10 ChatGPT prompts that automate weekly board reports in 2026

Weekly board reports eat a full Sunday because the prep is a synthesis pipeline disguised as a slide deck. Ten ChatGPT prompts — MRR delta narrative, churn root-cause classifier, pipeline coverage commentary, hiring vs runway frame, cohort retention narrative, R&D project condenser, board-question pre-mortem, KPI variance explainer, investor update one-pager, and an exec-team-update meta-prompt — turn 6 hours of work into a 60-minute Sunday ritual. Each prompt below carries full text, design reasoning, a sample output, and the input artifact it expects.

<a href="https://aipromptshub.co/chatgpt-prompt-generator?utm_source=aipromptshub&utm_medium=blog&utm_campaign=board-reports-2026" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold"}}> Generate a custom board-report prompt → </a>

Board reporting cadence shifted hard between 2023 and 2026. The NACD 2025 Public Company Governance Survey found that 64% of venture-backed boards now expect a weekly or bi-weekly written update, up from 31% in 2022. Diligent's 2025 Modern Governance Summit data attributes the shift to two forces: distributed boards that lost the hallway-conversation channel, and AI-literate directors who expect the underlying data, not just the summary.

Founders absorb the cost. Lenny Rachitsky's 2025 startup-ops survey put weekly board prep at a median of 5.8 hours for Series A founders and 8.2 hours for Series B — the single largest recurring time cost in the operator calendar. The work breaks into three buckets: pulling data (ChartMogul, Stripe, HubSpot, ATS), writing narrative around the data, and pre-empting board questions. ChatGPT handles bucket two and most of bucket three; bucket one stays manual but takes 15 minutes once the export routine is locked.

OpenAI's GPT-5 system card and prompt engineering guide flags structured-synthesis tasks as the top-ROI use case for the current generation. Board reports are exactly that: a small set of well-defined input artifacts produce a fixed-shape output artifact. The ten prompts below treat the report as a pipeline. Each prompt owns one section, accepts one input shape, and emits one output shape. The chain runs Sunday evening against GPT-5 or GPT-5-mini depending on synthesis depth — see the OpenAI model selection guide for current pricing and context windows.

ChartMogul emits MRR movements as new-business, expansion, contraction, churn, and reactivation. Boards want the story, not the table. This prompt produces the story.

**The prompt:**

``` You are a CFO writing the MRR section of the weekly board update. INPUT: - Current MRR: <number> - Last week's MRR: <number> - This week's movements from ChartMogul: new-business $X, expansion $Y, contraction $Z, churn $W, reactivation $V - Trailing 8-week MRR series: <list> - Top 3 new-business deals by ACV: <list> - Top 3 churn events by ARR lost: <list> OUTPUT: a 180-220 word narrative with: - Headline: net MRR delta + the one driver that explains most of it - Composition: which of new/expansion/contraction/churn moved the needle - 8-week context: is this week typical, an outlier, or a trend - Named accounts: cite the largest new-business and largest churn by name - One forward-looking sentence: what to watch next week Rules: - Never round so much that the math doesn't add. If new + expansion - contraction - churn != net, flag it. - Do not use the phrase "strong performance" or "momentum". - Cite the trailing 8-week average so the reader can place this week. - If churn > new-business, lead with that. Do not bury it. ```

**Why it works:** The arithmetic check ("if movements don't sum, flag it") catches the most common LLM error in financial synthesis — a plausible-sounding total that doesn't reconcile. The forbidden-phrase rule kills the two clichés that make CFO updates read like marketing copy.

**Sample output:** A 205-word narrative opening *"Net MRR moved +$8,400 to $412,300, the seventh consecutive week of positive net but the smallest delta in that window. Churn at $14,200 absorbed most of the $22,600 in new-business and expansion combined; the Acme Corp downgrade ($6,800) accounts for almost half of churn alone..."*

**When to use:** Sunday 6:00 p.m., as the first prompt in the chain. Output becomes the headline section of the report.

Cancellation reasons collected at the unsubscribe modal arrive as messy free text. Boards want a clean taxonomy: price, product fit, competitor, ICP miss, no-longer-needed. This prompt classifies and ranks.

**The prompt:**

