Deck Research, Stress-Tested
The research brief's rules, run against real decks: 11 famous early-stage decks read from the actual slide images, 13 real B2B sales decks, a forensic deep-dive on the Zuora deck, April Dunford's real client slides, founder-led selling accounts, and Gong's telemetry across 800k+ recorded meetings. Every verdict below says where the evidence came from.
The verdicts
HOLDS Lean slides — real winners run 18–25 words
Zuora's famous deck: median 18 words per slide. Moz: median 25. The controlled experiment's winning condition: 21. The one text-wall in the corpus is a HubSpot partner template — a thing nobody ever won with, published as a fill-in kit. Our email set sits at ~50, appropriate for an unaided read, and the trimming instinct stays.
Zuora 24-slide count · Moz transcript · Garner & Alley 2013
HOLDS No product screenshots
Zero product screenshots across all 24 verified decks — including Mixpanel, whose product is literally charts, and Zuora. The strongest unanimous pattern in the corpus.HOLDS Nothing about yourself first; team late
Team slides in real decks sit 76–92% of the way through, or don't exist. Reid Hoffman, on his own deck: "One common mistake is putting the team slide early… Instead, open with the investment thesis." Our team slide is second-to-last. LinkedIn 34/38 · Buffer 12/13 · Airbnb 11/14 · Hoffman's annotations
HOLDS Sales decks end with an explicit ask
Dunford's real structure (eight steps, corrected from the four-box simplification) ends with The Ask. Linear ends on a dual call-to-action, folk on a named account executive, ChartMogul on an email address. Investor decks often skip the ask — different genre. Ours ends on the 30-day pilot with success criteria.HOLDS The presented deck and the sent deck are different documents
Peter Kazanjy (the one founder with real sales slides published): "there's the one you present… and then there's the one that you'll send later." Reynolds and Duarte say the same from the design side. Our separator slide implements exactly this split.HOLDS Build the deck around the objections
Canva's Melanie Perkins: "Every time we'd get a really tricky question from investors we'd iterate on our deck to put the hardest questions they asked right at the front — so they'd stop asking us the same question." Dunford includes an objections step. Our objection slides and bridge follow the same construction.HOLDS — strengthened Discovery calls are for talking, not presenting
Gong, 803,402 recorded meetings: pulling up slides early in discovery correlates with worse outcomes — questions asked drop 21%, monologues run 25% longer. Slides earn their keep mid-to-late stage. Our internal rule ("talk 20%, show 3 slides max, slides only in the Connect phase") was already shaped this way; the data hardens it.REFINED Assertion headlines — for the argument, not everywhere
The corpus complicates the rule. Five successful seed decks (Airbnb, Uber, Buffer, Intercom, Coinbase) used zero assertion headlines — but those decks were narrated in the room. Mixpanel splits it perfectly: full-sentence claims on the narrative slides, terse labels on the data slides. Hoffman hedges: "helpful (but not mandatory) to put your thesis in each of the titles." The refined rule: slides that make your case assert; slides that show receipts can label — and the assertion rule matters most in a sent deck, where no voice fills the gap. Since ours is sent-first, our email set keeps assertions. Also real: question headlines (Dropbox used four) are a legitimate third form. And a correction for the record: Kevin Hale never stated the assertion rule — his rules are "legible, simple, obvious" plus the stranger test; crediting him with more would be fabrication.REFINED Pricing in the deck — right for us, not universal
Reality check: only 3 of 11 real sales decks show prices (folk: real per-seat prices; Splunk: a full list-price table right after the demo; HubSpot: slide 32 of 34). Enterprise decks with negotiated pricing omit it. The rule survives for PropFlow specifically: transparent per-unit pricing, founder-led motion, engagement data showing pricing is the most-read slide in sent decks — and our own Aug 6 meeting, which got confused precisely because pricing wasn't shown. Splunk proves listing real prices in a sales deck is a normal, safe practice.REFINED When you do present, keep the deck under 10 minutes
New number from 121,828 meetings: won deals averaged 9.1 minutes of deck time; lost deals 11.4. Added to the run-of-show slide ("2–3 slides, under 10 minutes"). Beware the same article's famous "every closed deal used a deck under 9 minutes" story — the author admits it's fictional; only 9.1 vs 11.4 is data.NEW The best founder-led sellers barely used decks at all
The most sobering finding: at the earliest stage, the famous wins came without slides. Gusto's Josh Reeves pitched verbally with a price — "Can I sign up?" — and got 17 of 20. Gong told beta users "beta is over, time to buy": 11 of 12 converted. Vanta's first sales artifact was a custom gap-assessment spreadsheet built for each prospect. Jen Abel's rule: three to four sentences, don't even talk about the solution. The deck supports the sale; the personalized, concrete follow-up is the sale. For us that means the deck stays light, and the real weapon after a discovery call is the custom follow-up: their numbers, their property, a start date.NEW Challenge new prospects; never provoke an existing customer
The only real randomized experiments in the field (vendor-funded, so labeled): leading with an unrecognized problem beat conventional pitches for new business — but the same provocative framing backfired on renewals (+10% switching risk) and price increases. Relevant beyond the deck: it's a rule for Sean's outreach vs. how we talk to Camellia.TRAPS New entries for the never-use list
- The circulated "Zuora sales deck" is a recreation Zuora assembled 12 days after Raskin's article; the deck he actually analyzed was never published. The article's outcome evidence is one anonymous anecdote ("Tim… signed the largest deal in his company's history").
