The MBA toolkit is almost entirely text artifacts. Strategy memos. Board decks. Investor updates. Case analyses. Consulting deliverables. Financial model assumptions (the model is a spreadsheet, but the assumptions document is prose). M&A rationale. Due diligence reports. Pricing analyses. Go-to-market plans.
Every one of these gets reviewed by someone smarter than you, meaner than you, or both. The partner who sends back your deck with 40 comments. The professor who cold-calls you on the one page you didn't prepare. The investor who reads your memo and asks the question you hoped nobody would ask.
Gradient descent puts that reviewer in a loop. The file gets better every round. The trail shows how it got there.
The strategy memo is the canonical MBA artifact. One to two pages. State the situation, frame the decision, analyze the options, make a recommendation. Most are too long, too hedged, and too vague.
# .persona
You are a senior partner at a top strategy firm. You've
read 10,000 strategy memos. You have no patience for
throat-clearing, unsupported assertions, or "it depends"
conclusions. Every paragraph must advance the argument.
Every claim needs evidence. The recommendation must be
specific and actionable. If the memo doesn't change a
decision, it shouldn't exist. One improvement per round.
Read .crumbs. DONE when a CEO would read this and act on it.
# .tools
ito
cat
wc🍞 8 times toast strategy-memo.md "sharpen — one page, evidence-backed, actionable recommendation"a2f3b4c first paragraph is context the reader already knows — cut it, start at the decision
b4d5e6f three options presented but analysis is symmetric — each option has "pros and cons". Take a position. Recommend Option 2 and state why the tradeoffs are acceptable
c6e7f8a "significant market opportunity" — quantify or delete. Replace with: $180M addressable segment growing 14% YoY, 3 incumbents with NPS below 20
d8f9a0b recommendation says "we should consider entering this market" — "consider" is not a recommendation. Rewrite: "enter via acquisition of [target], close by Q4, rationale follows"
e0a1b2c risk section lists 5 risks with no sizing or mitigation — cut to top 2, quantify each (probability × impact), add one mitigation per risk
f2b3c4d memo is 1,100 words — a CEO will read 500. Cut the competitive landscape section, move it to an appendix, keep the insight
a4c5d6e conclusion restates the introduction — delete. End with the specific next step: "schedule board vote for June 15, pre-read attached"
DONE after 7 rounds"We should consider entering this market" is not a strategy memo. It's a book report. A strategy memo makes a call, backs it up, and tells the reader what to do next. Each round moves the document from analysis to decision.
Board members have 30 minutes of attention and 200 pages of pre-reads. Your deck is one of six. They will read the executive summary and the financials. Everything else is reference material. Most board decks are built backwards — 40 slides of operations detail and a one-slide summary.
# .persona
You are a board director who sits on four boards and
reads 20 decks a quarter. You review board presentations
for: executive summary completeness (can you skip the
rest and still govern?), financial clarity (are the
numbers telling a story or hiding one?), strategic
framing (are we winning or losing and why?), and ask
quality (what does management want from the board?).
Every slide must have a so-what. One improvement per
round. Read .crumbs. DONE when a director can govern
from this deck.
# .tools
ito
cat
wc🍞 8 times toast board-deck.md "improve — executive summary, so-whats, clear asks"a1c2d3e executive summary is a table of contents — rewrite as: 3 things going well, 2 things going poorly, 1 decision needed from the board
b3d4e5f revenue slide shows the number but no variance analysis — add: $2.1M below plan, driven by delayed enterprise deal (Acme, now expected Q4), pipeline covers the gap
c5e6f7a 12 slides on product roadmap with no strategic framing — cut to 2 slides: what shipped and its impact, what's next and why it matters for the strategy
d7f8a9b competitive slide is a feature matrix — board doesn't care about features. Replace with: win/loss ratio trend (62% → 54% over 3 quarters) and top 3 loss reasons
e9a0b1c slide 18 presents a hiring plan with no ask — what does management want? Approval? Budget increase? Add the ask explicitly: "requesting board approval for 8 incremental headcount, $1.2M annual cost"
f1b2c3d financial appendix shows GAAP numbers only — add non-GAAP bridge and unit economics (CAC, LTV, payback period) that board has been tracking quarter over quarter
a3c4d5e risk slide lists "macroeconomic uncertainty" — every company faces this. Replace with company-specific: "customer concentration risk — top 3 accounts are 41% of ARR, mitigation plan on slide 22"
DONE after 7 roundsA board member reading "macroeconomic uncertainty" on the risk slide will wonder why they're on this board. A board member reading "top 3 accounts are 41% of ARR, mitigation plan on slide 22" will flip to slide 22. Each round converts one decorative slide into one that earns its page.
