Who's buying, when, and why they keep coming back.

Data Analytics works at the business level — total revenue, what sold, what time of day, which products, which channels. The unit of analysis is the transaction or the period. Customer Insights works at the customer level — who bought, how often, what they're worth over their lifetime, what triggers them to come back, who's about to churn. The unit of analysis is the individual person and their full history with you. Most clients want both, eventually. The bundled engagement saves you ~30% versus running them separately because the underlying data pipelines are shared.

All work happens on data you already collect and own under your existing privacy policy and customer consent. We don't add tracking pixels, don't deploy new cookies, don't share customer data with any third party, and don't train any model on your customers' personal information. For data-subject access and deletion requests (CCPA / GDPR / state laws), the source-of-truth records live in your POS and CRM — your standard process applies, and we mirror any deletion in our warehouse within 24 hours. If you operate in the EU, the analytics warehouse can be hosted in an EU region; for HIPAA-adjacent data we'll discuss a dedicated deployment posture before any work starts.

Almost never. We work with whatever you already have — HubSpot, Salesforce, Pipedrive, ActiveCampaign, Mailchimp, Klaviyo, your POS (Commerce7, WineDirect, Toast, Square, Lightspeed, Shopify), your e-comm platform, your ad accounts. If a system has any export — REST API, scheduled CSV, vendor portal — we can pull from it. The whole point of this work is making the systems you've already invested in tell a unified story; forcing a migration would undo most of that value.

We rebuild the integration for the new system — usually a 3–5 day project scheduled around your cutover. The historical customer-level data stays in the warehouse, so churn models, LTV trends, and cohort curves all survive the switch unbroken. A common gap: migrations from one CRM to another often drop historical engagement data that the new vendor can't import. Talk to us before the cutover and we'll snapshot everything.

Roughly 500 customers across 12+ months of history is the minimum where statistical patterns are reliable enough to drive decisions. Below that the insights start to look more like anecdotes — you can still get some value from segmentation and persona work, but predictive models (churn, LTV) need more signal. The sweet spot is 2,000–50,000 active customers: enough data for confident models, small enough that customer-level intervention is operationally realistic. Above 50,000 the work shifts toward automated trigger-based campaigns instead of per-customer touches.

Both. Dashboards include aggregate views (segment sizes, churn rates, LTV distributions) and individual customer drill-downs — who they are, what they've bought, when they last engaged, predicted next-purchase date, churn-risk score, and the segments they belong to. Your team can search a specific customer by name or email and see their full story. Read-only login per user, no per-seat charge.

Honest answer: it depends on your data quality and the predictability of your customers. We backtest every model against the prior 90 days of actual behavior before it ships and report the real accuracy (precision/recall for churn, MAPE for LTV) in plain English — not vendor marketing numbers. For most SMB clients we see 70–85% precision on churn ("of customers we flagged as high-risk, this share actually lapsed within 60 days") and ±15-25% on LTV forecasts. If a model can't beat the baseline of "assume everyone behaves like the average," we'll tell you and not ship it.

We surface the segment, the trigger, and the recommended message; you run the campaign in your existing email/SMS/loyalty tool (Klaviyo, Mailchimp, HubSpot, your POS's CRM module). We don't replace your marketing stack — we sit on top of it and tell it what to do. If you want us to also run the campaigns, that's a Google Marketing or Lead Generation engagement, often bundled. The split keeps things clean: insights is a strategic layer, execution stays where your customer relationships already live.

One PDF page (or web view), three sections. What moved: the segments that changed meaningfully versus last month and same month last year ("Tier-2 wine club churn jumped from 4% to 7% — concentrated in customers acquired during the 2024 spring flash sale"). Pattern of the month: a deeper read on one cohort or behavior we noticed ("Customers who buy reds-first have 2.3x the 24-month LTV of whites-first customers — implications for tasting-room flow and welcome-series sequencing"). Top moves: 3–5 specific, prioritized actions for the next 30 days, each with the expected impact and the segment to target.

Honest conversation: we'll tell you up front in the audit. The work pays back only when someone acts on it. If you don't have a marketing lead, customer-success person, or owner who can run the recommendations — and aren't willing to hire or contract one — we'll suggest starting with Data Analytics (lower commitment, foundation for later) or a one-time persona project. We'd rather scope a smaller engagement that gets executed than a bigger one that gathers dust.

Yes. Everything we build — the segments, the models, the SQL queries, the dashboard code, the briefing templates — is yours. We document the model specs (features used, training/validation split, accuracy metrics, retrain cadence) so an in-house data person can pick it up. 30-day notice to wind down, no exit fee. The warehouse and Postgres schema come with you in a documented dump.