Review the campaign economics
Supply one row per product–campaign with a credible no-promotion baseline, realized prices, variable costs and allocated campaign effects. Inspect the contribution bridge and identify the assumption that could change the decision.
Design the next useful experiment
Choose the independent assignment unit, primary outcome, worthwhile effect and measurement window. Supply the eligible-unit list, then generate a reproducible allocation and check the approximate sample size or detectable effect.
Analyze outcomes at the assignment level
Choose independent outcomes, matched pairs, two group summaries or unit-level before/after data. Read the interval alongside the design assumptions; statistical precision does not verify randomization or remove confounding.
Choose a feasible next portfolio
Supply candidate offers, expected volumes and shared product inputs. Set campaign budget, stock, capacity and markdown limits. Compare the selected plan with regular-price and greedy benchmarks, and inspect whether optimality was proved or the search stopped with a remaining gap.
Keep the decision and its source assumptions
Export the workspace's own JSON backup and relevant CSV or printable report. Campaign projects, experiment workspaces, planner projects and purchase recovery files serve different purposes; retain the correct file for each workflow.
Three separate input contracts
Campaign review
Use when the question is what a promotion contributed against a supplied no-promotion comparison. The buyer supplies the benchmark and source accounting; the workbench reconciles the result and sensitivity. Offline Personal and Agency support up to 25,000 product–campaign rows; the hosted preview supports 20.
You supply
- Required CSV columns: campaign, product, baseline_units, promo_units, regular_price, promo_price, unit_cost. One unique campaign/product pair per row.
- Optional columns: promo_spend, supplier_funding, period_days, baseline_days, returns, post_promo_dip_units, currency. Allocate fixed spending and funding once; supply both day-window fields or neither.
- The buyer provides expected no-promotion units, gross promotion units, realized prices, variable costs and any measured later displacement. A blank displacement field remains unmeasured, rather than an observed zero.
- Use one currency and a consistent tax/cost basis. This is a campaign-level input contract, not direct ingestion of raw order records or a connected commerce API.
You receive
- Contribution difference and a reconciled bridge for volume, price, unit cost, refunds, campaign spending, supplier support and supplied later displacement.
- Break-even response where defined, product–campaign ranking and sensitivity to the supplied baseline and scenario inputs.
- Results CSV, a printable decision report and portable project JSON. Local project saving is optional; exported backups are separate from browser storage.
- A transparent conditional comparison, rather than an automatically estimated counterfactual, causal effect or demand forecast.
Experiment studio
Use before an experiment to plan allocation and approximate power, or after it to compare independent assignment-unit outcomes. The buyer supplies the design and measurements. Personal and Agency include the offline studio with up to 25,000 assignment or unit/pair outcome rows; hosted unit-level previews support 100 rows.
You supply
- Assignment CSV: unit_id. Stratified allocation also needs stratum; paired allocation needs pair_id with exactly two units per pair. Supply the eligible-unit list, scheme and seed before outcomes.
- Independent-outcome CSV: unit_id, group, outcome. An optional stratum column supports within-stratum contrasts when multiple strata are supplied.
- Two-summary CSV: group, n, mean, sd, with exactly one control and one treatment row. n counts independent assignment units and sd is their sample standard deviation.
- Matched-pair CSV: pair_id, control, treatment. Each row is one independent pre-matched treatment/control pair, not an uncontrolled before/after observation.
- Two-period difference-in-differences CSV: unit_id, group, before, after, optionally stratum. Every unit has one before and one after value; the tool does not accept staggered-adoption or repeated-cross-section designs.
- Planning inputs include the outcome SD, worthwhile detectable difference, alpha, power, arm ratio, expected loss of whole units, and optional equal-cluster size and ICC. The tool does not estimate those design assumptions from a commerce feed.
You receive
- Reproducible assignment and a local analysis-plan draft, with assignment CSV and protocol export. This does not conceal allocation or formally register a study.
- Approximate sample-size, MDE and power calculations with explicit clustering and small-sample limitations.
- Treatment-minus-control effects with Welch or paired t intervals where estimable; two-period DiD uses within-unit changes. Insufficient variance information leaves uncertainty unavailable.
- Result, analyzed-row and applicable stratum CSVs, printable views and portable experiment-workspace JSON. The studio neither deploys offers nor retrieves live customer outcomes.
Portfolio planner
Use when the next decision is which mutually exclusive offers to choose under operational constraints. The offline planner is an Agency feature supporting up to 120 products and 2,000 candidate offers; the hosted preview supports 8 products and 32 offers. The buyer supplies candidate economics and response assumptions.
You supply
- Candidate CSV columns: product, offer, regular_price, promo_price, unit_cost, baseline_units, demand_mode, expected_units, elasticity, marketing_spend, supplier_funding, available_stock, capacity_per_unit, currency.
- Required for each row: product, offer, regular_price, promo_price, unit_cost and baseline_units. Manual demand mode requires expected_units; elasticity mode requires a supplied non-positive elasticity and a strictly positive offer price.
- All offers for one product share exactly the same regular price, variable cost, baseline units, stock and capacity per unit. Each product/offer combination is unique. Numeric CSV fields use decimal points without thousands separators or currency symbols.
- Set the global campaign budget, total capacity and maximum markdown in the planner. Funding affects contribution but does not offset the cash-spending budget. Total capacity includes regular-price sales for unpromoted products.
- Keep one currency and planning horizon. Supply expected offer volumes directly or choose an explicit constant-elasticity scenario. No fitted demand curve, cross-product substitution, returns process or later displacement is inferred.
You receive
- One candidate offer or a no-promotion policy per product, evaluated against the common supplied baseline.
- A feasible contribution plan when one is found, with selected spending, expected units, stock and capacity checks, and regular-price and greedy comparisons.
- A proved best combination when the bounded search establishes it; otherwise the best feasible plan found and an upper bound on possible remaining improvement. Search stops at its node/time limits, so a feasible plan is not always proved optimal.
- Selected-plan CSV, candidate export, printable planning report, scenario shocks and portable project JSON. The optimization gap is a search bound, not a statistical confidence interval or a revenue guarantee.