Causal Inference
Moves from correlation toward credible causal readouts with explicit identification, assumptions, diagnostics, robustness checks, and decision-ready interpretation.
S-Curve Data is building a team of agents that helps product teams understand performance, predict outcomes, measure impact, and plan what comes next—with shared metric definitions, traceable evidence, and human review.
Use an individual agent for a focused analytical problem. We’re building ADT to coordinate specialists across broader product decisions. Each discipline keeps its own methods, assumptions, diagnostics, and human review.
Moves from correlation toward credible causal readouts with explicit identification, assumptions, diagnostics, robustness checks, and decision-ready interpretation.
Turns historical patterns into forecasts, uncertainty estimates, scenarios, and planning baselines with backtests, diagnostics, statistical models, ML, and TimesFM.
Designs experiments for reliable decision readouts across intake, power, SRM and instrumentation checks, CUPED precision adjustment, guardrails, and decision records.
Builds predictions that can survive deployment with temporal validation, baseline hurdle gates, calibration, prediction ledgers, drift checks, and delayed-outcome monitoring.
General AI can generate analysis quickly. Trustworthy analytical work requires method-specific assumptions, validation, diagnostics, and guardrails.
The Autonomous Data Team (ADT) brings specialists together around shared product data. Analytics and Data Quality help define and validate the metrics; Insights & Memory connects past learning; Planning & Orchestration coordinates the work. Autonomy means coordinating analytical tasks—people review consequential decisions.
Try the Causal Inference, Forecasting and Experimentation portals today. The diagram shows the team architecture we’re building: the integrated ADT pilot is in development, and the full Insights & Memory evidence graph is planned.
Description
Forecasts, scenarios, targets, and assumptions make planning explicit before a decision is made.
Experiments, causal readouts, predictive signals, and diagnostics test whether the claim can be trusted.
Reusable decision packets keep sources, assumptions, uncertainty, caveats, and readouts together.
People review the tradeoffs, choose actions, and assign owners. Outcomes and decision records feed shared memory and the next planning cycle.
One product question can require several disciplines. ADT is designed to bring in the specialists the question needs, keeping evidence and assumptions attached at each handoff.
Analytics investigates funnels, cohorts, and segments. Data Quality checks the metric definition, source freshness, and instrumentation before the team treats the decline as real.
Insights & Memory retrieves relevant deep dives and experiments. Predictive Modeling can identify users at risk; Forecasting can estimate the likely trajectory. Prior hypotheses stay distinct from measured results.
Experimentation helps design an intervention and its guardrails. Causal Inference evaluates impact using an appropriate design, with assumptions and uncertainty visible.
Planning brings the evidence into a recommendation for human review. The decision, observed outcomes, and unresolved questions return to shared memory for the next cycle.
The intended result: a reviewable recommendation with sources, assumptions, uncertainty, an owner, and a follow-up measure.
I’m Sophia Chen, a product data scientist building S-Curve Data. My work spans defining success metrics for Google Maps Platform and helping build the experimentation culture and product data science foundation for QuickBooks Checking at Intuit.
I’m building toward an AI data team that can carry a product question from analysis to a decision—and bring what it learns into the next one. That means specialists with methodological depth, shared evidence, and work that people can inspect and challenge.
If you work in product or data science, I’d love to compare notes on where these agents could help, what you would need to trust them, and which decisions still need your judgment.