One platform. Every measurement method your AI can read.
A deterministic statistical engine runs geo, user-level, and platform incrementality plus marketing mix modeling. Your AI reads the results over MCP and the read-only API. Every model and every interval is shown — nothing blended, nothing hidden.
An AI scientist for incrementality experiments and MMM decisions
Shako Stats is not a generic marketing chatbot. It is a measurement platform built around causal tests, model outputs, diagnostics, and budget decisions, starting with a deep geo incrementality workflow and expanding into the rest of the marketing science stack.
Geo incrementality tests
Available nowDesign matched markets, check power before spend, run multiple counterfactual models, and read lift, iCAC/iROAS, intervals, and model agreement.
User-level experiments
Experiment workspaceBring holdout evidence for CRM, direct mail, lifecycle, and addressable channels into the same incrementality record your team can ask about.
Platform lift evidence
Central evidence layerStore and inspect Meta, Google, TikTok, and other platform lift reads alongside Shako-run tests so the AI sees the broader measurement history.
MMM incrementality
Coming nextMMM workflows will use incrementality evidence, response curves, and budget-planning context so the AI can explain model outputs and tradeoffs as they come online.
The measurement engine
your AI can actually read
Shako Stats runs the measurement workflows and stores the evidence: test designs, power analyses, published lift results, data dictionary entries, metric definitions, diagnostics, model outputs, and budget context. The read-only MCP/API layer makes that evidence available to the AI your team already uses, so every feature becomes part of the same statistics-backed source of truth.
Coming into the broader marketing science stack
Geo incrementality is the flagship workflow today. These modules keep the same measurement philosophy as the platform expands into MMM, forecasting, attribution, user-level tests, and audience science.
And More...
And more coming soon! We have many more features in the works that we are actively developing.
Talk to the statistics, not a black box
Shako Stats runs the marketing science engine. The MCP/API connection gives your AI agent read-only access to the data, instructions, diagnostics, and model outputs, so anyone can ask better questions and see exactly what supports the answer.
Ask in plain language
Your team asks the same questions they would ask a marketing scientist: power, design options, lift, iCAC, uncertainty, and budget implications.
Shako supplies the context
The MCP/API connection gives your AI read-only access to processed data, designs, diagnostics, calculators, and model outputs.
The statistical engine does the math
Shako runs the marketing science workflow: power analysis, matched markets, counterfactual models, intervals, and model agreement checks.
Everyone sees the evidence
The AI explains what worked, what did not, and where the result is uncertain, with the underlying numbers available for review.
Read-only by design
The AI can explain, compare, and draft reports from approved Shako context. Your statistical outputs stay traceable, reviewable, and grounded in the workspace.
Conversation
Design to decision
Can we design a Meta geo test that can detect a $55 iCAC before we spend more?
Yes. The power calculator finds a 6-week matched-market design with 82% power at your $55 iCAC threshold. The faster 4-week option only reaches 61% power.
Show the tradeoffs and pick the most defensible design.
Recommended: 8 test DMAs and 8 controls. It keeps pre-period revenue imbalance under 1.8%, avoids overlapping TV markets, and preserves your weekly budget.
The test finished. What happened, and do the models agree?
Incremental lift is 8.4% and iCAC is $47. Five counterfactual models agree within about a point. This looks decision-grade, with uncertainty shown clearly instead of hidden behind one black-box number.
Evidence
Models shown side by side
Augmented Synthetic Control
Primary production estimator
95% interval: 5.0% to 11.9%
Scaled Bayesian Synthetic Control
Posterior predictive bands
95% interval: 5.1% to 12.4%
Synthetic Diff-in-Differences
Unit and time weights
95% interval: 4.4% to 11.8%
Bayesian Structural Time Series
CausalImpact-style
95% interval: 4.9% to 13.0%
Matched-Market Regression
Fast baseline check
95% interval: 4.6% to 11.6%
Agreement check passed
Multiple models tell the same story, and the uncertainty is visible. If they disagreed, Shako would show that too.