- Drift's "greatest sales pitch" deck is a conference talk, not a sales deck — 72 slides, ends "Thanks.", no ask, no pricing.
- "Every closed deal used a deck under 9 minutes" — admitted fiction; "slides cut win rates by X%" — no X was ever published.
- "53% of customer loyalty comes from the sales experience" — untraceable to any primary source.
- Any "Square pitch deck" — no credible Square deck exists anywhere; all circulating copies are reconstructions.
The recommendations pass: hold
Verdict on the discovery deck: hold, with one edit made. The corpus confirms or is neutral on every structural choice we shipped — email set capped at seven, assertions on the argument slides, pricing present and simple, team second-to-last with founder access sold as a term, explicit pilot ask last, the send/live separator, objection-driven internals. The one edit from this pass: the run-of-show now caps live deck time at under 10 minutes. Still on offer, unchanged: the assertion variant of "Four ways we're different." — the data mildly supports it but doesn't demand it.
What was actually examined
| Corpus | What | Provenance |
|---|---|---|
| Early-stage investor decks (10) | LinkedIn '04 (with Hoffman's own slide-by-slide annotations), Moz '11 (a round that failed), Mixpanel '14, Front '16, Dropbox '07, Airbnb '08, UberCab '08, Buffer '11, Intercom '11, Coinbase '12 | Founder-published originals; slide images read directly for 9 of 10 |
| Real sales decks (13) | Zuora (both first-party copies, all 24 slides transcribed), Linear, folk, Salesforce, Splunk (real price table), Google Business Messages, HubSpot template, Asana, ChartMogul, Zenefits, ClickUp, + 2 flagged misattributions | PDFs and slide images read directly |
| Practitioner ground truth | Dunford's real before/after client slides (Help Scout), her 8-step structure from her own spoken words; Kazanjy's published sales slides and playbook; founder accounts (Vanta, Gusto, Retool, Gong, Jen Abel) | First-party talks, books, posts — verbatim quotes with URLs |
| Telemetry | Gong: 803k meetings (slides in discovery), 121k meetings (deck length won/lost) | Vendor data — engagement, not causation |
Excluded with reasons: Square (no deck exists), the "original" Zuora deck (never published), Canva's four decks (blocked to non-browser clients), the Drift talk (not a sales deck), every AI-generated statistics farm encountered. Raw slide images and transcripts are archived in the session workspace for re-checking.
Honest limits
- Survivorship cuts both ways: famous decks are famous because the companies won. The corpus includes one genuine failure (Moz's round died in diligence with a 40%-assertion, lean deck) — deck quality and outcome are loosely coupled at best, and the founders say so themselves.
- Investor decks ≠ sales decks: scored separately throughout; most public "famous decks" raised money rather than sold product.
- All telemetry is vendor telemetry (Gong, Storydoc): real behavior, but engagement ≠ closed deals, and nobody's data survived independent review.
Compiled 2026-08-10 from two provenance-first mining passes over the corpus named above. Companion pages: the research brief and the discovery deck it governs.