# 06 — Winning Hook Forensics

**Method.** Pulled 463 positive replies (positive/meeting_booked/soft_positive) and 1,027 negative replies (negative/objection_fit/unsubscribe) from `replies_full`. Extracted the deepest-nested quoted block from each reply (this is the ORIGINAL cold email being quoted back). Successfully extracted from 380 winners + 811 losers. Tagged each on opener/CTA/length/offer-words/personalization dimensions. Aug 2025 → May 2026.

**Top-line:** Aggregate Win-rate = 32%. Anything above is a hook that lifts.

---

## 1. Opener Patterns (ranked by Win-rate)

| Opener type | #W | #L | #Total | WinRate | Lift |
|---|---:|---:|---:|---:|---:|
| **`straight_to_point`** — "I'll keep it short / I'll get straight to the point" | 25 | 18 | 43 | **58.1%** | **1.82x** |
| **`are_you_using_question`** — "Are you using AWS or Google Cloud?" | 20 | 20 | 40 | **50.0%** | **1.57x** |
| **`saw_you_observation`** — "Saw you're an AWS partner in Europe" | 27 | 27 | 54 | **50.0%** | **1.57x** |
| `statement` (default declarative) | 270 | 626 | 896 | 30.1% | 0.94x |
| `open_question` (vague question) | 34 | 112 | 146 | 23.3% | 0.73x |

**Tier-A (≥10 in each side):** `straight_to_point`, `are_you_using_question`, `saw_you_observation` — three clear hook archetypes.

**Live winning examples:**
- `"I'll keep it short - Google and AWS have credits programs for startups..."` [email_id: 019c0506-a61e, campaign: Wideops - Cloud Credits Testing] → meeting booked
- `"Quick question, do you use AWS or Google Cloud? We help funded startups extend runway..."` [email_id: 019bec6c-ad48, campaign: Wideops - Cloud Credits Testing] → meeting booked
- `"Saw you're an AWS partner in Europe. We've built an outbound funnel that landed 15 Partner-of-Record signups..."` [email_id: 019e2293-ce20, campaign: GrowthGrid Evergreen - Cloud Partners 2026-04, 2026-05-13] → positive

## 2. Subject-line patterns

| Subject pattern | #W | #L | #Total | WinRate | Lift |
|---|---:|---:|---:|---:|---:|
| **`extend runway`** | 47 | 18 | 65 | **72.3%** | **2.27x** |
| **`cloud credits`** / `cloud credits for NAME` | 7 | 5 | 12 | **58.3%** | 1.83x |
| `partner` (alone) | 9 | 12 | 21 | 42.9% | 1.34x |
| `potential partnership` | 68 | 119 | 187 | 36.4% | 1.14x |
| `cloud_keyword` (e.g. "AWS Activate Portfolio") | 8 | 37 | 45 | 17.8% | 0.56x |
| `quick one` / `60 seconds` / `quick/brief` | 0 | 10 | 10 | **0.0%** | 0.00x |
| `thought(s)` | 0 | 10 | 10 | **0.0%** | 0.00x |
| `Re: שאלה` (Hebrew generic "question") | 3 | 34 | 37 | 8.1% | 0.25x |
| `question for NAME` (top single subject) | 10 | 1 | 11 | **90.9%** | 2.85x |

**Verified negatives:** `quick one`/`60 seconds`/`thought(s)` subjects produced ZERO positive replies in the dataset. These are dead.

**The `extend runway` subject is the single highest-yield workhorse** across the entire dataset (72% positive rate among repliers).

## 3. Length distribution

- Winners: median **327 chars**, mean 369, p25 291 / p75 381
- Losers: median 334 chars, mean 351, p25 289 / p75 398

Length is NOT a discriminator. Both winners and losers cluster tightly around 290-400 chars. **The 300-400 char cold email is the de facto standard.** Length below 300 has marginal advantage (33.8% vs 30.1% for 300-500) but the gap is small.

## 4. CTA patterns

| CTA type | #W | #L | #Total | WinRate | Lift |
|---|---:|---:|---:|---:|---:|
| **`worth_a_chat`** ("worth a quick chat?" / "worth a look?") | 12 | 13 | 25 | **48.0%** | **1.50x** |
| **`low_friction_ask`** ("send over", "share more", "reply with") | 38 | 57 | 95 | **40.0%** | **1.25x** |
| `interested_relevant` ("relevant?" / "interested?") | 105 | 227 | 332 | 31.6% | 0.99x |
| `open_to_question` ("open to a chat?") | 5 | 22 | 27 | **18.5%** | 0.58x |
| `question_close` (just `?`) | 5 | 38 | 43 | **11.6%** | 0.36x |

