Strategic Research Brief — 2026.04.30

Lead pipeline 10x — strategic direction.

Companion to the topology surface. Where the topology shows the diagnosis, this brief frames the strategic direction — current stack, build-vs-buy decisions, and the angles that move the number.
The Question— two ways to read "10x"
Reading 1 — what we shouldn't do
10x outbound volume on the FY24 pipe

FY24 cold outbound = 6,508 leads (corrected cohort, SVC excluded), 0.02% closed won, 99.95% never reached opp. Median B2B benchmark is ~2.9% conversion. Cypress's pipe is broken, not under-fed. The autonomous AI SDR vendors (11x, Regie) work for low-ACV / broad ICP — not the enterprise motion that produces Cypress's revenue.

Verdict · 10x volume = 10x drown rate. Don't do this.
Reading 2 — what we should do
10x revenue from the same volume

Achievable through (1) attribution lockdown to reverse the 2,090 strict LEADSOURCE overwrites (1,388 External System → Outreach dominant), (2) Snowflake Cortex scoring routing fewer-but-better leads, (3) PQL/inbound capture at 15–30% conversion vs MQL's 2–5%, (4) narrow signal-triggered outbound — not spray-and-pray. Maps to Tim's agentic-pipeline directive almost exactly.

Verdict · This is the play. 60-day plan below.
The Problem— Kelly's diagnostic, ANNA Thread 019ddf01
FY24 Outbound NNL — closed won
0.02%
6,508 leads (SVC excluded) · 99.95% never reached opp
Outbound volume FY24 → FY26
6,508 → 227
FY24 build-out collapsed to FY26 trickle (corrected)
Strict LEADSOURCE overwrites (SCD2)
2,090
1,388 External System → Outreach dominant · surfaced by Kelly's Outreach × SFDC cross-ref
Inbound MQL — closed won (FY24)
0.76%
Strongest converting segment but lower than prior estimates
What we already have— true stack as of 2026-04-30 standup
Layer Tool (in stack) Status What it does today / what changes
CRM Salesforce Core System of record for Lead/Contact/Account/Opportunity. Default Lead Assignment Rule is INACTIVE — assignment runs through Kubaru / ChiliPiper today.
Marketing / forms HubSpot Deprecating Forms, chat, MQL scoring, marketing automation. Decision: HubSpot is being retired. Chat is already off — Cypress GPT replaced it but creates zero lead records, so chat-originated intent (including explicit "have support reach out" messages) is being dropped on the floor right now. Active lead loss in production — fix is owned but not yet shipped. MQL scoring moves into Snowflake. Form capture: separate open question, see next row.
Form capture (replacement) — TBD — Open Net-new decision triggered by HubSpot retirement. Today's web forms post into HubSpot, which writes leads into SFDC. After HubSpot is off, this path needs a destination. Don't assume the answer. Evaluate: (1) SFDC Web-to-Lead — simplest, limited validation; (2) custom form → API Gateway → Lambda → Snowflake + SFDC — most flexible, fits the agentic spine; (3) a thinner marketing-automation tool (Customer.io, Loops, Pardot) if email nurturing is still needed. Owner: Kelly (lead-flow plumbing is her domain). Decision lands before any HubSpot form integration is unplugged.
Sales engagement Outreach Core Sequencing, email/call activity, send-event source. Bilateral sync with SFDC at contact level. Reporting on time-to-response.
Routing — forms ChiliPiper Concierge Keep Form-fill routing (e.g. UI Coverage webinar router). Click → routed concierge meeting. Stays as the form-handler.
Routing — accounts/triggers ChiliPiper Distro Expanding Today: Security Reports + paid-plan conversions + Chris's new license. 4/30 decision: Distro absorbs Kubaru — becomes the single trigger-based routing layer. Blocked on a SFDC API field fix that Kenny has already scoped — needs a roadmap slot. Todd unblocks via Kenny; this is not a Kenny-stuck item.
Routing — paid/TSC/DSR Kubaru Deprecating Routes paid accounts, TSC, DSRs today. Decided 4/30: consolidating into ChiliPiper Distro ("rest of kubaru to chili piper, I'd rather have everything in one tool than split"). Migration is a 60-day workstream.
Attribution HockeyStack Keep–Pending In stack. Renewal value contingent on SF parent-child account hierarchy fix. "HockeyStack Setup - Outbound" is 90+ days overdue. Decision needed at renewal — keep or drop based on hierarchy progress.
Warehouse Snowflake Core+ Raw data + analytics. 4/30 decision: Snowflake becomes scoring + routing source of truth — score writes to a SFDC field, ChiliPiper Distro executes. This is the architectural spine of the 60-day plan.
Analytics Hex (ANNA) · Metabase · dbt Core ANNA = semantic analytics layer (primary). Metabase = internal dashboards. dbt = model transformations. Sufficient — no buy decision here.
Compute / LLM platform AWS · Bedrock Core Cypress AWS account hosts production infra. Bedrock gives us managed Claude / Llama / Titan with per-token pricing and no data egress to a vendor. This is the build-platform for any "agentic" workstream — orchestration, personalization, enrichment can all live here instead of being bought.
