Uplift connects to your stack, runs real holdout experiments across every channel, and tells you which growth plays caused retention — and which just took the credit.
Under a mandate you set — goals, budget, brand, risk — Uplift closes the full loop autonomously, and gets sharper with every cycle.
A campaign "drove" 200 conversions last quarter. But 160 of those users would have converted anyway. Uplift holds out a real control group, so you see the 40 you actually caused — and the discount you wasted on the rest.
SDK experiments only see in-product features routed through them. Uplift measures the messy real-world motions that live everywhere else — email, ads, referrals, credits, events — through one incrementality engine, with a confidence tier for each.
Not every channel can be randomized. Uplift picks the strongest method available per channel — and never dresses an estimate up as proof.
You choose who's treated, so the control is clean by construction. Budget-grade causal proof you can ship to a CFO.
Can't randomize per user, so we model the counterfactual across regions and time. Directionally strong, with explicit confidence bands.
Lookalike controls correct for self-selection where randomizing is impossible. The best available for community-driven channels.
You set the goal, the budget ceiling, the brand rules and the risk appetite. Uplift escalates anything outside that envelope before it acts. Sensitive connectors stay self-hosted, tokens never leave your control plane.
# Uplift can act inside this envelope without asking. agent: uplift goal: "lift 30-day trial→paid by ≥5pp" window: 14d budget: monthly_cap_usd: 2000 per_experiment_cap: 200 hard_stop_at: 90% channels: allowed: [email, in_app, credits] ask_first: [social_post, ad_spend] guardrails: max_unsub_rate: 2.0% max_churn_drift: 1.5% min_sample_per_arm: 400 brand: voice: "direct, specific, no hype" banned_terms: ["guaranteed", "revolutionary"] auto_review: true # posts need human OKmandate.yaml · approved 2 days agoscope: company/3482
Your team owns the mandate. Uplift reads it, stays inside it, and asks for permission before crossing the line. Every action is logged against it — fully auditable, fully reversible.
Uplift can send email, post to social, and run real ad spend — gated by a mandate you own. Sensitive connectors stay self-hosted behind your own control plane. Every action is capped, gated, logged and reversible.
Send a win-back email to 680 idle users, hold 170 as control. Estimated cost $0, guardrail unsub < 2%, run for 7 days.
We're onboarding a small group of PLG SaaS teams as design partners. Two-week setup. First incrementality report in days, not quarters.