01
Choose a path
Direct Flight if you already know your test markets. Market Select if you want the power analysis done for you.
Measurement, without the guesswork
Know what your media actually drove.
Incrementality IQ runs every geo-experiment through three independent causal models — GeoTwin™, CausalCore™, and DiffLens™ — then reconciles them into one boardroom-ready verdict. No black boxes. No cherry-picking. No waiting a quarter.
Causal, not correlational3-model consensusResults in minutes
Always-on·Reproducible run hashes·Your data stays in your workspace
The workflow.
Six steps, ~40 minutes end-to-end.
From raw panel data to a board-ready decision in one afternoon — not one quarter.
02
Connect KPI data
Historical daily geo KPI from your warehouse, Meta, GA4, Shopify or Snowflake. Schema validated on upload.
03
Select markets
Monte Carlo across 58K+ pair permutations. Ranked by power, pre-period fit, and holdout cost.
04
Run analysis
Three causal models execute in parallel. Live engine log, reproducible run hash, under 3 minutes to verdict.
05
Consensus verdict
Direction agreement, CI overlap, estimate agreement, and a placebo bias check reconciled into one grade.
06
AI interpretation
Analyst-grade narrative: what happened, why we trust it, what to do next. Board-ready in one click.
The correlation era is over.
Platform-reported attribution tells you who saw an ad before a sale. It can't tell you whether the sale needed the ad to happen. Causal experimentation can. We built the infrastructure to run it continuously.
Design phase
Power before you spend.
ML ranks 58K+ test/control market combinations by pre-period predictive match. You launch with statistical power guaranteed — not hoped for.
GeoTwin™
Synthetic-twin counterfactual.
A weighted blend of control DMAs whose historical pattern mirrors your test market within a 0.012 pre-fit L2 error. When the campaign runs, the gap is the lift.
Tri-model
Three models. One verdict.
Every study runs through GeoTwin™ (SCM), CausalCore™ (BSTS), and DiffLens™ (DiD). We report only the range all three agree on. No cherry-picking.
Placebo
Self-validating.
Every model also runs on 22 sham windows where no lift should exist. If residual lift appears there, the calibration is off — and we flag it before you trust the result.
Always-on
Continuous, not quarterly.
Most MMM studies run once a quarter. MemoLogs runs experiments continuously — so channel incrementality shifts surface in days, not months.
Reproducible
Signed run hashes.
Every verdict ships with an immutable run hash, seed, donor weights, and data snapshot. Re-run tomorrow, you get the same answer. Auditors love it.
The tri-model consensus engine.
Each model answers the same question a different way. When they agree, you can act. When they don't, we surface why — and what to change before you scale.
SCM
Builds a synthetic twin market from a weighted blend of donor DMAs and measures the gap between the twin and the treatment trajectory.
- Donor DMAs weighted
- 14 / 22
- Pre-fit L2 error
- 0.012
- Placebo p-value
- 0.031
- Lift estimate
- +7.6%
BSTS
Bayesian structural time-series forecasts the counterfactual. Posterior intervals quantify uncertainty with mathematical rigor.
- Posterior draws
- 10,000
- R² (pre-period)
- 0.961
- Tail probability
- 0.028
- Lift estimate
- +7.1%
DiD
Two-way fixed-effects regression — the classic econometric benchmark. Robust, simple, widely trusted by CFOs and auditors alike.
- Parallel-trends test
- Pass
- Clustered SE
- 0.0192
- Coefficient p-value
- 0.048
- Lift estimate
- +6.4%
See the lift. See the decay.
See the counterfactual.
Every consensus verdict ships with the visualizations that prove it — the trajectory your campaign caused, the trajectory your market would have taken without it, and how long the lift persists after the flight ends.
GeoTwin™ · Synthetic control
Observed vs. Counterfactual — daily orders
ObservedCounterfactual twinLift zone
Effect persistence
Effect Decay Curve — how long does the lift last?
Half-life14.0dPeak lift+9.8%Residual @ 21d+3.1%
Analyst-grade interpretation.
One click.
Every verdict is paired with an AI Insights Engine that scores confidence, flags caveats, and produces a boardroom-ready narrative — plus a per-DMA decomposition so you know exactly which markets carried the lift and which underperformed.
Market-level insights
Per-DMA decomposition of the consensus lift · model-adjusted
- +8.4%
Strongest of the four treatment DMAs. Pre-period donor fit is excellent (L2=0.008). Oct 18 holiday slightly inflates local SCM; stripping that day shifts Seattle to +7.9%, still above average.
- +7.2%
On-trend with the consensus. Effect builds steadily through week 2 and holds. Historically low variance and tight correlation with Seattle make Portland's estimate robust to donor-pool perturbations.
- +6.9%
Slightly below group mean but well within the CI. Denver's counterfactual weights Phoenix, Austin and Minneapolis heaviest. Lift is stable across all three models — no methodology-specific artifact.
- +5.8%
Weakest donor-pool fit of the treatment group — event-driven tourism spikes reduce pre-period correlation. Still positive and directionally consistent, but swapping in Salt Lake City next flight will tighten control.
From the verdict to the next dollar.
Measurement isn't the end of the workflow. Every trusted lift result anchors the Budget Planner — a causal media-mix model whose backbone is your own live geo-experiments, not a periodic calibration — so you can plan next quarter from what you measured, not what a model inferred from history.
Allocate cross-channel
Split the next budget to maximize measured incremental outcome — bounded to the spend range your experiments actually validated, never extrapolated.
Forecast the reallocation
See the projected lift of moving spend from current to recommended, with credible intervals — not a single point guess.
Gated to what it can defend
Every plan carries a trust state. When the model isn’t trustworthy, the planner shows your measured test history instead of a confident number.
Ask it in the Co-Pilot, or open the Budget Planner — every recommendation carries its trust state.
“We had been spending on branded search for years because the platform said it was our best channel. MemoLogs proved it was 11× inflated. That one insight paid back the whole contract.”
VP Growth · Consumer fintech
See it on your account.
Scoping call · 30 min · free · zero obligation.