BRAND FOOTPRINT REVIEWS
One brand, from the ad to the page.
I want to see whether the message holds together across the ad, the social presence and the destination. The review separates the evidence from my interpretation and the next test.
THE EVIDENCE LAB / INTERACTIVE PORTFOLIO DEMONSTRATIONS
What does the evidence
actually support?
I use competitor ads to develop ideas. Then I decide what needs testing, what can be measured and what conclusion the evidence would support.
The starting examples and experiment results are synthetic. This workspace supports manual evidence capture; it does not fetch live ads or infer competitor performance.
01 / COMPETITOR & CREATIVE AUDIT
What I would learn
from the creative.
I would fix the category, market and review window first. Then I would review a consistent sample and keep each recommendation tied to what I actually observed.
Explore brand reviews
Ad Library
Review visible creative, copy and advertiser details. Coverage includes active ads, with additional regional and regulated-ad archives.
Open Meta Ad Library ↗Two different lenses
The Commercial Content Library covers EEA, UK and Switzerland ads. Top Ads is a curated, advertiser-authorised inspiration set, not a full competitor census.
Open Commercial Content Library ↗Explore Top Ads ↗Ads Transparency Center
Inspect ads across Search, YouTube and Display. Record the advertiser, format, market filters and available dates alongside your observations.
Open Google’s ad library ↗The audit protocol
1. Define the sample. Fix competitors, country, date window and inclusion rules. Record ads reviewed and coverage gaps.
2. Capture the evidence. Save the detail URL or ID, capture date, market and verbatim observations. Deduplicate repeated creative variants.
3. Apply one rubric. Tag hook, offer, proof and CTA. Score messaging clarity using the same anchors; use a second reviewer for calibration.
4. Write a hypothesis. Separate “the ad shows a demonstration” from “a demonstration may improve qualified conversion.” Validate with your own test.
Longevity, ad counts and engagement are not proof of profitability. Missing ads do not prove a competitor is inactive. Platform samples and metrics are not directly comparable.
Creative evidence board
Fictional advertisers below illustrate the workflow. Add your own observations from real library detail pages.
Add an ad from a real library
Open a library above, review an ad, then record its detail link and your observations. Entries stay in this browser only. Ad details stay in this browser; this tool does not scrape ads, connect an advertising account or submit your notes externally. Optional website analytics is explained in the privacy notice.
What I would take into the discussion
I would show the pattern, the underlying ads, an alternative explanation and the test I would prioritise. Public ads can suggest an opportunity; they cannot tell me which competitor is profitable.
Before I trust a test result, I check the measurement.
The flagship case walks through a filtered experiment report and faulty SQL, with downloadable data and executable checks. Read the experiment-integrity case.
02 / INCREMENTALITY EXPERIMENTATION
What happened
because of the ads?
I want to separate credited conversions from additional conversions. A well-designed holdout helps estimate what changed because of the advertising.
EXAMPLE QUESTION
Does this campaign increase the share of eligible people who purchase?
Randomly assign eligible people before delivery. The treatment group is eligible for the campaign; the control group is withheld from that campaign. Analyse everyone as assigned, whether or not they received an impression.
User holdout
Use a supported platform lift study or a reliable person-level randomisation and suppression system. Fix assignment and measure outcomes consistently across groups.
The calculator below assumes independent users and one binary outcome per user.Geo experiment
When user holdouts are impractical, use enough non-overlapping regions, assess pre-period balance, and randomise within a defensible design.
Analyse at the randomisation unit. Use geo-aware inference and account for spillover. The user-level calculator below does not apply.Explore a user-level holdout
Synthetic fixed-horizon example · pre-specified 50/50 assignment · two-sided 95% normal approximation
Absolute lift
Relative liftVersus control purchase rate
Incremental purchasersWithin treatment-sized population
Modelled incremental ROASAssumes fixed purchaser value
95% INTERVAL / ABSOLUTE LIFT
COST & VALUE
How the calculations work
Absolute lift: treatment purchase rate − control purchase rate.
Relative lift: absolute lift ÷ control purchase rate.
Incremental purchasers: absolute lift × treatment users. This scales the control rate to the treatment population.
95% interval: absolute lift ± 1.96 × √[pT(1−pT)/nT + pC(1−pC)/nC]. An approximate, unadjusted interval for independent binary outcomes.
iCPA: incremental campaign spend ÷ incremental purchasers. Modelled iROAS: incremental purchasers × assumed value ÷ incremental campaign spend.
Revenue is modelled using a fixed value, not measured revenue lift. A real revenue experiment needs its own revenue variance and uncertainty analysis. Return on revenue is not profit; consider margin and fulfilment costs.
Plan the sample before launch
Approximate two-proportion planning at 80% power, two-sided 5% alpha, equal-sized groups. Fix the MDE from business relevance before looking at results. Validate with the study provider; clustering, multiple tests and attrition can increase requirements. Duration ≈ required total users ÷ eligible users per day, plus outcome maturation.
What I would agree before implementing the test
Pre-register the decision.
Specify the campaign, eligible population, primary metric, economically meaningful effect, sample size, analysis window and scale/hold/stop rule. Record secondary metrics separately.
Protect the counterfactual.
Randomise before delivery. Keep assignments stable. Suppress only the tested intervention in control and document background activity, overlap, leakage and cross-device limitations.
Measure business outcomes.
Use a consistent purchase definition and deduplicated outcome source for both groups, independent of an ad click. This binary demo counts purchasers, not repeat orders. Check refunds, exclusions and conversion lag.
Check the experiment before interpreting lift.
Audit assignment ratios, pre-treatment balance, missing outcomes and contamination. Analyse by original assignment. Do not switch to exposed-only users, which breaks randomisation.
Finish the planned window.
Wait for the fixed sample, duration and conversion lag. Repeatedly checking and stopping when significant inflates false positives unless a valid sequential method was planned.
Make a bounded decision.
Report the effect, interval, cost and limitations together. An interval spanning zero is inconclusive, not proof of no effect. Re-test after material changes to audience, creative or spend.
How I would connect the audit to a test
First test whether a creative approach beats the current creative under a controlled A/B design. To test advertising incrementality, add a no-tested-campaign holdout. A creative A/B comparison alone cannot tell you how many purchases would have happened without the campaign.
METHOD & PLATFORM REFERENCES
Platform scope checked 12 November 2026. Availability and fields can vary by region and account.
← Back to selected work