Flagship workbench · Measurement & modeling
Campaign Impact Analyzer estimates the modeled incremental effect of your brand campaign: the sales, conversions, and demand the model attributes beyond the baseline that would likely have happened with no campaign at all. Upload your timeline, get a board-ready observational read with the statistics and caveats to defend it.
Evidence boundaryObservational modeling is not randomized causal proof; controls and data quality constrain the claim.
Your numbers almost always rise during a big brand push — but some of that growth was coming regardless: seasonality, your existing trend, promotions, other channels. The only honest question is the counterfactual one: what WOULD have happened with no campaign? The gap between what actually happened and that campaign-free model is your estimated incremental impact. Everything this tool does is built to estimate that gap rigorously, stress-test it, and make clear that a geo or holdout test is still the cleanest proof of causality.
From one spreadsheet to a defensible number in three steps.
A daily or weekly CSV/XLSX with channel spend and outcomes. GA4/Ads preambles, semicolon CSVs, EU decimal commas, Indian lakh/crore grouping, duplicate headers, and totals footers are handled with visible warnings.
Models learn your baseline from the pre-campaign period — your trend, weekly seasonality, and the other channels — then project what would have happened without the brand campaign.
Actual minus that baseline = incremental impact, with an honest confidence range, a placebo test, and three independent methods that must agree before we call it trustworthy.
One simple file: a timeline of your spend and your results. The parser handles standard GA4, Google Ads, Excel, EU, and Indian export quirks, then surfaces warnings where values need human verification.
Your file should span three phases of time — give us as much of each as you can.
The “before” window
The foundation of the whole analysis. Give us as much clean pre-campaign history as you can so we can learn your normal pattern. Minimum ~2 weeks; recommended 8–12 weeks or more. Rule of thumb: at least 2× the length of your campaign, and ideally 6+ weeks so daily data can reveal your weekly rhythm and trend.
The “during” window
The exact dates your brand campaign ran — you’ll set these after upload. Works from ~3 data points; best with the full flight (commonly 1–6 weeks).
The “after” window
Optional but valuable. Brand effects linger, so include 2–4+ weeks after the campaign ends and we’ll measure the (adstock) and whether the lift held.
Aim for 30+ rows total for confident results; ~14 is the practical floor — we’ll flag it transparently if your data is thin, never hide it.
Daily data is best (it captures weekday/weekend patterns); weekly works if you have 6+ months of history.
A handful of columns — one row per day (or week). Name them however you already track them.
dateOne row per day (or week). ISO dates stay timezone-safe, and slash dates use a column-wide convention with a warning when day/month order is ambiguous.
<channel>_spendRequiredSpend per marketing channel, e.g. dv360_spend, meta_spend, google_search_spend, youtube_spend, tiktok_spend. Include your brand channel plus the others (they act as controls).
<channel>_impressions, <channel>_clicksOptionalOptional, but they sharpen the model when you have them.
sales / revenue / conversions …RequiredYour outcomes / KPIs — at least one: sales, revenue, conversions, leads, users, signups… named however you track them.
Every brand’s data looks different — different channels, different KPIs, different naming. Nothing here is hardcoded. The moment you upload, we automatically detect your channels and your outcome metrics from the column headers, then let you pick your brand channel and primary KPI from dropdowns. Bring Meta + TikTok + retail media + offline sales, or just DV360 + conversions — and keep the original export formatting. If the parser sees mixed locales, repeated headers, or a totals footer, it tells you exactly what it changed before the model runs.
Each analysis answers one plain-English question, in language you can take straight to the board.
Counterfactual / Interrupted Time Series
How much additional sales activity does the model attribute to the campaign?
Why it matters: Use the modeled incremental revenue estimate alongside its interval, diagnostics and assumptions when evaluating campaign ROI.
How to read it: We show actual vs. the predicted campaign-free line. The shaded gap during your campaign window is the incremental impact. Bigger gap, bigger effect.
Moving-Block Bootstrap
How sure are we — what’s the realistic range?