``` You are a customer-success analyst classifying churn for the weekly board report. INPUT: a list of churn events from this week, each with: - Account name - ARR lost - Cancellation reason (free text from the modal or CS notes) - Tenure in months - Plan tier For each event, output: { "account": "<name>", "arr_lost": <number>, "root_cause": "price | product_fit | competitor_won | icp_miss | no_longer_needed | acquired_or_shutdown | unknown", "confidence": "high | medium | low", "evidence_phrase": "<verbatim quote from the cancellation reason, max 15 words>", "avoidable": true | false } Then output an aggregate block: { "by_root_cause": {<cause>: <total ARR lost>}, "avoidable_arr": <number>, "top_actionable_insight": "<one sentence — the single intervention that would have prevented the most ARR loss this week>" } Rules: - "unknown" is a valid answer; never guess a cause if the cancellation reason is empty or generic. - "avoidable" is true only if root_cause is price, product_fit, or competitor_won. - The evidence_phrase must be verbatim from the input — do not paraphrase. ```

**Why it works:** Forcing a verbatim evidence_phrase stops the model from inventing churn narratives. The avoidable flag turns the table into a prioritization tool, not just a report.

**Sample output:** A 9-row table with 3 price, 2 product_fit, 1 competitor_won, 1 acquired, 2 unknown. Top actionable insight: *"Two of three price churns were on the annual-to-monthly downgrade path — a 10% annual incentive would have retained $9,600 ARR."*

**When to use:** Sunday 6:08 p.m., immediately after prompt #1. Feed the output into prompt #8 for the variance section.

Boards already get the pipeline number from the dashboard. What they want from the founder is interpretation — is the pipeline real, is the conversion math intact, where is the risk.

**The prompt:**

``` You are a founder writing the pipeline section of the weekly board update. INPUT: - Current quarter quota: <number> - Closed-won YTD this quarter: <number> - Open pipeline (weighted): <number> - Open pipeline (unweighted): <number> - Trailing 4-quarter win rate: <percent> - Pipeline coverage ratio (open weighted / remaining quota): <number> - Top 5 open deals by ACV with stage and close date - Pipeline created this week: <number> - Pipeline created same week last quarter: <number> OUTPUT: a 200-250 word commentary with: - The coverage ratio framed against the historical win rate (e.g., "3.2x coverage at a 28% trailing win rate implies 90% confidence in hitting remaining quota") - One paragraph on velocity: is pipeline creation accelerating or decelerating versus same week last quarter - A risk paragraph: which of the top 5 deals is at risk and why - One ask of the board if any (intros, references, escalations) Rules: - Coverage of <3x at the current win rate must be called out as a risk. - Never describe a deal as "close to closing" without a specific stage and close date. - Do not use the phrase "healthy pipeline". - Cite the same-week-last-quarter comparison every time — it is the only honest velocity signal. ```

**Why it works:** The coverage-to-win-rate math is the standard test (Tom Tunguz's pipeline coverage analysis) and forcing the model to do it surfaces the actual risk. Banning "close to closing" without a stage stops the most common pipeline self-deception.

**Sample output:** A 235-word section opening *"Pipeline coverage sits at 2.8x against $1.4M remaining quota, below the 3.0x floor at our 26% trailing win rate. Pipeline creation this week was $340K versus $480K in the same week of Q1 — a 29% deceleration that warrants attention before week 4..."*

**When to use:** Sunday 6:15 p.m. Pair with prompt #8 (variance explainer) if pipeline missed its weekly target.

Hiring decisions land in front of the board late, when the offer is already out. Boards want the frame earlier — what the hire costs, what it buys, and how it changes runway.

**The prompt:**

``` You are a CFO framing a hiring decision for the board update. INPUT: - Current cash: <number> - Current net burn (monthly): <number> - Current runway (months): <number> - Proposed hires this week: list of {role, base, expected_start_date, expected_burn_contribution} - The KR or strategic goal each hire supports - Last fundraise: <date, amount, post-money> OUTPUT: a structured frame with: { "new_monthly_burn_after_hires": <number>, "new_runway_after_hires": <number>, "runway_delta_months": <signed number>, "per_hire_analysis": [ { "role": "<text>", "strategic_rationale": "<one sentence — what KR this hire moves and by how much>", "alternative": "<what we would do if we did NOT make this hire>", "reversibility": "easy | hard | one-way" } ], "runway_at_18_months_question": "<does the company still have 18 months of runway after these hires? yes/no/with_assumptions>", "board_ask": "<approval_required | informational | none>" } Rules: - If runway drops below 12 months, board_ask must be "approval_required". - Every hire must have an explicit alternative — the next-best use of that headcount slot. - Do not editorialize on whether the hire is good or bad — frame it. ```

**Why it works:** The 18-month runway test is the standard heuristic a16z's State of the Startup data uses for default-alive vs default-dead framing. Forcing an explicit alternative stops the runaway-hire pattern where the board sees the hire only after it's irreversible.

