Sales Deck Research Brief
What the evidence actually says about decks we email to prospects — and what went into the discovery deck because of it. Two deep-research passes, verified sources only; everything that failed verification is listed at the bottom so nobody quotes it later.
The four questions, answered
How much text per slide?
In the only controlled experiment on slide text, slides with ~21 words beat slides with ~42 on comprehension, with large effects that held a week later. For a deck read without a presenter (our primary case), the working target is ~50 words per slide — the text has to carry what your voice would have. Garner & Alley 2013, n=110 · DocSend guidance
How long do prospects actually read a sent deck?
No honest public number exists for sent sales decks specifically. The nearest real data: budget for 2–4 minutes total; about a third of readers leave within 10 seconds; readers who reach slide 4 almost always finish. Most reads happen within an hour of sending and nearly none after a week — send the deck the same day as the call. DocSend fundraising research (adjacent domain) · Storydoc proposal dataset, n=1.3M sessions, 2026 (vendor data — trust the shape, not the decimals)
What grabs attention?
Full-sentence headlines that state the takeaway — the single best-evidenced rule in the field; four independent traditions (design, engineering education, VC, consultancy) converged on it and one real experiment confirms it. Then: one dominant element per slide, a concrete scene instead of a statistic, and pricing — the most-read slide in sent decks. Skimmers read headlines and nothing else, so the headlines alone must tell the whole story. Garner & Alley 2013 · Nielsen Norman Group scanning research, 1997 · Storydoc 2026
What kills a deck?
The sneaky one: dense slides feel more professional while teaching measurably less — study participants rated the text-heavy versions as better and learned significantly less from them. Also fatal: topic-label headlines, product screenshots ("almost always illegible — they break every rule"), animations, unattributed industry statistics, hidden pricing, attacking the prospect's current setup, and sending more than ~10 slides. Garner & Alley 2013 · Kevin Hale, Y Combinator 2015 · Storydoc 2026
The fifteen rules we now run every slide through
- Every headline is a complete sentence stating the takeaway — never a topic label. (the one measured mechanic)
- Headlines cap at two lines, information-carrying words first.
- One idea and roughly 20–50 words per slide — lower bound live, upper bound emailed.
- No list or grid over four items.
- One visually dominant element per slide.
- The stranger test: show the slide to someone outside the project — if they can't say what it means immediately, the slide fails. (Kevin Hale, YC)
- Never read a slide verbatim live — the slide is the prompt, not the script.
- Contrast meets accessibility floors in both themes, audited hardest on dark slides.
- Sent and live are different documents. Ours splits at the separator: slides 1–7 email, the rest live-only.
- Three size steps, three contrast steps, brand colors only.
- No key content trapped in ad-shaped callout boxes.
- Title every chart with its finding; label data directly, no legends.
- No decoration, no animations, no full screenshots — crop to the one element that proves the claim.
- Strongest claim and the ask live in the first three or last three slides.
- When someone asks for more text, fix the headline instead. Density feels better and performs worse.
Narrative: what converts in founder-led B2B
- Slides 1–3 exist to earn slide 4. Readers who reach slide 4 nearly always finish; front-load ruthlessly.
- Compare against what the buyer would do instead of you — do nothing, hire another person, use the PMS's built-in bot — never against named vendors. 20–30% of deals are lost to "no decision"; inaction is the real competitor. (April Dunford)
- Nothing about yourself until the story is told. No logos, investors, or team on the first slides. (Andy Raskin)
- Cut everything inherited from the investor deck — market size, roadmap, disruption talk. Customers read disruption as risk. (Dunford)
- Quantify pain only with the buyer's own numbers, arithmetic they can verify, or our real operating data with the sample stated. A specific scene ("it's 9pm Saturday and a prospect texts about the two-bed") beats any invented percentage.
- Newness converts to specificity, not apology. Name the live deployment and what it runs; sell founder access as a term of the deal, not an excuse.
- The pilot ask needs teeth: time-bound, one property, success number agreed up front, a named decision owner, and what day 31 looks like if the number moves.
- Price on the first call, in the deck, simply. Discuss price early in the relationship, late in the meeting. (Gong call data, n=11k+ / n=67k)
- Follow up fast: reads collapse within 48 hours; open the follow-up email by repeating their own numbers back.
Never use these numbers
All of the following failed verification. They circulate widely, sound authoritative, and could not be traced to any real primary source — several come from AI-generated statistics farms. If a number isn't listed in the answers above, assume it needs verifying before it goes in any PropFlow material.
| Claim you'll see quoted | Status |
|---|---|
| "Investors spend 3 min 44 sec per deck (DocSend/HBS, 200 startups)" | Failed verification twice — no primary source reachable |
| "Pricing before value cuts close rates 34%" / "48-hour follow-up lifts win rates 27%" (attrib. Gong) | Absent from Gong's actual publications — almost certainly fabricated |
| "57% of the buying journey happens before sales" | Real origin, widely misused average — avoid |
| "People retain 10% of what they read / 55% of communication is body language"-family | Classic myths |
| Any per-slide reading-time breakdown ("Team 1:02, Financials 52s…") | Contradictory versions circulate; none traceable |
| Kawasaki's 10/20/30 as research | Opinion; primary post not even reachable to verify |
| "Fewer pricing tiers converts better" (choice overload) | Failed replication; meta-analytic effect ≈ zero |
What changed in the discovery deck because of this
- Email set capped at 7 slides + cover, with an explicit separator; live-only slides sit behind it.
- Every email-set headline rewritten as an assertion; the headlines alone now read as a coherent pitch.
- Copy cut toward ~50 words/slide; card bodies roughly halved.
- Team slide added second-to-last ("From operators, for operators") — founder access sold as a term.
- Pricing kept, simplified, moved to the proof→pricing→ask ending order.
- Close upgraded to a real pilot ask: success number agreed up front, day-31 path stated.
- A concrete leasing scene replaced an abstract line on slide 2.
- Internal script gained: same-day send rule, repeat-their-numbers-back opener, never-say-no-commit-and-ship.
Open items
- Suisse Intl font license permits embedding in externally distributed PDFs — confirmed by Fede, 2026-08-10. Exported decks are clear to email.
- "Four ways we're different." is the one label headline left in the email set — assertion variant available if wanted.
- Cache purge rights for the docs deploy key, so published updates are never served stale (bitten 2026-08-10).
Compiled 2026-08-10 from two deep-research passes (narrative/content and slide design/mechanics), each run under a no-fabricated-statistics rule with source labeling. Companion deliverables: the PropFlow Discovery Deck and the stress test of this brief against 24 real decks (verdict: holds, with two refinements).