Lead → MQL
42hrs
+6h QoQwas 36h · target <4h
01 GTM Performance · Q2 Snapshot
Every GTM metric moved the wrong way last quarter
This is not metric drift. It is one architecture failing under load — and the warehouse is not the missing piece.
Win Rate · Ent
16%
−2 pp QoQwas 18% · target 25%
Forecast Accuracy
62%
−6 pp QoQwas 68% · target 85%
Net Retention
98%
−4 pp QoQwas 102% · target 115%
Rep Data Entry
14h/wk
+2h QoQwas 12h · target <5h
The mechanism
77.5% of GTM reporting runs on spreadsheets. That burden reaches reps as 14 hrs/week of manual entry — 35% of the selling week — and the same rushed hands type the stage probabilities the forecast is built from. Hygiene and forecast accuracy are not correlated variables; one is the other's input.
01
Model what is already in Snowflake
CS and Finance already land in the warehouse. Retire the two worst spreadsheet estates (85%, 90%) first — modeling work, not plumbing. Weeks 1–6.
02
Ship only the AI whose data is ready
UC-3 (expansion) and UC-1 (lead enrichment) both score 5/6 on deliverability. UC-2 scores 2/6 and is deferred — its dependency is the silo we are removing.
03
Measure speed, trust, and value
Time-to-insight 4.6 days → ≤1 day. Spreadsheet dependence 77.5% → ≤30%. NRR back above 100% in Q1, ≥104% by end of Q2.
02 Semantic Layer & Data Modernization
Snowflake already exists. The missing layer sits above it.
Two of four GTM functions already land in the warehouse — a modeling and contract problem, not an infrastructure one.
01
No conformed grain
4 systems, 4 definitions of the same metric; one cross-funnel question costs 13.5 days of sequential latency.
02
Spreadsheets are the system of record
77.5% average dependence. No lineage, no tests — every number gets re-litigated.
03
Half the funnel never lands
Marketing and Sales have no warehouse path, so Snowflake cannot answer end to end.
Function scorecard priority = % sheets × time-to-insight
| Function ↕ | Time-to-insight (days) ↕ | % on spreadsheets ↕ | Priority ↓ | In DW |
|---|---|---|---|---|
Customer Success Gainsight · SFDC · Snowflake |
6.0 | 85% | Yes | |
Finance / Ops NetSuite · Stripe · Snowflake |
5.0 | 90% | Yes | |
Marketing HubSpot · Clay · CSV exports |
4.5 | 75% | No | |
Sales Salesforce · local sheets |
3.0 | 60% | No |
Read: the two highest-priority functions are already in Snowflake — the fix is modeling, not migration. Click any column header to re-sort.
Target model — medallion + semantic contract select a layer
Migration sequence 13-week quarter
Daily snapshots are non-negotiable
Forecast accuracy cannot be scored without history to backtest against. Silver carries fact_opportunity_snapshot at daily grain from Wave 1 forward.
03 AI Transformation Roadmap
Two bets whose data is ready. One deliberately deferred.
Deliverability = effort (3 = low … 1 = high) + data readiness (3 = high … 1 = low), scored identically across all three candidate use cases.
| Use case | Target metric (Q2 → goal) | Effort | Data readiness | Score | Call |
|---|---|---|---|---|---|
UC-3 CS Expansion Predictor |
Net retention 98% → 115% | Medium4–6 wks | HighSnowflake consolidated | 5/6 | Build |
UC-1 Lead Enrichment Agent |
Lead → MQL 42h → <4h | Low2–3 wks | MediumHubSpot API cleanup | 5/6 | Build |
UC-2 Deal Risk & Forecasting |
Forecast accuracy 62% → 85% | High8–10 wks | Lowsiloed SFDC & Gong | 2/6 | Defer |
Select a row — or a chip in the matrix — to read the impact, feasibility, and risk assessment for that use case.
Effort × readiness
High
Med
Low
Low
Med
High
↑ Readiness
Effort →
Why UC-2 is deferred, not dropped
Its data dependency — joining siloed SFDC and Gong — is precisely what the semantic layer exists to fix, so building it now means hand-wiring the integration twice and throwing one away. It returns after Wave 3, when the daily snapshot fact becomes its training set.
04 Execution & Metrics
Three KPIs: is the layer fast, is it trusted, and is it worth it?
Each target is paired with an anti-gaming guardrail. Rails show the committed path: baseline → end of Q1 → end of Q2.
01
Time-to-Insight
Median days from request to certified answer, cross-functional.
4.6 daysBaseline
≤ 2.5End of Q1
≤ 1.0 dayQ2 commit
Guardrail
Request volume must not fall — otherwise speed is bought by deflecting the hard questions.
02
Certified-Layer Coverage
Share of exec-reported metrics served from the gold layer — every one under an automated contract test.
77.5%On sheets today
≤ 55%End of Q1
≤ 30%Q2 commit
Guardrail
A metric counts as migrated only when its contract test passes. 100% of certified metrics under test.
03
Funnel Impact of AI
UC-1 lead latency and UC-3 net retention, read from the semantic layer.
Lead → MQL
42hBaseline
≤12hEnd of Q1
≤4hQ2 commit
Net retention
98%Baseline
≥100%End of Q1
≥104%Q2 commit
Guardrail
MQL → SQL conversion must hold — speed cannot come from lowering the qualification bar.
What I am not committing to
115% NRR and 85% forecast accuracy are destinations, not two-quarter promises. Forecast accuracy is deliberately absent — UC-2 is deferred, so I will not commit to a number I have not built the mechanism to move.
05 Insights & Analytical Methodology
What the data supports, what it does not, and how I would close the gap
The requested segment-level cut depends on deal-grain data that has not yet been provided — so the answerable is separated from the unanswerable.
01
Win rate by segment and lead origin — not computable from today's data
Current reporting carries a single Enterprise win rate (18% → 16%) with no segment or lead-origin dimension and no deal-grain records. Rather than infer a cut the data cannot support, I built the harness and flagged the gap.
Ready to runWin rate = won ÷ (won + lost) at deal grain, cut by segment × lead origin, with Wilson confidence intervals and a minimum cell size of n ≥ 30 before any cut is ranked.
02
Data hygiene → prediction precision — a causal path, not a correlation
Table 2 names the mechanism outright: reps manually input stage probabilities in SFDC. The forecast's primary input is a hand-typed field maintained by the same reps absorbing 14 hrs/week of admin. Directionally, admin rose 2 hrs/week as accuracy fell 6 pp.
Test to runRegress absolute forecast error at deal grain on a per-rep hygiene index — field completeness, stage-update recency, stage-skip count — controlled for segment and deal size. n = 2 quarters is an association, not a correlation.
03
NRR contraction — the pre-renewal risk-detection window
NRR fell 102% → 98%, crossing from net expansion into net contraction. Table 2 gives the cause: churn risk signals identified post-renewal date. Table 1 explains why — CS carries the worst time-to-insight (6.0 days) at 85% spreadsheet dependence.
MechanismDetection latency exceeds the intervention window, so the save motion never fires. Contraction sits on the churn / down-sell side — which is why Wave 1 repairs CS detection before UC-3 adds expansion upside.