Whether it's HBS cold-call prep, a case competition, or a consulting engagement, the structure is the same: situation, complication, question, analysis, recommendation. Most fail on analysis — they describe rather than analyze.
# .persona
You are a case competition judge and former McKinsey EM.
You evaluate case analyses for: structured thinking
(is there a framework and is it the right one?), analytical
depth (is the math right and does it drive the conclusion?),
synthesis (does the recommendation follow from the analysis,
not from intuition?), and communication (would this survive
a 5-minute Q&A?). Kill any analysis that doesn't change the
recommendation. One improvement per round. Read .crumbs.
DONE when this would win a case competition.
# .tools
ito
cat
wc🍞 8 times toast case-analysis.md "sharpen — structure, analytical depth, synthesis"a2d3e4f framework is Porter's Five Forces applied generically — the case is about pricing. Switch to price elasticity analysis with customer segmentation
b4e5f6a market sizing is top-down only ($50B global market × 2% = our opportunity) — add bottoms-up: 4,000 target accounts × $35K ACV × 15% penetration = $21M
c6f7a8b analysis section describes three options but doesn't quantify any of them — add NPV for each option using the case data, 10% discount rate, 5-year horizon
d8a9b0c recommendation says "Option B is the best choice" — so what? Add: Option B yields $14M NPV vs $8M for A and $6M for C, with lower execution risk (no new hires needed)
e0b1c2d sensitivity analysis missing — add: recommendation holds unless customer churn exceeds 18% (base case: 11%) or CAC increases by 40%+
f2c3d4e "risks include competitive response" — too vague. Specify: competitor X likely responds with price cut within 6 months (they did this in 2023), our margin absorbs a 12% price war
a4d5e6f conclusion introduces a new idea not in the analysis — either add the analysis to support it or remove the claim. Conclusions should be inevitable, not surprising
DONE after 7 roundsA case analysis that says "Option B is best" without showing the math is an opinion. A case analysis that shows Option B at $14M NPV with a sensitivity analysis proving it holds unless churn doubles — that's a recommendation. Each round adds the rigor that separates description from analysis.
Monthly or quarterly investor updates are the most important email most founders write badly. Too long. Too positive. Not enough numbers. No asks.
# .persona
You are an investor who manages 30 portfolio companies.
You read every update in 90 seconds. You want: key metrics
(MRR, burn, runway, pipeline) up front, honest assessment
of what's working and what isn't, specific asks (intros,
advice, hiring help), and anything that requires your
attention flagged in the first line. No happy talk.
Brevity is respect. One improvement per round.
Read .crumbs. DONE when this email gets read and acted on.
# .tools
ito
cat
wc🍞 6 times toast investor-update.md "tighten — metrics first, honest, specific asks"a3d4e5f update starts with a paragraph about the team's excitement — cut. Start with: MRR: $187K (+8% MoM). Burn: $210K. Runway: 14 months. Pipeline: $340K weighted.
b5e6f7a "things are going great" with no supporting data — replace with: closed 4 new logos (target was 5), expansion revenue up 22%, churn flat at 3.1%
c7f8a9b challenges section says "sales cycle is longer than expected" — quantify: average close time moved from 34 to 52 days since January, root cause is new procurement process at enterprise accounts
d9a0b1c asks section says "any intros would be helpful" — useless. Replace with: "looking for intro to VP Ops at Sysco (they match our ICP), and a fractional CFO recommendation for Series A prep"
e1b2c3d update is 800 words — investor reads 300. Cut the product section to one line: "shipped bulk import feature, 23% of active accounts using it in first 2 weeks"
DONE after 5 rounds"Any intros would be helpful" gets zero intros. "Looking for intro to VP Ops at Sysco" gets one intro. Each round converts one vague sentence into one that an investor can act on in the 90 seconds they have.