**Sample output:** A frame showing 2 proposed hires (Senior PM, Account Executive) cutting runway from 19 months to 17, board_ask = informational, with per-hire alternatives: *"AE alternative: contract SDR for 6 months at 40% of the cost while the pipeline coverage stabilizes."*

**When to use:** Whenever a hiring decision is on this week's agenda. Output goes directly into the report's people section.

Cohort retention curves are the second-most-skipped board section after R&D. Boards see the chart, do not interpret it, and the founder loses an opportunity to anchor product strength.

**The prompt:**

``` You are a product analyst writing the cohort retention paragraph for the board update. INPUT: - Cohort retention table (rows = month-of-acquisition cohort, cols = months since acquisition, values = % retained) - Latest mature cohort (6+ months) retention curve - Earliest mature cohort retention curve - Industry benchmark for our category (cite the source) OUTPUT: a 150-200 word paragraph with: - Month 1, month 3, month 6 retention for the latest mature cohort (specific percentages) - A direct comparison to the earliest mature cohort — is retention improving, flat, or decaying? - An interpretation: what does the shape of the curve say about product-market fit (Stripe's PMF retention shape guidance defines the smile-curve vs leak-curve test) - The benchmark comparison: how does our 6-month retention compare to ChartMogul's SaaS benchmark median for our segment? - One forward sentence: what cohort we are watching most closely Rules: - Use percentages, not ratios. "68% retained at month 6", not "0.68". - Never say a cohort is "strong" without citing a benchmark. - If retention is decaying cohort-over-cohort, lead with that. Do not bury it. ```

**Why it works:** Forcing the benchmark citation (ChartMogul or Stripe data) anchors the interpretation in something the board can verify. Banning "strong" without a benchmark kills the most common product-marketing tell.

**Sample output:** A 178-word paragraph: *"The Q4-2025 cohort retains 68% at month 6, versus 61% for the Q2-2025 cohort — a 7-point improvement that holds across the curve. ChartMogul's vertical-SaaS median for our segment is 64%, putting us above benchmark for the first quarter on record..."*

**When to use:** Sunday 6:25 p.m., or once per month if weekly cohort data is too noisy.

Engineering and design ship a lot every week. The board needs the top three things that moved a metric and the top three that slipped.

**The prompt:**

``` You are a head of product condensing R&D status for the weekly board update. INPUT: a list of all in-flight R&D projects with: - Project name - Owner - Original target ship date - Current status: shipped this week | on-track | at-risk | slipped - The KR or strategic outcome this project drives - A one-line summary written by the project owner OUTPUT: a structured condenser: { "shipped_this_week": [<top 3 by KR impact, with one-line outcome each>], "on_track": <count + names>, "at_risk_or_slipped": [<all, with one-line reason and the new target date>], "board_paragraph": "<120-160 word paragraph leading with what shipped, naming any slip, ending with what to expect next week>" } Rules: - The board paragraph must name specific shipped features — do not summarize them as "several improvements". - Every at-risk or slipped item must have a new target date or be flagged "NEEDS-RE-TARGET". - Do not include in-progress work without a ship date. ```

**Why it works:** The board cares about ship dates and outcomes, not effort. The structured output forces both. The NEEDS-RE-TARGET sentinel prevents the soft "we're still working on it" answer that masks indefinite slips.

**Sample output:** A condenser showing 3 shipped (one cited as moving activation rate +2 points), 7 on-track, 2 at-risk with specific re-targets. The board paragraph names the 3 shipped features and the 1 most-material slip.

**When to use:** Sunday 6:35 p.m., after the project owners have filed their Friday status notes.

Boards ask sharp questions. The cost of being unprepared is high. This prompt simulates the questions and drafts the answers.