Context engine (planned) Grove Draft Company-wide context layer Todd designed: Aurora pgvector + Lambda + EventBridge + Bedrock. /v1/context/{query,route,ingest,feedback}. Phase 1 connectors include Salesforce, Snowflake, ANNA, GitHub. If Grove ships, every "buy a black-box agent" line item below becomes a Grove + Bedrock workstream. Spec is drafted, pending review.
Core / Keep — in stack, not at risk Expanding / Keep-Pending — in stack, scope changing Deprecating — decision made to retire
Architectural Decision— locked in the 4/30 Data/GTM standup
Decision · 2026-04-30
Snowflake-as-scorer + ChiliPiper Distro-as-router. Snowflake computes the composite score (fit × intent × behavior), writes to a Salesforce field, ChiliPiper Distro reads that field and executes routing. The shape of this is dictated by a SFDC platform constraint Kelly surfaced 2026-04-20: cohorts cannot be sequenced directly from Snowflake — final cohorts must land in SFDC for sequencing to fire. The standup ratified the architecture; Kelly named the constraint that forces it. Kubaru and HubSpot are both being deprecated — Kubaru's routing responsibilities consolidate into Distro; chat is already on Cypress GPT; MQL scoring moves into Snowflake. Form capture is an open question that needs a decision before HubSpot is fully unplugged — Kelly owns that call (see stack table). This architectural spine is locked; the form-capture replacement is the next decision to make.
Five Strategic Angles— what 2026 best-practice looks like
Angle 01 · Architecture
Agentic pipeline — build it, don't buy it
2026 SOAReal-time signal capture → AI enrichment → composite scoring (fit × intent × behavior) → routing → personalized multi-channel "Plays" → human handoff at conversation depth. The agent does the research and the cold open; humans take the live conversation.
Cypress PathSnowflake (data + score)AWS Lambda (orchestration)Bedrock (LLM)Grove (context, planned)Distro (routing)Outreach (send)
vs. VendorsUnify (Plays)Clay (orchestration)Pocus (PLG signals) — these are productized versions of what AWS+Bedrock+Grove can run natively against our own data.
PilotOne ICP + 3 trigger types (OSS repo activity, PQL event, intent surge). Lambda subscribes to Snowflake row inserts → Bedrock enriches/personalizes → Distro routes → Outreach sends. No new vendor for the orchestration layer.
RiskBuild path depends on Sarah's strategic ownership of the data spine plus engineering capacity from Muaz / Ryan. Grove is still a draft spec. If Grove slips, Phase 1 runs on direct Lambda → Bedrock without the context engine, then back-ports.
Angle 02 · Scoring Layer
Snowflake Cortex — alignment confirmed 4/30
2026 SOACortex AISQL exposes AI_CLASSIFY · AI_AGG · AI_SUMMARIZE at SQL level. Build feature views in Snowflake, score in-warehouse, write non-destructive score field to SFDC, Distro routes from there. 4/30 standup locked this architecture — Snowflake is the scorer.
VendorsCortex AISQL in-stackMadKudu (A/B baseline)Forwrd.ai
PilotCortex prototype on one ICP. Score = fit × intent × behavior with human-readable explanation string. Output writes to cortex_score field on SFDC Lead/Contact. MadKudu A/B baseline only if there is budget — not on the critical path.
RiskAISQL token billing is separate from warehouse credits. Bound prototype to one cohort and instrument credit usage. MadKudu A/B is optional — primary path is in-stack Cortex.
Angle 03 · Outbound that Converts
Signal-triggered, not list-based
2026 SOAWhat's working in 2025–26: narrow ICP × high-quality signals (Bombora topics, OSS activity, G2 intent) × AI personalization × human SDR for live qual. Median B2B = 2.9% conversion; the tail above that is built on signals, not volume.
VendorsBombora (faster onboard)6sense (richer, slower)Regie.ai (Outreach native)11x (low-ACV only)
Pilot1,000 sends to a tight ICP triggered by Bombora surge + product/OSS signal. Regie for personalization. Target: 1–3% reply, meeting-booked → opp-opened conversion measured. Revenue lag 60–90 days; outbound metrics are leading indicators, not the result.
RiskAutonomous AI SDRs underperform on enterprise complex deals. Use them on mid-market only; humans handle enterprise.
Angle 04 · Attribution Fix
Stop the LEADSOURCE bleed
2026 SOALock canonical lead_source_raw at creation, never overwrite. Compute attribution centrally in Snowflake (already the warehouse + scoring source), export derived fields back to CRM. Reverses the LEADSOURCE-overwrite mechanism (2,090 strict overwrites detected) for new leads.
VendorsHockeyStack in-stack, renewal pendingDreamdata (BigQuery-native)Bizible (SFDC-native)
