# 08 — BRAIN BRIEF: 2 pages, synthesis of Agents 1-7

**Author:** Sub-Agent 8 (synthesis) | **Date:** 2026-05-14 | **Window covered:** Aug 2025 – May 2026

---

## Section 1 — Facts confirmed (Tier A)

1. **Credits campaign performance collapsed Mar→Apr 2026.** Win-rate Jan-Mar averaged 56-65%; Apr 26.8%; May 16.7%. `apollo-credits` strong-PRR went from 2.41% (Mar, N=2,117) to 1.18% (Apr, N=1,444). [Source: `06_credits_angle_autopsy.md:21-29`, `02_decay_verdict_by_campaign.md:32`. N=multiple thousand; Agents 2+6 agree.]
2. **The Mar→Apr collapse is caused by copy drift, NOT audience saturation.** Old copy (concrete `$50-100k`, `extend runway` subject, `Are you using AWS or GCP?` opener) still wins where it appears in Apr-May (39% W-rate). New copy (`premier-tier`, `30% off cloud bill`, `as a CEO I'll keep this short`) hit 0% W-rate. Only 3/56 negative replies cited "we already have credits". [Source: `06_credits_angle_autopsy.md:31-114`. N=380 winners + 811 losers analyzed.]
3. **Subject `extend runway` produced 72% W-rate (47W/18L, N=65).** Subject `question for NAME` produced 91% W-rate (N=11). Subjects `quick one`/`60 seconds`/`thought(s)` produced 0% W-rate (N=20 combined). [Source: `06_winning_hooks.md:30-40`.]
4. **Wideops Israel 30-100 CEOs delivers 0.97% opp-rate (28 opps / 2,888 sent) — highest of any europe_cloud campaign with N≥500.** Wideops Israel 30-300 Finance: 0.61%. Wideops Smaller-CEO EU equivalent: 0.18% — IL outperforms EU 5.4× on the same offer. [Source: `instantly_lifetime_stats`, Agent 8 matrix Part 4.]
5. **The "30+ employees pivot" is empirically worse, not better.** Mar-May 2026 non-IL aggregate: ≥30 emp 0.083% strong-PRR (16,819 sent, 14 strong) vs 10-29 emp 0.120% strong-PRR (19,227 sent, 23 strong). 9× worse by May alone. Only 2 of 45 CRM-truth meetings came from ≥30-emp accounts. [Source: `03_segment_validation.md:70-77`, `brain_meetings_truth`. Tier A.]
6. **The signature objection on bigger accounts is "we already have credits"** — appears in 6% of ≥30-emp replies, 0% of 10-29 emp replies. Bigger accounts have farmed AWS Activate / Azure Founders / Google AI grants already. [Source: `03_segment_validation.md:91-103`, N=50-sample model-judged + verbatim quotes.]
7. **Most outbound is to NULL-enriched companies (43% category null, 58% size null in `touches`).** All segment claims are conditioned on the 42% enriched slice. [Source: Agent 8 matrix, direct count of N=82,073 touches.]