Why it matters: A single number invites a fight. A defensible range (“between X and Y, 95% confident”) wins the room. We deliberately use the HONEST interval that accounts for the fact that days near each other move together — most tools quietly report a too-narrow range.
How to read it: If the whole range sits above zero, the effect is real. The width tells you how much to trust the precise figure.
We tested our own method
Is this lift real, or would your method find “lift” anywhere?
Why it matters: The most persuasive slide in the deck. We re-run the exact analysis on FAKE campaigns in periods where nothing happened. If the method is honest, it finds roughly zero there — and your real campaign stands far outside that noise.
How to read it: Fake campaigns cluster near 0%. Your real campaign sits far to the right. The further out, the more bullet-proof the result.
Was the data strong enough?
Could this data even detect an effect this size?
Why it matters: Stops you from over- or under-claiming. If the data was too thin to detect the effect, we say so up front — credibility you can’t fake.
How to read it: The “minimum detectable lift” is the smallest effect your data could reliably catch. If your effect is comfortably larger, you were well-powered.
Three independent methods, one answer
Does the answer hold up under different assumptions?
Why it matters: We estimate the lift three independent ways — a full model, a simple before/after, and a trend projection. When all three agree, no one can dismiss it as “just your model”. When they disagree, we tell you honestly.
How to read it: Three bars close together = robust. The simple before/after usually OVER-states lift because it ignores your growth trend — seeing that gap is exactly why modelling matters.
Pearson, Spearman, Kendall, Partial, Distance
Which channels actually move with your outcomes?
Why it matters: A fast read on which levers matter — and a guard against being fooled. We auto-pick the right correlation for your data and control for other channels so you see each channel’s UNIQUE association, not borrowed credit.
How to read it: Stronger, darker cells = tighter relationship. Correlation isn’t causation — that’s why we also run the counterfactual.
Adstock
How long until spend shows up in results — and how long does it linger?
Why it matters: Brand effects aren’t instant. Knowing the delay and the carryover (“half-life”) tells you when to expect payback and how long to keep measuring.
How to read it: The peak shows the best delay between spend and response. The half-life is how many days/weeks the effect keeps working after you stop.
Does spend lead outcomes?
Does past spend genuinely help predict future outcomes?
Why it matters: A formal directional check that spend LEADS outcomes, not the other way around — a stronger statement than correlation.
How to read it: A directional flag means past brand spend improves the forecast of the outcome beyond the outcome’s own history. It is predictive evidence, not randomized proof.
Multi-channel Regression + Elasticity
Where is your money actually working — and where are you wasting it?
Why it matters: Turns measurement into a budget decision. Contribution shows each channel’s share of results; elasticity shows efficiency — how much each 1% more spend returns.
How to read it: Higher contribution = bigger driver. Higher elasticity = more responsive to extra budget. We also flag channels whose effects overlap (so you don’t double-count).
Before / During / After
What changed across the campaign’s life?
Why it matters: The simplest, most intuitive view for any audience — average performance before, during, and after — with a significance test so it’s not just eyeballing.
How to read it: Compare the bars. We test whether the during-vs-before jump is statistically real, not noise.
Where the next dollar should go
If I have one more dollar, where should it go?
Why it matters: A single table that ranks every channel by correlation, contribution, efficiency and timing — the executive summary of where to lean in.
How to read it: Scan the brand row first, then the channels with the highest contribution and elasticity.
Ranked actions, efficient frontier
What should I actually do first, and with how much budget?
Why it matters: Findings become a ranked queue of evidence-graded recommendations, and a downside-robust optimizer shows the reallocation that survives uncertainty — not just the point estimate.
How to read it: Work the queue top-down; the optimizer panel shows the efficient frontier and the water-filled reallocation per channel.
Every figure ships with the method behind it, so your team — and your CFO — can check the working.
An interactive results dashboard with plain-language explanations on every metric, plus a polished, client-ready PowerPoint deck you can drop straight into a CMO presentation.
Campaign Impact Analyzer is available to invited beta accounts, with 5 analyses per day.