Why has RTB House's retargeting campaign performance declined by 15% in the last month for e-commerce clients?

Introduction

A 15% decline does not identify a cause. First define the performance metric and reproduce the result for the same mature client cohort, attribution window, currency, and conversion rules. Then locate the first break across measurement, eligible traffic, auction delivery, creative exposure, site conversion, and incremental outcomes. An algorithm change is only one hypothesis and should not be assumed.

Verified context and interview assumptions

• RTB House currently describes its offering as Deep Learning-powered performance advertising for web and app, including next-generation retargeting and personalized shoppable creative based on first-party signals. This describes the product context, not the cause of the reported decline.
• RTB House’s own definition of incrementality testing in retargeting compares an exposed group with a control group that does not receive ads to distinguish advertising-driven conversions from conversions that would have happened anyway.
• The IAB Tech Lab’s OpenRTB standard documents the programmatic bid-request and response protocol. It supports an auction-stage decomposition but does not imply that an auction fault occurred.
• The Media Rating Council’s Outcomes and Data Quality Standards stress that outcome attribution depends on accurate delivery, exposure, and data quality.
• The 15% decline, its timing, and its e-commerce scope are interview premises. No supplied evidence establishes a model release, broad client impact, creative fatigue, competition, or any root cause.

Step 1

Clarifying Questions (3 minutes)

• **What does “performance” mean?** Is the 15% a relative fall in attributed ROAS, incremental ROAS, conversion rate, revenue, conversions, or another KPI? Is it a weighted portfolio total or an average of client-level changes?

Why it matters: Revenue can fall while ROAS holds, and attributed ROAS can move differently from incremental value. Ask for: Formula, numerator, denominator, weighting, currency conversion, comparison dates, and whether 15% means relative percent or percentage points.

• **Is the comparison mature and like-for-like?** Are both periods complete under the same click and view attribution windows, refund policy, time zone, identity, and late-conversion treatment?

Why it matters: Recent days often look worse before conversions mature, and a reporting change can mimic a real decline. Ask for: Conversion-lag curves, data freshness, event versions, attribution settings, deduplication, cancellations, and restatements.

• **Which clients and funnel stages contribute to the gap?** Is the decline broad or concentrated by client, vertical, market, device, browser, app or web, inventory source, campaign, or new versus established setup?

Why it matters: A portfolio mix change needs a different response from a shared delivery defect. Ask for: Same-client cohort results, contribution to total loss, and counts and rates from eligible user through purchase.

• **What changed and who was exposed?** Were there changes to tracking, consent, feeds, creatives, campaigns, budgets, bidding, models, supply integrations, attribution, or client sites?

Why it matters: A dated change is not causal unless affected traffic or users actually received it. Ask for: Release and configuration logs, experiment assignments, feed health, consent trends, site releases, and supply-partner incident records.