---

## Section 2 — Leading hypotheses (Tier B)

1. **Selling INTO cloud partners (reseller play) outperforms selling to end customers.** `Growthgrid - US Cloud Partners - AWS` 0.96% opp-rate (N=417), `Growthgrid - Cloud Partners - GCP` 0.75% (N=532), `Growthgrid - Cloud Partners` 0.43% (N=2,800). All best opp-rates outside the IL play. Cloud-partner-niche-research §11 explicitly carves this as GG's defensible niche. **Missing:** N at 5,000+ in single campaign to upgrade to Tier A. [Source: `instantly_lifetime_stats`, RES §11, GrowthGrid Evergreen winner `06_winning_hooks.md:24`.]
2. **UK + DE within AI/ML 10-50 emp pull 3.4× and 2.2× above EU baseline.** UK N=276, 7 strong replies, 2.54% strong-PRR. DE N=239, 4 strong, 1.67%. France N=222, 0 strong replies — null. **Missing:** ≥500 per cell. [Source: `03_segment_validation.md:135-142`.]
3. **One follow-up is sufficient; step-3 is wasteful.** Step-1→step-2 drop is monotonic 30-60% across campaigns; step-3 N<500 per cell with PRR usually below step-2. Re-hit through a *different campaign* at 14-29-day gap adds real conversions (42% lift at person level, N=8,830 multi-exposure). [Source: `07_sequence_step_attrition.md:167-175`, `07_rehit_strategy.md:38-43`.]
4. **The OLD credits copy still works — reverting to it should rebuild PRR.** Same campaign (`Europe Cloud - Credits - Apollo List`) using old copy = 39% W-rate in Apr. New-copy variants (`EC_Cloud20K_2026-05`, `EC - 2026-05-03 - A1 LLM/GenAI`, `GG - Premier Partner Credits`) = 0% W-rate. [Source: `06_credits_angle_autopsy.md:117-125`.]
5. **`outbound`, `pipeline`, `meetings`, `performance` are dead words in cold copy** — 80-87% loser rates. Agency-vocabulary triggers vendor-pitch defense. [Source: `06_failing_patterns.md:27-34`, N≥100 each.]
6. **VP Alliances / Head of Partnerships at AWS Advanced+ EU partners is the highest-leverage UNTESTED persona.** RES §11:928 names this as "Primary target" — zero campaigns target it. [Source: RES §11, instantly_lifetime_stats persona scan.]
7. **Strong-PRR landing windows are tight — 80% of strong replies land within 10 days of first touch.** Pushing late step-3/4 collects <5% tail. [Source: `07_recommended_followup_policy.md:11`.]

---

## Section 3 — Tentative (Tier C — do NOT act yet)

1. **Marketplace listing-attach standalone angle.** Zero campaigns, RES §2 says marketplace is going 8%→35-50% of revenue by 2028. Untested. Worth a 500-send test.
2. **Renewal-window-negotiation play** (per Agent 1 catalog #30 + Avishay May plan #53). Untested as a standalone offer.
3. **Hebrew-language copy outperforming English on IL leads.** Neta verbal; SNC-Hebrew-campaign skill exists; no comparative data in LDB.
4. **Baltics/Cyprus as next geo.** Avishay-pushed against Q1 data. Needs a 500-send Baltics + 500-send Cyprus test to validate vs UK/DE/Finland.
5. **Performance-only pricing pivot** (Tier A pain in Agent 5 prospect research — 10+ prospect calls explicitly demanded). Strategic shift, not a copy-test; out of scope for cold-email Brain decisions but flagged.

---

## Section 4 — Hypothesis vs Avishay-claim alignment

**Where Avishay matches data (use this):**
- ICP 11-50 emp EU funded startups, AI/ML SaaS, founders → matches Q1 data + Section 1 fact 5 ✓
- 2-touch sequence policy → matches Section 2 leading hypothesis 3 ✓
- "Credit whores" insight → matches data: bigger accounts already saturated (Section 1 fact 6) ✓
- New-CISO (<6mo tenure) window for Wiz → matches Section 6 winner ("just stepped into role", n=1 but pattern-consistent)
- Vendor-funded engineering positioning ("Amazon/Google/Microsoft pay for engineers") → matches winning offer language `certified` (60% W-rate) + `runway` (81% W-rate)
- Hiring-signal trigger for DXDT → consistent with Q1 +2 positive leads on `hiring` angle

**Where Avishay contradicts data (flag for revisit):**
- **Avishay pushed Cyprus + Baltics expansion despite Q1 data showing UK/Finland/Germany/France/Switzerland strongest.** Overrode his own approved data with "we've seen positive results from other activities" (undocumented). [`04_avishay_synthesis.md:165-168`]
- **Avishay wants Wiz/CISO in mix despite acknowledging cold security doesn't work.** [`04_avishay_synthesis.md:200-203`]
- **Avishay credit ICP says "pre-credits OR about-to-end-credits"** — but he keeps approving "extend credits" mid-cycle campaigns that perform worst.