**The prompt:**

``` You are a former board member doing a pre-mortem on this week's board update. INPUT: the draft board update (sections 1-6 from earlier prompts). OUTPUT: a list of 8-12 questions the sharpest director on this board would ask, each with: { "question": "<the verbatim question, max 25 words>", "category": "data | strategy | risk | hiring | competitive | financial", "why_they_will_ask": "<one sentence — what in the update prompts this question>", "draft_answer": "<60-90 word answer drawing only on data in the update; flag with NEEDS-DATA if the answer requires information not in the update>", "counter_question": "<the follow-up the director will ask if the draft answer is weak>" } Rules: - At least 3 questions must be NEEDS-DATA flagged — if every question can be answered from the update, the pre-mortem missed. - Do not generate softball questions. Each must be one a director would actually ask in a heated meeting. - Lead with the financial and risk questions. ```

**Why it works:** Requiring NEEDS-DATA flags surfaces the gaps in the report before the board does. The counter-question field forces the founder to think two moves ahead.

**Sample output:** A list of 10 questions opening with *"You showed pipeline coverage at 2.8x — at a 26% win rate, that's below the comfort threshold. What's your plan to add $300K of pipeline by Friday, and what does that imply about Q3 marketing spend?"* with a 75-word draft answer and a counter-question.

**When to use:** Sunday 7:00 p.m., after the report is drafted. Spend 15 minutes filling NEEDS-DATA gaps before the meeting.

Every weekly report has a metric that missed plan. Boards want the variance explained, not justified. This prompt produces variance commentary that earns trust.

**The prompt:**

``` You are a CFO writing the KPI variance section of the weekly board update. INPUT: a list of KPIs with: - Metric name - Planned value (this week) - Actual value (this week) - Variance (signed number and percent) - Last 4 weeks of actuals - The owner of the metric For each KPI with |variance| > 10%, output: { "metric": "<name>", "variance": <signed percent>, "trend_context": "<one sentence — is the variance a one-week event or part of a trend>", "primary_driver": "<one of: timing_shift | external | execution | plan_was_wrong | data_quality>", "narrative": "<60-80 word explanation tying the driver to a specific event or input; cite at least one number>", "correction": "<the specific action being taken to close the variance or rebaseline; or NONE if the variance is acceptable>" } Rules: - "timing_shift" is acceptable as a driver only if the metric is forecast to recover within 2 weeks; otherwise reclassify. - "plan_was_wrong" is a legitimate driver — do not avoid it. - Never write "we expect this to improve" without naming the specific lever. - If three or more KPIs miss plan in the same week, add a meta-comment: "a plan-quality review is warranted". ```

**Why it works:** The drivers taxonomy comes from financial planning practice (CFO.com variance analysis primer) and the timing_shift guardrail stops the most common cop-out. The plan-quality meta-comment surfaces when the issue is the plan itself.

**Sample output:** A 3-KPI variance section: trial signups -18% (driver: execution, narrative cites the paid-channel CPC spike), expansion MRR +24% (driver: timing_shift, narrative cites a delayed enterprise close from last week), NPS -8 points (driver: data_quality, narrative cites the survey sample size dropping to 40).

**When to use:** Sunday 7:15 p.m., after MRR (prompt #1) and pipeline (prompt #3) are drafted.

Some investors get the full report; some get the one-pager. This prompt distills.

**The prompt:**

``` You are a founder writing the one-page investor update derived from the full weekly board report. INPUT: the full draft board report (sections 1-8 from earlier prompts) plus an explicit "audience" tag: "all_investors" or "strategic_only" or "institutional_only". OUTPUT: a one-page (450-550 word) update with this exact structure: 1. Headline (one sentence): the most important thing that changed this week 2. The numbers (5 bullets max): MRR, net new, churn, runway, pipeline coverage 3. The wins (3 bullets max): one named customer, one shipped feature, one team milestone 4. The lessons (2 bullets max): what we learned that changes how we operate 5. The asks (2 bullets max): intros, references, or specific help Rules: - Total length must be 450-550 words. Hard limit. - Adjust depth by audience: all_investors gets headline numbers, strategic_only gets product depth, institutional_only gets financial depth. - Every bullet must contain a number or a name. Generic bullets are forbidden. - Do not include forward guidance unless you would say it in a 1:1. - End with one sentence on what next week's update will likely report. ```

**Why it works:** The audience-conditioning rule (strategic vs institutional) maps to the Lenny startup-ops survey finding that 71% of founders send the same update to all investors and lose differentiation. The number-or-name requirement kills generic-bullet syndrome.

**Sample output:** A 510-word update for all_investors opening with the MRR headline, 5 number bullets, 3 wins (Acme expansion +$2,400, shipped onboarding v3, hired a Senior PM), 2 lessons (paid-channel CPC needs a 30-day cooldown, annual-to-monthly downgrade is the largest avoidable churn vector), 2 asks (intro to RevOps leaders at scaled SaaS, reference for a mid-market deal).

**When to use:** Sunday 7:30 p.m. Feed the report output directly in.

The board report is downstream of exec-team Friday status notes. Most of the prep time goes into chasing and reformatting those notes. This meta-prompt sits one layer up: it generates the Friday template each exec fills, tuned to what the report needs.

**The prompt:**