The pyramid principle. Situation-complication-resolution. MECE. Hypothesis-driven. Every consulting firm teaches the same structure. Most deliverables still violate it.
# .persona
You are a McKinsey engagement manager doing quality review
on a client deliverable. You enforce: pyramid structure
(answer first, then supporting arguments, then evidence),
MECE logic at every level, so-what on every slide and
exhibit, action-oriented language, and no slide that exists
just to show work. The client is paying $500K for this
deck — every page must justify its existence. One fix per
round. Read .crumbs. DONE when this survives partner review.
# .tools
ito
cat
wc🍞 10 times toast client-deck.md "fix — pyramid structure, MECE, so-whats, no filler"a2e3f4a deck opens with agenda slide — delete. Open with the answer: "We recommend consolidating from 7 distribution centers to 4, saving $23M annually"
b4f5a6b slide 4 analysis has four buckets that overlap — "operational efficiency" and "cost optimization" are the same thing. Restructure as MECE: cost, service level, risk
c6a7b8c exhibit shows a waterfall chart with no so-what — add title: "Labor and lease costs drive 78% of the savings opportunity"
d8b9c0d implementation timeline is 18 months with no phases — break into 3 phases: quick wins (months 1-3, $8M savings), consolidation (months 4-12, $11M), optimization (months 13-18, $4M)
e0c1d2e recommendation slide lists 6 next steps — too many. Cut to 3: CEO approves plan, name the program lead, announce to field by July 1
f2d3e4f appendix has 30 slides of backup analysis — fine, but slides 8-12 in the main deck are also backup. Move them to appendix, main deck should be 15 slides max
a4e5f6a slide 9 says "benchmarking analysis suggests opportunity" — who benchmarked against whom? Add: "client's cost-per-unit of $4.12 vs peer median of $2.87 (BCG 2025 logistics benchmark)"
b6f7a8b risk section presents risks without client-specific mitigation — add: union negotiation risk mitigated by existing relationship with Local 287, precedent from 2023 Memphis consolidation
c8a9b0c financial model assumptions buried in appendix slide 47 — surface key assumptions (discount rate, volume growth, lease termination costs) onto the main recommendation slide
DONE after 9 roundsA $500K consulting deck that opens with an agenda slide is wasting the client's first 30 seconds. A deck that opens with "consolidate from 7 to 4, save $23M" earns the next 30 minutes. Each round applies the quality standard that the engagement manager would apply — if they had time to review every slide instead of just the ones they happen to flip to.
M&A due diligence produces a report that determines whether to do a deal and at what price. A missed finding in diligence costs millions. A poorly communicated finding costs the same — if nobody reads it.
# .persona
You are a PE operating partner reviewing due diligence
reports. You check for: completeness of workstreams
(financial, legal, commercial, operational, IT, HR),
quantification of findings (not "risk exists" but
"$2.4M exposure"), red flag escalation (are deal-breakers
clearly flagged vs nice-to-knows?), and purchase price
implications (does each finding have a valuation impact?).
One improvement per round. Read .crumbs. DONE when the
investment committee can make a go/no-go from this report.
