This page is a question taxonomy — useful before any methodology choice. If a question is on Rung 2, the right tool is one that answers do-queries. If it's on Rung 3, the right tool answers counterfactuals on a structural model. Tools designed for one rung give wrong answers when used on the other.
For the buyer-facing pitch on the bridge tool that answers these questions, see the Strategy page. For the methodology, see the Methods page.
The procedure
Two steps. The first surfaces the question in the form a stakeholder uses; the second locates it on the ladder so the methodology choice becomes obvious.
Step 1 — Ask the question. State it in plain language, the way a stakeholder would actually phrase it. Don't pre-translate it into causal notation; the phrasing itself carries clues about which rung it belongs on.
Step 2 — Categorize the question. Place it on Pearl's ladder of causation:
- Rung 1 — Associational. "What is?" — seeing. Questions about what the data shows: correlations, conditional probabilities, patterns. Answerable from observational data alone.
- Rung 2 — Interventional. "What if we do?" — doing. Questions about the effect of an action we take. Require a causal model; cannot be answered from observation alone, even with infinite data.
- Rung 3 — Counterfactual. "What would have been?" — imagining. Questions about a world that didn't happen, given what did. Require the strongest assumptions and a structural model.
Scope of work
The procedure above is universal — it applies to any causal question in any domain. What's in scope on this site is narrower: the ten marketing questions below, with a status pill on each indicating where the work has reached. Three statuses:
- Productized Built tooling exists that directly answers this question. The Strategy page describes the deliverable; the bayes model in the repo runs the query.
- Adjacent A related pattern exists — either methodology that transfers to this question through bespoke work (pricing), or a model file in the repo that solves the structurally identical problem in a different domain (retention). Closer than a gap; not yet a product.
- Not yet addressed A real gap. Something would need to be built — typically a model extension, a data integration, or both. The coverage matrix names what.
Rung 2 — the planning questions
The questions that dominate planning conversations: forecasts, tests, budget asks. Each one assumes an action the team is considering and asks what would follow.
"What happens if we do X?" — doing. Questions about the effect of an action we're considering.
-
Spend lift. What's the lift if we increase paid media spend by X%? — the perennial budget-defense question.
Productized — Strategy: Corrected ROIs -
Pricing & promo elasticity. What happens to conversion if we change price or discount depth?
Adjacent — Economics: methodology transfers, bespoke only -
Campaign forecasting. How much incremental revenue will this campaign generate? — pre-launch forecasting for a specific activation.
Not yet addressed -
Channel expansion. What's the impact of launching in a new channel (CTV, retail media, influencer)?
Not yet addressed -
Mix reallocation. How does shifting budget from channel A to channel B change total outcomes?
Productized — Strategy: Sensitivity-analyzed allocation
Rung 3 — the measurement questions
The questions that dominate measurement and attribution conversations: incrementality studies, MMM audits, holdouts. Each one asks what would have happened in a world that didn't happen, given what did.
"What would have happened if…?" — imagining. Questions about a world that didn't happen, given what did.
-
Individual incrementality. Would this customer have converted anyway without seeing the ad? — the core question behind every uplift / geo / holdout test.
Not yet addressed (aggregate-only stance) -
Promo incrementality. What sales would we have gotten last quarter without the promotion? — cannibalization vs. true incremental.
Productized — Strategy: Per-campaign attribution -
Brand-equity attribution. If we hadn't run brand campaign X, where would awareness, search volume, or direct traffic be today?
Not yet addressed -
Retention counterfactual. Would this churned customer have stayed if we'd sent the retention offer? — counterfactual CLV.
Adjacent — BankChurnCounterfactual.bayes pattern in repo -
Retrospective allocation. What would CAC and revenue look like if we'd allocated last year's budget differently?
Productized — Strategy: $50M TV / $20M unmaterialized example
Coverage matrix
The same ten questions, with status, where each is addressed, and what would close the gap on the four that aren't. Status definitions are in Scope of work above.
| Rung | Question | Where it's addressed | Status |
|---|---|---|---|
| Rung 2 | Spend liftLift from a paid-media spend change. | Strategy — Corrected ROIs deliverable; do(TVSpend) / do(SearchSpend) in MarketingMixSCM-AllRungs.bayes. |
Productized |
| Rung 2 | Pricing & promo elasticityEffect of a price or discount-depth change. | Economics — methodology transfers; bespoke engagements only, no productized bridge analog. | Adjacent |
| Rung 2 | Campaign forecastingPre-launch revenue forecast for a specific activation. | Indirect — corrected channel ROIs let you forecast scale-ups, but campaign-level forward mode isn't built. | Not yet addressed |
| Rung 2 | Channel expansionImpact of launching in a new channel. | No Robyn coefficient exists to correct; needs a cold-start prior-elicitation workflow. | Not yet addressed |
| Rung 2 | Mix reallocationShifting budget across channels. | Strategy — Sensitivity-analyzed allocation with the confounder slider. | Productized |
| Rung 3 | Individual incrementalityWould this customer have converted anyway? | Aggregate-only stance. Closest path: hierarchical user-level effects nested under aggregate channel effects, calibrated by geo holdouts. | Not yet addressed |
| Rung 3 | Promo incrementalitySales without the promotion. | Strategy — Per-campaign attribution deliverable; the structural decomposition of observed sales into causal share vs. baseline. | Productized |
| Rung 3 | Brand-equity attributionAwareness, search volume, direct traffic without the brand campaign. | Outcome node is Sales only. Needs brand-equity mediator nodes and long-lag (DBN or slow-state) modeling — pushes off linear-Gaussian closed-form. | Not yet addressed |
| Rung 3 | Retention counterfactualWould the churned customer have stayed? | BankChurnCounterfactual.bayes in the repo — twin-network pattern that transfers structurally; not adapted to a marketing-retention case study. |
Adjacent |
| Rung 3 | Retrospective allocationDifferent past-budget what-ifs. | Strategy — headline numbers: $2.83 vs. $1.84 TV ROI on a $50M budget = ~$20M of revenue that wouldn't have materialized. | Productized |
Score: 4 productized, 2 adjacent, 4 not yet addressed. The four gaps cluster on user-level questions (individual incrementality, individual retention) and outcomes other than weekly sales (brand equity, forward-looking campaigns), consistent with the site's stated scope: aggregate-causal MMM on top of Robyn / Meridian, sales as the outcome.
The practical split
Rung 2 dominates planning conversations — forecasts, tests, budget asks. Rung 3 dominates measurement and attribution conversations — incrementality studies, MMM audits, holdouts. The two rungs use different machinery and the answers are not interchangeable.
Most marketing analytics debates are really arguments about whether a Rung 2 answer is being mistaken for a Rung 3 one. A regression coefficient on past spend is a Rung 1 quantity that a planning meeting reads as Rung 2. A holdout test produces Rung 3 evidence that an attribution dashboard reports as Rung 2. The methodology that survives scrutiny is the one that names the rung first and picks the tool to match.
Going deeper
If the question you're working on is a Rung 2 spend-lift or mix-reallocation question, or a Rung 3 promo-incrementality or retrospective-allocation question, the bridge tool is built around exactly those queries. The Strategy page explains what the bridge produces, what the engagement looks like, and how the structural correction differs from the regression-only ROIs that come out of Robyn or Meridian.
Before reaching for a methodology, name the rung. The right tool is the one matched to the rung, not the one closest to hand.
info@rung3.ai