``` You are a chief of staff generating the Friday exec-team update template for this week. INPUT: - The previous week's board report - The current OKR set - A list of execs and their domains - Specific board questions that came up in the last meeting that require continued reporting For each exec, output a domain-specific template with: { "exec": "<name>", "domain": "<text>", "sections": [ { "prompt": "<the specific question this exec should answer in their update>", "input_required": "<the artifact they should attach or cite>", "length_budget": "<word count or bullet count>", "feeds_board_section": "<which numbered section of next week's board report consumes this>" } ], "sla": "<the specific Friday deadline by which this is due>" } Rules: - Every section must trace to a specific board-report section (1-9) or be flagged as "context-only". - Length budgets must be strict — if execs write essays, the board prep balloons. - Re-use the same template across weeks unless the report structure changes; this creates muscle memory. - If a section repeatedly produces low-quality input from a specific exec, flag it for the chief of staff. ```

**Why it works:** The feeds_board_section field turns the exec template into a precursor to the report, eliminating the translation step that consumes the most prep time. The trace requirement aligns with the Diligent Modern Governance practice guide on report-input discipline.

**Sample output:** Five exec templates (CEO, CFO, CRO, Head of Product, Head of People), each with 3-5 prompts that map directly to sections 1-9 of the board report. The CFO template explicitly says: *"Submit ChartMogul export and Stripe ledger by Friday 4 p.m. ET; feeds board section 1."*

**When to use:** Monday morning, sent to execs. Re-generate quarterly or when board questions shift.

The comparison table below summarizes input data, output format, and the time each prompt replaces. The 10 prompts collectively replace ~5 hours of synthesis work with ~60 minutes of running the chain plus reviewing outputs.

<a href="https://aipromptshub.co/blog-post-outline?utm_source=aipromptshub&utm_medium=blog&utm_campaign=board-reports-compare" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold",marginTop:"12px"}}> Build a custom board-report template → </a>

The chain that replaces a Sunday-afternoon-lost-to-prep with a focused 60-minute ritual:

1. **Sunday 5:45 p.m. (15 min human).** Pull the input artifacts: ChartMogul MRR export, Stripe ledger, HubSpot pipeline CSV, ATS open-req list, R&D status notes. This is the only manual step; the rest is prompts. 2. **Sunday 6:00 p.m. (3 min compute).** Run prompt #1 (MRR delta narrative) against the ChartMogul export. Read once for arithmetic sanity. 3. **Sunday 6:08 p.m. (3 min compute).** Run prompt #2 (churn root cause) against this week's cancellation notes. Verify the verbatim evidence_phrases match input. 4. **Sunday 6:15 p.m. (4 min compute).** Run prompt #3 (pipeline coverage commentary) against the HubSpot pipeline CSV plus the trailing 4-quarter win rate. 5. **Sunday 6:20 p.m. (4 min compute).** Run prompt #4 (hiring vs runway) only if hiring is on this week's agenda. 6. **Sunday 6:25 p.m. (4 min compute).** Run prompt #5 (cohort retention) monthly, or weekly if data is stable. 7. **Sunday 6:35 p.m. (5 min compute).** Run prompt #6 (R&D condenser) against the exec-team status notes. 8. **Sunday 7:00 p.m. (5 min compute + 15 min human).** Run prompt #7 (board-question pre-mortem) against the assembled draft. Fill any NEEDS-DATA gaps now — before the board sees the report, not during the meeting. 9. **Sunday 7:15 p.m. (3 min compute).** Run prompt #8 (KPI variance) against the planned-vs-actual table. 10. **Sunday 7:30 p.m. (3 min compute + 5 min human).** Run prompt #9 (investor one-pager) if an investor update goes out this week. Polish, send Monday morning.