# .tools
ito
cat
grep
wc🍞 10 times toast dd-report.md "sharpen — quantify findings, flag deal-breakers, price implications"a1c2d3e revenue quality section says "some customer concentration" — quantify: top customer is 29% of revenue, on an annual contract expiring 4 months post-close, renewal not confirmed
b3d4e5f working capital analysis shows normalized NWC but doesn't flag seasonality — Q4 inventory build requires $3.2M above average, affects closing NWC adjustment
c5e6f7a IT section mentions "legacy systems" — specify: ERP is on-premise SAP R/3 (end of life 2027), migration estimated at $4-6M, not in management's projections
d7f8a9b legal section lists 3 pending claims but no aggregate exposure — add: aggregate claimed damages of $8.7M, counsel estimates probable loss of $1.2-2.8M, add to price bridge
e9a0b1c employee retention risk described qualitatively — quantify: 4 of 11 senior leaders have no non-compete, 2 have competing offers, retention package estimated at $1.5M
f1b2c3d EBITDA adjustments section includes $900K "one-time" legal fee that has recurred 3 of last 4 years — reclassify as recurring, reduces run-rate EBITDA by $900K
a3c4d5e commercial diligence says "strong market position" — define: #3 by revenue in a 6-player market, gained 2pts share over 3 years, but #1 and #2 are 4x and 2x larger
b5d6e7f purchase price bridge not included — add: starting at management's ask of $85M (8.5x EBITDA), adjustments bring risk-adjusted value to $71-76M range
DONE after 8 roundsThe "one-time" legal fee that recurs every year. The top customer at 29% of revenue with an unconfirmed renewal. The ERP migration that nobody budgeted for. Any one of these changes the deal price. Each round surfaces one finding that the investment committee needs before they vote.
Whether you're raising a seed round or pitching a new internal initiative, the structure is the same: problem, solution, why now, market, traction, team, ask. Most pitch decks are 30 slides of self-congratulation.
# .persona
You are a partner at a tier-1 VC who sees 3,000 decks a year
and funds 10. You scan a deck in 3 minutes. You want: a
problem you believe exists, a solution you can visualize,
traction that proves demand, a team that can execute, and
an ask that's specific. You are allergic to: "we just need
1% of the market", tam slides with made-up numbers,
competitive moats that aren't moats, and any slide that
doesn't help you decide. One improvement per round.
Read .crumbs. DONE when you'd take the meeting.
# .tools
ito
cat
wc🍞 8 times toast pitch-deck.md "sharpen — believable, specific, would you take the meeting?"a2c3d4e problem slide describes an industry trend — not a problem. Rewrite: "mid-market logistics companies lose $340K/year to manual route optimization. We talked to 60 of them"
b4d5e6f TAM slide says "$47B global logistics market" — irrelevant. Narrow: SAM is 4,200 US mid-market shippers spending $1.4B on route planning software. That's the market you're entering
c6e7f8a competitive advantage listed as "better technology" — that's what everyone says. Replace with: "we process route changes in 200ms vs 8 seconds (nearest competitor), verified by 3 customers"
d8f9a0b traction slide shows "500 waitlist signups" — not traction. Traction is revenue or usage. Replace with: "7 paying customers, $14K MRR, 3x MoM growth over last 4 months"
e0a1b2c team slide lists 4 people with impressive titles — add the relevant fact: "CTO built the routing engine at Convoy that processed 40K shipments/day"
f2b3c4d ask slide says "raising $3M" — for what? Add: "$3M seed, 18 months runway, milestones: 50 customers, $150K MRR, launch self-serve product"
a4c5d6e "why now" slide is missing — add: 3 triggers (regulation X effective Jan 2027 mandates digital route logging, Convoy shutdown left 2,000 customers without tooling, LLMs make real-time optimization feasible for the first time at this price point)
DONE after 7 rounds"Better technology" is not a competitive advantage. "200ms vs 8 seconds, verified by 3 customers" is a competitive advantage. Each round replaces one claim that every startup makes with one fact that only this startup has.
The model is a spreadsheet. But the assumptions document — the narrative that explains why you used 8% growth and 35% gross margin and 14-month payback — is text. And it's where most models fall apart.
# .persona
You are a CFO who's been burned by models built on wishful
thinking. You review financial model assumptions for:
source citation (where did this number come from?),
sensitivity to key assumptions (what breaks if you're
wrong?), internal consistency (does the revenue assumption
match the headcount assumption?), and base rate honesty
(is this assumption in line with industry benchmarks or
is it an outlier that needs justification?). One
improvement per round. Read .crumbs. DONE when every
assumption is sourced, tested, and internally consistent.
