Customer Analytics Checklist

List every system feeding the analytics layer — Salesforce or HubSpot CRM, Segment or RudderStack CDP, GA4 properties, Meta and LinkedIn ad accounts, MAP (Marketo, Pardot, Klaviyo). Note the owner, refresh cadence, and whether the source is the system of record or a downstream copy.

Confirm OneTrust or Cookiebot is gating tags by region — EU traffic should not load Meta Pixel or any analytics tag before opt-in. Verify the CCPA "Do Not Sell or Share" link is wired to suppress downstream sharing. Walk a fresh EU session and a California session through Tag Assistant to prove gating works.

Check the link-builder doc or tool against the last 30 days of GA4 source/medium/campaign values. Drift here breaks attribution downstream — campaigns named differently across email and paid become uncomparable. Update the UTM convention doc and re-share with channel owners.

Walk every key conversion (lead form submit, signup, purchase) end-to-end with GTM Preview and GA4 DebugView open. Common failure: form_submit firing on blur instead of submit, inflating reported conversions. Document any discrepancies before relying on the data for segmentation.

Open a ticket in the engineering queue with the failing event names, the GTM tag IDs, and DebugView screenshots. Block downstream segmentation work on this fix — segmenting on broken conversion data produces broken segments.

Anchor segments to current ICP and revenue goals — firmographics for B2B (industry, headcount, ARR band), behavioral for B2C (RFM, product affinity). Avoid clustering on every available field; pick the 3-5 dimensions the GTM team will actually act on.

Pull a 12-month cohort from the warehouse (Snowflake, BigQuery, Redshift) and run k-means or hierarchical clustering. Test silhouette scores across k=3 through k=8; pick the lowest k that produces actionable, distinct groups.

Hold out 20% of the cohort and confirm cluster assignments are stable. Check that segment-level conversion rate, AOV, or LTV differ meaningfully — if segments don't behave differently, they're not segments worth acting on.

Have PMM draft persona narratives anchored to the cluster centroids — title, top 3 jobs-to-be-done, common objections, preferred channels. Validate with 3-5 sales reps who talk to these accounts daily; their gut-check kills bad personas fast.

Push segment membership to HubSpot or Marketo lists and to Meta / LinkedIn custom audiences via the CDP. Confirm hashed PII transit (SHA-256) and that the ad-platform audience match rate is at least 50% before treating it as ready for paid activation.

Match model to question: logistic regression or gradient boosting for churn / lead scoring, survival models for time-to-churn, regression for LTV prediction. Start with the simplest model that works — interpretability matters when sales and lifecycle teams have to act on the score.

Use a time-based train/test split — never random — to avoid leakage from future events. Watch for features that wouldn't exist at scoring time (e.g., "days_until_renewal" leaks the label for churn models).

Report AUC plus precision and recall at the operating threshold sales will actually use. A churn model with 0.85 AUC but 20% precision at the top decile is useless — sales will stop trusting it after one bad week.

Re-examine feature engineering, class balance, and hyperparameters. Common wins: add product-usage features from the CDP, fix label definition (e.g., what counts as "churned"), correct for seasonality.

Write daily scores into a Salesforce custom field or HubSpot property and into the MAP for nurture branching. Document the score's range and meaning on the field's help text — a 0-100 score with no docs becomes folklore within a month.

Track score distribution and PSI (population stability index) weekly. When PSI exceeds 0.2 or the score distribution shifts noticeably, schedule a retrain. Models silently degrade — this monitor is what catches it.

List every touchpoint per persona — paid ad, organic search, blog, webinar, sales call, onboarding email, support ticket, NPS survey, renewal. Note the owning team for each; orphaned touchpoints (no clear owner) are where customer experience breaks down.

Join CDP event data with CRM activity history at the account or person level. Calculate time-between-touchpoints, drop-off rates, and conversion rates per stage. Reconcile any gaps where events exist in one system but not the other.

Build the map as swim lanes per persona, with quantitative metrics overlaid on each touchpoint (volume, conversion rate, time to next stage). Keep it on one page — multi-page maps don't get used.

Rank pain points by drop-off rate × stage volume to focus on places where intervention moves the most revenue. Cross-reference with support tickets and NPS verbatims for qualitative confirmation.

Write a brief per friction point with the data, the proposed change, and the success metric. Lifecycle owns email/in-app fixes; product owns UX fixes; demand-gen owns paid-funnel fixes. Set a 30-day check-in to measure impact.

Pin the quarter's KPIs to MQL→SQL conversion, pipeline contribution, CAC payback, and segment-level LTV. Resist adding vanity metrics like raw pageviews — every KPI on the dashboard is a KPI someone has to defend.

One page, current-period vs. prior-period vs. plan, with a comment cell next to each KPI for the analyst's interpretation. Self-serve filters by segment and channel. Test the dashboard on a phone — execs read it on phones.

Lead with the headline (what changed, what we're doing about it), then the KPI scorecard, then segment deep-dives, then risks. Avoid 40-slide data dumps — one chart per insight, every chart titled with the conclusion.

30-minute slot with CMO, demand-gen lead, lifecycle lead, sales ops, and PMM. Capture decisions and owner-action items in a single doc; circulate within 24 hours.

Collect what worked and what was confusing — which charts got questioned, which segments stakeholders ignored, which KPI sparked the most debate. Roll changes into the next quarter's run.