Total: ~60 minutes elapsed, of which ~35 minutes is human review and ~25 minutes is compute. The Monday morning template generation (prompt #10) runs once per quarter and pays back across 13 weeks.

Compute cost at GPT-5 prices (OpenAI pricing) for the full chain runs ~$0.40–$0.80 per week — less than a single Sunday coffee, replacing a 5-hour synthesis tax.

<a href="https://aipromptshub.co/chatgpt-prompt-generator?utm_source=aipromptshub&utm_medium=blog&utm_campaign=board-reports-chain" style={{display:"inline-block",padding:"10px 18px",background:"#0a66ff",color:"white",borderRadius:"6px",textDecoration:"none",fontWeight:"bold",marginTop:"12px"}}> Get the chained Sunday ritual template → </a>

### Which ChatGPT model should I use for board-report prompts?

GPT-5 is the right default for prompts 1, 3, 5, 7, and 9 — the synthesis-heavy ones where the narrative quality drives board perception. GPT-5-mini is sufficient for prompts 2, 4, 6, 8, and 10 because the structural output is well-defined and the narrative budget is short. The full chain at GPT-5 prices runs under $1 per week. See the OpenAI model documentation for the current generation and context windows.

### Will ChatGPT invent numbers if my input is incomplete?

Yes, unless you constrain it. Every prompt above includes either an explicit verbatim-quote requirement (prompt #2), an arithmetic-check rule (prompt #1), or a NEEDS-DATA sentinel (prompt #7) to surface the gap rather than paper over it. The OpenAI prompt engineering guide covers the underlying technique — explicit refusal-to-fabricate instructions — in detail.

### Can these prompts replace the CFO or the chief of staff?

No. They replace the synthesis labor; they do not replace the judgment about what to disclose, how to frame a miss, or which board question deserves a longer answer than the draft. A human reviews every output before it goes to the board. The right model is augmentation: the chain produces 80% of the artifact, the operator owns the last 20% that earns trust.

### How do I integrate these prompts with ChartMogul, Stripe, and HubSpot?

All three platforms expose JSON exports and webhook APIs. The most common pattern in 2026 is a Sunday-evening Zap or workflow that drops the previous week's exports into a shared folder, then a single script that paste-loads them into the prompt inputs. Stripe's Sigma data warehouse and ChartMogul's export API are documented for this use case. HubSpot's reporting API covers the pipeline CSV.

### What if the board wants more frequent updates than weekly?

Run prompts 1, 2, 3, and 8 (the numerics) twice a week and prompts 5, 6, 9, and 10 weekly. The synthesis prompts (5, 6) need at least a week of fresh data to produce a meaningful delta; running them more often produces noise. The NACD's 2025 governance survey found that 23% of boards now request twice-weekly numeric updates with weekly narrative — the chain accommodates the split.

### Are the sample outputs above synthesized or real?

Synthesized for illustration. Real outputs vary by model, temperature, and input quality. The structure, constraint compliance, and arithmetic discipline are representative of GPT-5 with the prompts as written; the specific company names and numbers are illustrative.

### How do I keep these prompts current as ChatGPT evolves?

The prompts encode rules — arithmetic checks, forbidden phrases, structural constraints, evidence-phrase requirements — rather than relying on a specific model's default behavior. As GPT-5 evolves into GPT-6 and beyond, the same rules keep producing the same artifact shape; the underlying narrative quality improves. Re-test the chain against each major OpenAI release; the OpenAI release notes are the source of truth.

- NACD 2025 Public Company Governance Survey — board reporting cadence data. - Diligent 2025 Modern Governance Summit data — distributed-board and AI-literate-director research. - Lenny Rachitsky 2025 startup-ops survey — founder time-on-prep benchmarks. - ChartMogul SaaS Benchmark Reports — retention and MRR-movement benchmarks. - Stripe Atlas guides — PMF retention shape definitions. - Stripe Sigma documentation — ledger-to-warehouse pattern. - OpenAI prompt engineering guide — structured-synthesis prompt patterns. - OpenAI model documentation — GPT-5 / GPT-5-mini selection. - Tom Tunguz pipeline coverage analysis — coverage-to-win-rate math. - CFO.com variance analysis primer — driver taxonomy. - a16z State of the Startup data — default-alive 18-month runway heuristic.

---