PilotWeek 1: SFDC validation rules + lead_source_raw lock (no new vendor needed). Week 2: Outreach send events → Snowflake. Week 3–4: revive HockeyStack attribution dashboards or build the same logic directly in Snowflake/dbt if HockeyStack renewal is dropped. Target: Attribution-Lost from 44% → <20% on new leads.
RiskHockeyStack renewal hinges on the SF parent-child account hierarchy fix. If hierarchy slips again, reconsider whether HockeyStack still earns its keep — Snowflake-native attribution is a real fallback now that scoring lives there.
Angle 05 · Inbound 10x · The Real Lever
PQL capture + programmatic SEO + GenAI discovery
2026 SOAPQLs convert at 15–30% in dev tools vs MQL 2–5%. Programmatic SEO can deliver 300–500% traffic gains within six months at low ongoing cost. GenAI/LLM discovery (structured content, schema markup) is the new SEO. Sub-1hr contact rule on PQL events doubles conversion. Inbound MQL is already Cypress's only working segment — this is where 10x is actually achievable.
VendorsDataForSEO (API)Semrush (research)Common Room (OSS signals)
Pilot50 programmatic landing pages for one integration/use case. PQL webhook on trial API key + repo star + first test run. Sub-1hr contact rule. OSS signal pipeline (GitHub events → Snowflake → Cortex score).
RiskThin programmatic content gets penalized. Quality threshold is non-negotiable. Treating all inbound as MQL — PQL signals materially outperform.
Build vs Buy— the agentic spine Cypress can actually run
Reframe · agentic ≠ vendor purchase
Most of the "agentic" SaaS recommendations (Clay, Regie, Common Room, Unify, Pocus) are productized orchestration on top of LLMs + a vector store + an event bus. Cypress already owns that platform: AWS + Bedrock + Snowflake + Grove (planned). The build-vs-buy decision per layer is below.
Buy when: data is licensed and we can't reproduce it (Bombora topics, 6sense intent graph, DataForSEO SERP corpus).
Build when: the value is orchestration / LLM-on-our-data / scoring / routing — workloads that run cleanly on Lambda + Bedrock + Snowflake.
Defer when: Grove dependency is on the critical path and Grove is still draft. Build a thin v1 on direct Lambda → Bedrock; back-port to Grove when the context engine ships.
Buy-Decision Matrix— gaps to fill on top of the existing stack
Stack table above shows what we already own. This table covers net-new evaluations only. Each row also lists the build alternative on AWS+Bedrock+Grove.
Layer Buy option Build alternative on AWS+Bedrock+Grove Verdict Buy spend
Scoring (A/B) MadKudu
Forwrd.ai		 Already the build path. Snowflake Cortex AISQL = scorer. MadKudu A/B is optional validation, not a replacement. Skip unless budget Cortex covers the spine. $15–40k/yr
Intent (third-party data) Bombora PICK
6sense		 Not buildable. Bombora and 6sense are licensed cooperative data sets — Cypress can't generate this signal in-house at any price. Pure buy. Buy Bombora for the pilot. $25–80k/yr
Outbound personalization Regie.ai
11x.ai		 Buildable. Bedrock Claude/Titan reads ICP context from Grove + signal payload, writes personalized cold-open into Outreach via API. Outreach send-events pipe back to Snowflake for closed-loop learning. ~2-week build. Build Bedrock + Outreach API. $15–60k/yravoided
Orchestration Clay Buildable, already half-done. EventBridge + Lambda + Snowflake = Clay's core function. Add Bedrock for the LLM-enrichment cells. The 4/30 architecture (Snowflake → SFDC field → Distro) already obsoletes most of Clay's role. Build Lambda + Bedrock. $3–15k/yravoided
Programmatic SEO (data) DataForSEO API PICK
Semrush		 Not buildable. DataForSEO sells the SERP corpus + keyword index — that's licensed data, same logic as Bombora. The page-generation layer on top can be Bedrock-powered. Buy data, Build page-gen. $0.5–2k/mo
Community / OSS signals Common Room Buildable. GitHub Webhooks + Discord API → Lambda → Snowflake. Grove ingests as a connector (already in Phase 1 plan). No third-party data dependency — these signals are public. Build via Grove connector. $20–50k/yravoided
Net-new buy commitment if we follow the build-where-we-can rule: Bombora ($25–80k/yr) + DataForSEO ($6–24k/yr) ≈ $31–104k/yr. Avoided spend on buildables: $38–125k/yr.

Sources · Tavily research 2026-04-30 — 53 reviewed, 16 cited

[01]unifygtm.com — agentic pipeline architecture
[02]snowflake.com — Cortex for marketing
[03]docs.snowflake.com — AISQL functions
[04]snowflake.com — lead scoring guide
[05]forwrd.ai — Snowflake integration
[06]madkudu.com — scoring patterns
[07]6sense — buyer experience report
[08]11x.ai — outbound at scale
[09]regie.ai — case studies
[10]marketbetter — meta-analysis 2026
[11]hockeystack.com — implementation scope
[12]srpro.marketing — lead source discipline
[13]getpassionfruit — programmatic SEO
[14]daily.dev — PLG for dev tools
[15]pocus.com — LaunchDarkly case
[16]prospeo.io — Bombora vs 6sense
Cypress.io · Chief of Staff · Strategic Research Brief · Stack-Reconciled
Stack source: Data/GTM Standup 2026-04-30 · Re-run: 2026-07-30