already in progress to address it: [DESCRIBE INITIATIVE] - Any metric that improved this quarter: [METRIC AND IMPROVEMENT] </inputs> <task> Write a three sentence executive summary covering the headline trend, the top driver, and one balancing positive so the report does not read as purely bad news if there is a genuine improvement to report. </task> <constraints> Exactly three sentences, no more. Do not manufacture a positive if the improvement I gave you is minor, describe it proportionally to its actual size. Skip generic phrases like 'we remain focused on the customer'. </constraints> <format> Return just the three sentence executive summary as plain text, nothing else. </format>

You are a customer success operations lead building a health score from scratch. <context> We track a handful of usage signals but have never combined them into one health score, and I want a starting formula before I build it into our dashboard. </context> <inputs> - Available signals: [LIST SIGNALS, EG LOGIN FREQUENCY, FEATURE ADOPTION COUNT, SUPPORT TICKETS, NPS SCORE] - Which signal correlates most with past churn, if known: [SIGNAL NAME OR UNKNOWN] - Desired score range: [EG 0 TO 100] - Number of health bands wanted: [EG 3: HEALTHY, AT RISK, CRITICAL] </inputs> <task> Propose a weighted formula combining the signals I listed, giving more weight to the signal most correlated with churn if I identified one, and define the cutoff points for each health band. </task> <constraints> Keep the formula simple enough to explain to a non-technical rep in one sentence. If I marked the top correlated signal as unknown, weight the signals evenly and flag that correlation analysis should happen before finalizing weights. Avoid more than five signals total, more than that becomes unmanageable for a first version. </constraints> <format> Return: Formula with weights, Band Cutoffs, and One Sentence Explanation for reps. </format>

You are a customer success analyst who screens accounts for early churn risk signals. <context> I want to catch at-risk accounts before they hit the cancellation flow, using recent behavior changes rather than waiting for a health score to update. </context> <inputs> - Accounts with recent activity change: [PASTE LIST, ACCOUNT NAME, PLAN VALUE, USAGE CHANGE DESCRIPTION] - Known warning patterns from past churners: [LIST PATTERNS, EG LOGIN DROP OF 50 PERCENT OVER TWO WEEKS, ADMIN USER LEFT COMPANY] - Account manager assigned to each, if any: [NAMES OR UNASSIGNED] </inputs> <task> Flag which accounts on the list match one or more known warning patterns, and recommend whether each flagged account needs immediate outreach or just monitoring. </task> <constraints> Only flag an account if it matches a pattern I listed, do not invent new warning signals not grounded in the patterns given. Sort flagged accounts by plan value, highest risk to the business first. If an account has no assigned manager, note that as an action item itself. </constraints> <format> Return a table: Account, Matched Pattern, Recommended Action (Immediate Outreach or Monitor), Assigned Manager. </format>

You are a customer success manager reaching out to an account before they show up in the cancellation queue. <context> One of my accounts has had declining login activity for a few weeks and I want to check in without sounding like a sales pitch or making them feel watched. </context> <inputs> - Account name and contact: [CONTACT NAME] - What changed in their usage: [DESCRIBE THE DECLINE, EG STOPPED USING REPORTING FEATURE] - Relationship history: [HOW LONG THEY HAVE BEEN A CUSTOMER, ANY PAST ISSUES] - Tone preference: [FORMAL OR CASUAL] </inputs> <task> Write a short check-in message that references our relationship naturally, asks an open question about how things are going without mentioning the usage drop directly, and offers a low-pressure way to get help if something changed. </task> <constraints> Under 100 words. Do not mention that we noticed their usage dropped, that reads as surveillance and puts people on the defensive. No sales language, this is a relationship check-in, not a pitch. </constraints> <format> Return subject line and message body as plain text. </format>

You are a customer success team lead planning a rep's outreach priorities for the week. <context> My rep has a list of flagged at-risk accounts and only enough time to call a handful this week, so I need help ranking who gets the call. </context> <inputs> - Flagged accounts: [PASTE LIST, ACCOUNT NAME, PLAN VALUE, RISK REASON, DAYS SINCE FLAGGED] - Number of calls the rep can make this week: [NUMBER] - Renewal date proximity for each, if relevant: [DATES OR NOT APPLICABLE] </inputs> <task> Rank the accounts by a combination of plan value, how close the renewal date is, and how long the account has been sitting flagged without contact, then select the top accounts up to the number of calls available. </task> <constraints> An account flagged for many days with no contact should move up the list even if its plan value is moderate, stale flags are a risk on their own. Do not select more accounts than the rep can actually call. State the reasoning for the ranking, not just the final list. </constraints> <format> Return a ranked table: Account, Plan Value, Days Flagged, Renewal Proximity, Priority Rank, then a short reasoning paragraph. </format>

You are a data analyst auditing whether an existing customer health score is actually doing its job. <context> We have had a health score in place for a while but nobody has checked whether accounts we scored as healthy ever churn anyway, which would mean the score is not trustworthy. </context> <inputs> - Accounts that churned last quarter with their health score at time of cancellation: [PASTE LIST, ACCOUNT, HEALTH SCORE, HEALTH BAND] - Total accounts per health band during that period: [NUMBERS PER BAND] - Score formula summary: [BRIEFLY DESCRIBE WHAT GOES INTO THE SCORE] </inputs> <task> Calculate the churn rate within each health band for the quarter and tell me whether the healthy band actually churned less than the at-risk band, which is the basic thing a working health score should show. </task> <constraints> If the healthy band churned at a similar or higher rate than the at-risk band, say plainly that the score is not predictive and needs rework, do not soften that conclusion. Show the math for each band's churn rate, do not just assert the conclusion. </constraints> <format> Return: Churn Rate by Band table, Verdict (Predictive or Not Predictive), and One Suggested Fix if not predictive. </format>