Winning CTA archetypes:
- `"Worth a look?"` / `"Worth a quick chat?"` — low commitment, soft framing
- `"Should I check for you?"` / `"Can I share more?"` — implicates ME doing the work, not them

Dead CTAs:
- Bare question marks at end of email (no clear ask)
- `"open to a chat?"` (too vague, sounds like every other vendor)

## 5. Offer-wording lift (which specific words correlate with positives)

| Phrase flag | #W | #L | WinRate | Lift |
|---|---:|---:|---:|---:|
| `runway` mentioned | 34 | 8 | **81.0%** | **2.54x** |
| `certified` | 136 | 91 | **59.9%** | **1.88x** |
| `$50-100k credits` amount specified | 91 | 69 | **56.9%** | **1.78x** |
| `GCP / Google Cloud` | 129 | 113 | 53.3% | 1.67x |
| `premier` partner | 40 | 36 | 52.6% | 1.65x |
| `credits` (general) | 155 | 143 | 52.0% | 1.63x |
| `cloud` | 204 | 226 | 47.4% | 1.49x |
| `AWS` | 246 | 295 | 45.5% | 1.43x |
| `partner` (any context) | 304 | 419 | 42.0% | 1.32x |
| `outbound` | 39 | 169 | **18.8%** | **0.59x** |
| `meetings` / `qualified calls` | 21 | 92 | **18.6%** | 0.58x |
| `performance` / "pay on results" | 5 | 28 | **15.2%** | 0.47x |
| `pipeline` | 10 | 65 | **13.3%** | 0.42x |

**Winning offer language:** runway, certified, specific $ amounts, GCP, premier-tier.
**Losing offer language:** outbound, pipeline, meetings, performance — the vocabulary of agencies pitching agencies. People don't reply to those words positively.

## 6. Personalization signals

| Signal | #W | #L | WinRate | Lift |
|---|---:|---:|---:|---:|
| `saw / noticed` phrase | 33 | 29 | **53.2%** | **1.67x** |
| `linkedin` mention | 57 | 141 | 28.8% | 0.90x |
| `funding / raised / series A-C` | 16 | 51 | **23.9%** | 0.75x |

Counter-intuitive: explicit funding mentions hurt slightly (people resent "I know your business"). The vague observation pattern ("Saw you're a partner / Saw you're hiring") wins — it shows light effort without surveillance vibes.

## 7. Length by month — no drift

| Month | Win median | Loser median |
|---|---:|---:|
| 2026-01 | 298 | 306 |
| 2026-02 | 357 | 362 |
| 2026-03 | 338 | 316 |
| 2026-04 | 360 | 296 |
| 2026-05 | 343 | 340 |

**Length stable across months.** This confirms message DRIFT (different copy) is not a length issue.

## 8. Monthly Win-rate trajectory — the real story

| Month | #W | #L | W% |
|---|---:|---:|---:|
| 2025-08 | 11 | 35 | 23.9% |
| 2025-09 | 33 | 190 | 14.8% |
| 2025-10 | 30 | 70 | 30.0% |
| 2025-11 | 12 | 41 | 22.6% |
| 2025-12 | 33 | 94 | 26.0% |
| **2026-01** | **94** | **133** | **41.4%** |
| **2026-02** | **69** | **97** | **41.6%** |
| **2026-03** | **64** | **65** | **49.6%** |
| 2026-04 | 29 | 65 | 30.9% |
| 2026-05 | 5 | 21 | 19.2% |

Sharp run-up Jan-Mar 2026 (credits angle peak). Steep cliff in April. See `06_credits_angle_autopsy.md` for the why.

## 9. Tier-A pattern catalog (≥10 in winners AND ≥10 in losers — actionable, statistically grounded)

| Pattern | Tier | Hit-rate when present |
|---|---|---:|
| Opener: `straight_to_point` | A | 58% |
| Opener: `are_you_using_question` | A | 50% |
| Opener: `saw_you_observation` | A | 50% |
| Subject: `extend runway` | A | 72% |
| Subject: `question for NAME` | A (single subject) | 91% |
| Offer wording: `certified` | A | 60% |
| Offer wording: `$50-100k credits` | A | 57% |
| CTA: `low_friction_ask` | A | 40% |
| Personalization: `saw/noticed` phrase | A | 53% |

## 10. Counter-list (Tier-A losing patterns)

- Subject `quick one` / `60 seconds` / `thought(s)` → 0% positive
- Offer words `outbound` / `pipeline` / `meetings` / `performance` → <20% positive
- CTA `question_close` (bare `?`) → 12% positive
- CTA `open_to_question` → 18% positive
- Opener `open_question` (vague) → 23% positive