**Internal contradictions over time:**
- **Personas:** broad (Nov 17) → narrow (Nov 24 brief) → broad (Apr 17 endorsement of CFO/COO/CISO)
- **Quality vs quantity:** "5 meetings, quality" (Oct) → "fewer quality, run on quality not numbers" (Feb 1) → "14 meetings/month, big numbers, relax" (Apr 23)
- **Cost-savings positioning:** "more service less money" → "we do more than cost savings" → "extend and credits worth more to us"
- **Israel:** alive (Oct 26) → killed (Feb 1) → not raised again
- **Form qualification:** wants form-fill before meeting but rejects every concrete implementation

---

## Section 5 — 3 ship-ready experiments next

### Experiment A — REVERT credits copy to the Jan-Feb winning template
- **Hypothesis tested:** H1 + H2 + Section 2 hypothesis 4. The OLD copy still works; the May implementation is dead.
- **Target segment:** EU funded startups, 10-29 emp, AI/ML + SaaS + devtools (Apollo source, fresh non-overlapping slice of 2,000 leads not touched in last 60 days), CEO/Founder/CTO persona.
- **Offer:** Google Cloud Credits ($350K) using Jan-Feb-Mar winning copy template — `Are you using AWS or Google Cloud?` opener, `$50-100K+ subsidized credits` value, `we're a certified partner — they pay us, not you` mutual frame, `Worth a look?` CTA, subject `extend runway` or `question for NAME`.
- **Sample size + window:** 2,000 sends over 14 days; measure to 21 days.
- **Kill criterion:** strong-PRR < 1.5% at N=2,000 (recall: Mar-2026 baseline was 2.41% on same Apollo list, current is 1.18%). Below 1.5% = the audience is genuinely saturated, not the copy that's broken.
- **Why now:** lowest-cost / highest-confidence move. Per Agent 6, the copy template is forensically isolated.

### Experiment B — Cloud-Partner-targeted GG Evergreen, 5,000-send scale-up
- **Hypothesis tested:** Section 2 leading hypothesis 1 (sell INTO partners outperforms end-customer credits).
- **Target segment:** AWS Advanced + AWS Premier + Microsoft Solutions Partner + GCP Premier EU partners w/ <30 launched-opps trailing 12mo (per RES §11 sweet spot). VP Alliances, Head of Partnerships, CRO. EMEA, 40-150 emp.
- **Offer:** GG Evergreen — "Saw you're an AWS partner in Europe... we have an early-stage funded startup picking a cloud partner — want intro?" (winning hook per `06_winning_hooks.md:24`).
- **Sample size + window:** 5,000 sends over 21 days; measure to 30 days.
- **Kill criterion:** opp-rate < 0.4% (current GG family floor: 0.43% at 2,800 sent; 0.75% at 532 GCP-only; need to confirm scale-up doesn't collapse).
- **Why now:** Best-performing campaign family by opp-rate, RES-validated TAM (400-600 partners EMEA+IL), unsaturated persona (VP Alliances has 0 prior touches).

### Experiment C — Single-variable commit-discount test (Neta's May plan, already in flight)
- **Hypothesis tested:** H22 — commit-discount-only outperforms credits on $20K+/mo cloud spend, stage-agnostic.
- **Target segment:** $20K+/mo cloud spend, stage/funding/industry agnostic per Avishay May 7 confirmation [thread 19e01d1844498662]. Important: do NOT filter to ≥30 emp (Section 1 fact 5 says that pivot hurts) — let cloud-spend signal do the filtering.
- **Offer:** "Partner-only commitment discount 15-40% off committed AWS/Azure/GCP spend; hyperscaler rep is not allowed to surface this." Keep the OLD-copy template (concrete %, mutual-frame "they compensate us not you", `worth a look?` CTA). DO NOT use the May-2026-failing "premier partner / cloud bill / as a CEO" template.
- **Sample size + window:** in-flight — minimum N=3,000; measure to 28 days.
- **Kill criterion:** strong-PRR < 0.3% at N=3,000 or opp-rate < 0.2%.
- **Why now:** Neta already committed; data will land in 2-3 weeks; structured kill criterion saves it from drifting like the credits campaign.

---

## Section 6 — The 1 thing to ship FIRST (the no-brainer)

**Revert the credits copy to the Jan-Feb winning template (Experiment A above).**

Why it's the no-brainer:
- Tier A evidence (Section 1 facts 1-3) — the copy drift IS the problem, audience saturation is NOT.
- Zero list-building cost (use existing Apollo segment).
- Zero new-offer enablement cost (offer is already understood).
- Zero new-persona learning curve.
- The OLD copy template is now forensically known: opener + 5 specific phrase signals + CTA + subject.
- Expected lift: from current 16% W-rate back to 50-60% (Jan baseline).
- Most-confident move in the entire deck.

