Research & Proof
Marketing science that survives a board review.
We publish our methods before we publish our wins — design-partner case studies, attribution benchmarks vs. last-click, and our co-research with NVIDIA on accelerated marketing inference.
Attribution benchmarks
Confidence ranges vs last-click attribution
The reason last-click survives in marketing reporting is not that it's accurate — it isn't. It survives because it's cheap, deterministic, and easy to explain. What WatEase Research gives a board review is the one thing last-click cannot: a calibrated answer to "how confident are you?" Below is the gap we see consistently across channel-mix benchmarks on India D2C accounts in the ₹50 lakh–₹15 crore monthly-spend range.
| Reporting question | Last-click answer | WatEase Research answer |
|---|---|---|
| Channel contribution to revenue | Single point estimate per channel; sums to 100%; no error bars. | Per-channel contribution with an explicit confidence range; uncertainty is made visible. |
| Brand-campaign incrementality | Routinely zero — brand-driven conversions attributed to "direct" or "organic search". | Delayed conversion contribution assigned back to the brand channel that caused it. |
| Saturation / diminishing returns | Not modeled. Spending ₹2 cr or ₹4 cr on Meta returns the same per-rupee number. | Saturation curve per channel; marginal ROI drops as spend climbs past the inflection point. |
| Cross-channel synergy | Invisible. Meta-then-Google sequences attributed to whichever fired last. | Co-spend lift captured by joint modeling; geo-experiments inform the synergy estimate. |
| Confidence interval on ROAS | None. The number is the number. | Explicit confidence range; the CFO sees the floor and the ceiling, not just the point estimate. |
| Budget-reallocation guidance | "Shift to whatever had the highest last-click ROAS last week." | Reallocation proposal with expected uplift and a risk band, not a single point estimate. |
Live benchmark numbers from design-partner cohorts publish here as each 90-day measurement window closes. Definitions of the underlying terms are documented in the glossary with explanations calibrated for marketing leaders, not statisticians.
Research methodology
How the 90-day design-partner window works
Each design-partner engagement is structured as a 90-day window because that's the shortest period that captures a full marketing rhythm — always-on baseline, one promotional spike, one creative refresh — and exposes the carryover and saturation effects that shorter windows miss. The four phases run in sequence; the artefacts at each step become inputs to the next.
- Phase 1 · Weeks 1–2
Data ingest & baseline calibration
Two years of weekly impressions, spend, and conversions per channel are pulled through the WatEase connectors. Expectations on channel carryover, saturation, and brand-vs-performance split are codified with your team before the model runs. Flat assumptions waste data you already have.
- Phase 2 · Weeks 3–4
Model fitting & validation
The model is fitted to your historical data and validated against held-out weeks. Back-fits compare model-generated weekly revenue to actuals. Anything that doesn't reproduce gets flagged before the forward predictions are trusted.
- Phase 3 · Weeks 5–10
Geo-holdout incrementality test
One channel is dark-paused in a matched-pair geo cohort (typically two tier-2 cities matched on prior 26-week trajectory) and run for 4–6 weeks. The observed geo gap becomes a strong input on that channel's incrementality in the next model refresh. This is the step that separates "the model says X" from "the model agrees with the experiment that says X."
- Phase 4 · Weeks 11–13
Reallocation & readout
The optimizer reads the calibrated model output, proposes a budget reallocation with expected uplift and a risk band, and the team executes it via the WatEase ad-platform connectors. The readout deck contains the per-channel contribution range, the incrementality-test result, the proposed reallocation, and the back-test of last quarter's decisions against what the model would have recommended.
Frequently asked
What marketing leaders ask before booking a research call
- How do you defend cross-channel attribution to a CFO who has only ever seen last-click?
- Last-click gives a single number with no error bar. WatEase Research delivers a contribution range per channel — the CFO sees both the central estimate and the confidence band. Marketing leaders walk into the budget review with both numbers and explain the delta. The confidence band is what survives finance scrutiny because it admits uncertainty instead of pretending it doesn't exist.
- What data do you need to start a research engagement on WatEase?
- Two years of weekly data is the most useful window. For each marketing channel: impressions, spend, and (where available) clicks or unique reach. For the business outcome: weekly revenue or unit sales, broken out by SKU or category if you want product-level attribution. Optional but valuable: incrementality experiments, pricing changes, distribution events, and macro covariates like festival seasons or competitor launches. More data produces tighter confidence bands.
- How does WatEase Research compare to running open-source attribution libraries in-house?
- Open-source libraries give you a model. WatEase ships the model plus a managed data pipeline, ad-platform connectors, the optimizer that translates results into a budget reallocation, the execution layer that deploys the decision, the UI your CFO can read, and the ops team that keeps it running. The choice is the same as Apache Airflow vs Astronomer or Postgres vs RDS — model autonomy versus time-to-first-insight.
- How long does a measurement window need to be before the model is trustworthy?
- The model converges fast — typically 4–8 weeks of weekly observations is enough to produce reasonable confidence bands on the largest channels. Tail channels (those at <5% of spend) need more data, often 6 months or more, because the signal-to-noise ratio is lower. We run the 90-day measurement window in design-partner engagements because it covers a full marketing-rhythm cycle (always-on baseline, one promotional spike, one creative refresh) which exposes seasonality and carryover effects the way a shorter window does not.
- Can WatEase quantify the incremental impact of brand campaigns, not just performance?
- Yes — this is where research-grade attribution produces leverage that last-click cannot. Brand campaigns generate delayed conversion contribution that last-click attributes to "direct" or "organic search" weeks later. Joint modeling of all channels with carryover and saturation effects separates the brand-channel contribution from the lower-funnel contribution. Pairing it with an incrementality experiment (geo-holdout for the brand channel) tightens the result further.
Marketing-science glossary
Plain-English definitions of the methods this program runs on.
Case studies
Case studies launching with our design-partner cohort.
We're working with our first design partners on WatEase AI. Public case studies land here as each cohort completes its 90-day measurement window.
What will appear here
- Design-partner stories — named only with written consent, metrics shown as ranges with confidence intervals.
- Methodology posts — how our Bayesian MMM, lift tests, and budget optimiser actually work.
- Benchmark notes — analyses of public data (conversation pricing, UPI adoption) with cited primary sources.
What will never appear here
- Anonymous “300% ROAS” miracle numbers that can't be traced to a method.
- Cherry-picked date windows or screenshots without the surrounding context.
- Point estimates without uncertainty — if we can't bound it, we won't print it.
Run the math on your own data.
Paste 3 months of Meta + Google spend. We'll show you the cross-channel-vs-last-click directional delta on your numbers.
Run the ROAS calculator