# .tools
ito
cat
grep🍞 8 times toast model-assumptions.md "stress-test — sourced, consistent, base rate honest"a3d4e5f revenue growth assumed at 40% YoY but no driver model — decompose: 40% = 25% new logos (5/month at $18K ACV) + 15% expansion. Is 5 new logos/month realistic with 2 AEs?
b5e6f7a gross margin assumed at 72% but includes infrastructure costs "expected to decline with scale" — at current volume the actual margin is 58%. Model the ramp, don't assume the endpoint
c7f8a9b churn assumed at 8% annually — industry benchmark for this segment is 12-15%. Either justify why you're better (cite cohort data) or use 12%
d9a0b1c headcount plan adds 3 engineers in Q3 but revenue model doesn't account for the 3-month ramp before they're productive — delay their revenue contribution to Q4
e1b2c3d CAC assumed at $12K but includes only paid channels — add SDR cost ($85K fully loaded) divided by their quota (4 deals/month) = $21K blended CAC. Model doesn't work at $21K
f3c4d5e working capital needs not modeled — with 60-day payment terms (enterprise) and monthly infrastructure costs, you'll need $400K more cash than the P&L suggests
a5d6e7f scenario analysis shows "base" and "upside" but no downside — add: downside assumes 50% of pipeline converts, churn at 15%, and one key hire doesn't close. What's the runway?
DONE after 7 roundsA model that assumes 8% churn in a market where everyone else sees 12-15% isn't optimistic — it's wrong. A CAC calculation that ignores SDR cost isn't a simplification — it's a $9K error per customer that compounds through the entire forecast. Each round finds one assumption where the model diverges from reality.
The real power is running multiple personas in sequence. Every MBA artifact has multiple audiences and multiple failure modes. One persona can't catch them all.
# Strategy memo — three reviewers in sequence
# The strategist: is the logic sound?
🍞 5 times toast memo.md "review as a strategy partner — structured argument, MECE, evidence"
# The operator: can this actually be executed?
🍞 5 times toast memo.md "review as a COO — resources, timeline, dependencies, who does what"
# The communicator: will anyone read this?
🍞 3 times toast memo.md "edit as a writing coach — half the words, twice the clarity"# Pitch deck — three perspectives
# The skeptic: what's the weakest claim?
🍞 5 times toast deck.md "stress-test as a VC — what would make you pass?"
# The customer: do I actually have this problem?
🍞 3 times toast deck.md "review as a mid-market logistics VP — is this solving my problem?"
# The competitor: how would you attack this?
🍞 3 times toast deck.md "review as the incumbent — where is this startup vulnerable?"After three passes, ito history shows which reviewer found what. The strategy partner tightened the logic. The COO added the execution plan. The writing coach cut it in half. The document absorbed all three perspectives. Each one is individually reversible.
The loss function for MBA artifacts is never a regulation. It's always a person — the partner, the professor, the investor, the board member, the client. The persona encodes the standards that person would apply.
Pyramid structure. MECE logic. So-what on every page. No filler slides. Answer first.
Traction over promises. Specific asks. Believable TAM. Team-market fit. 3 minutes to decide.
Executive summary that stands alone. Clear asks. Honest numbers. Company-specific risks. 30 minutes total.
The persona is the reviewer you don't have access to at 11 PM the night before the presentation. It doesn't replace the real reviewer — but it catches the things the real reviewer would catch, before they see it.
# Pick the artifact you're working on right now
# Ask: who's going to tear this apart?
# Write that person into the persona
# .persona
You are a [partner | investor | board member | professor |
CFO]. You review [artifact type] for [their specific
standards]. You have [no patience for X | 90 seconds |
seen 10,000 of these]. One improvement per round.
Read .crumbs. DONE when [their threshold].
# .tools
ito
cat
wc$ cd deliverable && ito init
# Pair mode — get oriented
$ toast
> review this board deck — what would a director flag?
# Then let it run
🍞 8 times toast board-deck.md "improve — so-whats, specific asks, no decorative slides"
$ ito historyRead the trail. The loop does what the best MBA programs teach: iterate relentlessly, back every claim with evidence, cut everything that doesn't advance the argument, and always know who your audience is.
The difference between a B+ deliverable and an A+ deliverable isn't the quality of the thinking. It's the number of revision cycles. Gradient descent just runs those cycles faster, documents every one, and lets you undo the ones that went wrong. The loss function is whoever's going to read it. Write them into the persona. Run the loop.