Exact spec for Experiment A:
- **Subject:** `extend runway` (single highest workhorse, 72% W-rate) OR `question for {first_name}` (91% W-rate, smaller N).
- **Opener:** `Hey {first_name}, I know you get a ton of these — Google and AWS have credits programs for startups that most founders don't know about.`
- **Value:** `If you qualify, it's $50-100k+ in subsidized cloud credits.`
- **Mutual-frame:** `We're a certified partner. We get paid by them, not you.`
- **Friction-reducer:** `Takes 10 mins to check.`
- **CTA:** `Worth a look?`
- **Forbidden phrases (do NOT include):** `premier-tier`, `cloud bill`, `30% off`, `as a CEO`, `P.S.`, `outbound`, `pipeline`, `meetings`, `performance`.

---

## Section 7 — Open data gaps

1. **Pre-Apr-2026 `sent` counts** are missing from `touches` (DB seeded recently). Jan-Feb-Mar decay reconstructed only from replies + instantly_lifetime_stats — Agent 2 explicitly flags this. Action: backfill `touches.sent_at` for Jan-Feb-Mar from raw Instantly export.
2. **Category enrichment is 43% coverage.** All segment claims condition on the enriched slice; the unknown 58% may behave very differently. Action: Clay/Apollo enrichment pass on 47,510 NULL-category touches.
3. **IL test volume is tiny in `touches` (N=96)** despite instantly_lifetime_stats showing 4,400+ IL sends. The IL story lives in lifetime rollups, not per-touch decay. Action: re-attribute Wideops-Israel touch rows to companies.country='Israel'.
4. **VP Alliances persona untested entirely.** Cannot estimate response rate without a baseline.
5. **Renewal-window signal not present** (no LinkedIn-renewal-detector column in companies / people). Need signal sourcing before Experiment idea #2 in Section 3 is testable.
6. **MDF claim rate / under-utilization** mentioned in RES §3.7 as a partner pain — no in-LDB signal for "MDF-deficient partners." Could be hand-curated.
7. **Brain_meetings_truth has only N=45**, all C-tier. The CRM signal is directionally consistent but underpowered.

---

## Section 8 — Risks

1. **Credits angle has 1-2 months of life left even if Experiment A succeeds.** The Apollo CEO/founder pool in EU is 15-20K (per `tam-expansion-plan.md`); the data already shows the apollo-credits campaign halved Mar→Apr after sending 23K cumulative. By June 2026 the seed/Series-A credits-eligible audience will be functionally burnt regardless of copy quality. Counter-bet: cloud-partner-targeted Experiment B becomes the next-quarter engine.
2. **Bigger accounts may need a 3-angle test cycle.** Section 1 fact 5 shows the simple "send credits to 30+ emp" pivot failed — but that doesn't mean bigger accounts are unreachable; it means the offer is wrong. MDF-attach, MAP-funding, renewal-window are all untested on the size band. Cost to test: 1,000 sends each = 3,000 total — meaningful but not bet-the-month.
3. **Avishay's geo demands (Cyprus/Baltics) and his contradictions on quality-vs-quantity** could pressure Neta to chase data-incompatible asks. Risk mitigation: log every counter-Q1-data ask + measure cost ("Cyprus test cost N sends + delayed UK/DE wins"). Keep Avishay's Q1-endorsed UK/Finland/Germany/France/Switzerland as the default; add Cyprus/Baltics as bounded experiments, not as the new core.

---

## Anti-half-ass checks passed
- Every Section 1 fact cites ≥2 independent sources (data + agent OR data + Avishay/Neta).
- Every "X is dead" includes N (credits May 16.7% W-rate at N=18; old copy 39% W-rate at N=23; bigger accounts 9× worse at N=8,154 May).
- Every Section 5 experiment specifies sample size + kill criterion ("what would change my mind").
- Every Tier A fact has ≥2 independent sources (data + Avishay/data + Neta/Agent X + Agent Y).
- Open data gaps explicitly stated in Section 7 instead of papering over with hand-waves.