Diagnose a Structural Break in KYC Submission Rate

Scenario

You are a Data Scientist supporting the onboarding and compliance product teams at a global business payments platform.

During a weekly business review, leadership presents a line chart showing the KYC (Know Your Customer) Submission Rate—defined as kyc_submitted / account_created—over the trailing 12 months. The metric exhibits three distinct patterns:

1. A structural break: An abrupt 8 percentage point drop occurring over a single 48-hour window in mid-Q2.
2. A gradual secular drift: A steady 0.5% month-over-month decline in the submission rate across the rest of the year.
3. Year-end seasonality: A dip in submission rates every December, coupled with an increase in median time-to-submit.

Your task is to outline a diagnostic framework to uncover the root causes of each pattern, separate true user behavior changes from data or mix artifacts, and prioritize operational fixes.

Available Data

You have access to the following event-level tables:

• user_signups: user_id, created_at, country_code, channel (e.g., SEO, Paid Search, Referral, Partner API), device_type, business_type (Enterprise, SMB, Sole Proprietor).
• kyc_events: event_id, user_id, step_name (e.g., identity_document, proof_of_address, beneficial_owner), event_type (viewed, submitted, failed, passed), timestamp, error_code, app_version.
• system_logs: timestamp, service_name, endpoint, http_status, latency_ms.

Key Deliverables

1. Taxonomy of Hypotheses: Group potential explanations for the mid-Q2 structural drop into four buckets: Data/Tracking, Product/UX, Compliance/Policy, and Traffic Mix. State how each could cause an abrupt drop.
2. Disentangling Traffic Mix from True Drop: Define the exact metric decomposition (e.g., using a fixed-weight index or Kitagawa-Blinder-Oaxaca style shift-share decomposition) to determine how much of the gradual secular drift is attributable to changing user acquisition mix versus genuine within-segment drop-off.
3. Isolating Seasonality: Propose a method to normalize or validate the December dip against historical years, accounting for holiday business cycles versus onboarding backlog.
4. Investigation Plan: Specify 2–3 concrete SQL queries or statistical tests you would run first to pinpoint the exact failure point of the mid-Q2 drop.