Aurum Quanta Analytics.
Churn, lifetime value, and next-best-action scores that land in the CRM your team already uses.
Models that decide which customers to retain, which to grow, and which to let go gracefully. Churn prediction, lifetime value, segmentation, propensity scoring. We build on top of the warehouse you already maintain, and avoid standing up a parallel data store that the consultancy ends up owning.
The output goes where your team already works: a ranked list in the CRM, a flag on the account record, a Slack alert when something material changes. A model that lives in a Jupyter notebook nobody opens is a model that doesn't deliver value.
Drop in cohort data. See where the retention story actually lives.
Representative monthly cohorts for a mid-market SaaS product. The retention curve is healthy, and a small lift in the most recent cohorts illustrates what onboarding improvement looks like in this view.
Pick a scenario or paste your own cohorts. The calculator builds a triangular retention heatmap, derives the retention floor, the M1 trend across cohorts, the half-life, and an implied LTV multiplier. Try the mobile preset: the floor, not M1, is what determines whether the unit economics work.
Note · this is a simplified demo
A real engagement would wire this same shape into a live data pipeline. Cohort definitions get nuanced (acquisition channel, plan tier, geo); we'd Bayesian-smooth small cohorts so noise doesn't masquerade as signal; we'd add survival models for time-to-churn; we'd instrument leading indicators that predict retention before the cohort fully matures. The numbers you see here are a snapshot - production analytics is the discipline of making them decision-grade.
Posteriors, not just predictions.
Customer behaviour is noisy and the data is never enough. A Bayesian model carries its uncertainty forward instead of pretending it doesn't exist - so when you act on a number, you also know how much to trust it.
Concrete deliverables.
Churn and LTV models
Customer-level scores updated on a schedule, with per-prediction explanations (SHAP values, top three drivers in plain English) so your team can see why a customer was flagged before they pick up the phone.
Segmentation and clustering
Behavioural segments built to drive campaigns, with one recommended action attached to each segment. The act of writing the recommended action is what stops segmentation from becoming a dashboard nobody actions.
Next-best-action scoring
Ranked actions per customer, fed into the CRM your sales and marketing teams already use. They see one number and one recommendation; the modelling work stays out of their way.
CRM-ready outputs
Lands in Salesforce, HubSpot, Braze, or directly in your data warehouse. We don't ship a separate dashboard for your team to log into. They've got enough of those already.
Significance isn't enough. Effect size has to count.
# analytics/cohort.py: only act on cohorts that are statistically AND meaningfully different
def is_actionable(treatment: pd.DataFrame, control: pd.DataFrame, alpha: float = 0.05) -> bool:
_, p_value = mannwhitneyu(treatment["ltv"], control["ltv"])
if p_value >= alpha:
return False
effect = cliffs_delta(treatment["ltv"], control["ltv"])
return abs(effect) > 0.2 # significant AND meaningfulA p-value gate alone ships noise. Cliff's delta keeps the effect-size threshold honest.
How it would unfold.
Discovery
Data audit, stakeholder interviews, target metric (retention rate, LTV uplift) agreed.
Pilot
Models built with backtests on historical cohorts, lift-tested against a holdout group.
Production
Scheduled pipeline, CRM integration, campaign feedback loop.
Ongoing
Quarterly model review, A/B testing of actions, retraining as behaviour shifts.
Tools we reach for on this kind of work.
Common questions.
If we have a CDP, do we still need this?
A CDP stores customer data; we build the models that score it. The two are complementary. Integrating with the major CDPs (Segment, mParticle, Tealium) is straightforward and doesn't require any platform changes on your side.
Will this break GDPR or the Privacy Act?
Models run inside your cloud against your data and don't transfer anything to a third party. We design with jurisdiction-specific consent and deletion requirements in mind from the first scoping call, because retrofitting them later is painful.
How do we measure if it's working?
Lift testing against a holdout group, on retention uplift, LTV uplift, and campaign ROI. The metrics and the holdout get agreed in writing before any modelling, and measured after the campaign has had time to land.
Let's build it.
A 30-minute discovery call. We'll tell you whether we're the right shop for this.
Book a discovery call →