Google Ads published a measurement discussion on September 2 about situations in which real-time attribution, periodic incrementality tests and media mix models appear to conflict. The official summary also references longer purchase journeys, Qualified Future Conversions and the limits of chasing short-term return on ad spend. These issues are amplified in global B2B. A buyer may discover a video, search the brand, read a technical page, share it with colleagues and submit a requirement weeks later through a different device or person. No real-time platform metric can explain that entire commercial path on its own.
Start with the decision the number is meant to support
Real-time attribution is useful for operational questions: did delivery start, did the page load, did an event fire and did a creative prompt an observable action? An incrementality experiment, where conditions support one, asks whether an intervention created additional change relative to a comparison. A media mix model addresses longer periods, multiple channels and external factors for resource planning.
Different outputs do not automatically mean that one method failed. The apparent conflict often begins when a daily attributed event is expected to answer a quarterly investment question, or when a long-horizon model is used to diagnose a page outage that happened this morning.
Build three decision clocks
The daily clock covers availability and anomalies: spend, delivery, visits, page errors, forms and messaging routes. A weekly or monthly clock reviews movement from content to a defined requirement: target-account visits, specification use, document requests, qualified conversations and sales acceptance. A quarterly clock considers the market, channel and content portfolio for resource allocation.
Each clock should specify the earliest reliable read, normal reporting delay, responsible role and permitted action. A daily fluctuation may justify a tracking check; it should not automatically trigger a long-term market withdrawal. A quarterly pattern may justify investment review; it should not excuse a broken form that needs immediate repair.
Preserve future states in a long buying journey
When a lead reaches the CRM, retain company, market, product, application, buying role, expected timing, next step and the evidence currently available. “No order today” is not a useful universal status, and a sales team should not be pushed to close an attribution gap with a premature judgment.
Use operational stages such as unverified, requirement defined, technical review, commercial review and closed, and timestamp each transition. This allows an early content interaction to be considered within an appropriate window without assigning certainty where identity or causality is missing. Multiple participants and offline contacts should remain visible as limitations.
What this means for Chinese exporters
Chinese exporters often combine TikTok, search, an independent website, exhibitions, email and WhatsApp. Buyers cross channels, devices and company roles. Each platform sees only part of the process. A decision clock helps each team ask an answerable question: operations can repair a broken path, growth can compare content and destinations, sales can verify requirement quality, and management can assess a portfolio over a longer period.
This approach does not eliminate uncertainty. It prevents short-term signals from being promoted into long-term business results. It also reduces the habit of changing bids, pages and sales scripts simultaneously whenever two dashboards disagree.
Action checklist
- Define daily, weekly or monthly, and quarterly decisions with metrics, delay, owners and allowed actions.
- Keep platform attribution, first-party site events and CRM business stages as separate raw records.
- Document the identity key, time window and deduplication method used to connect those records.
- Preserve unmatched visits, multi-person buying groups and offline touches as visible gaps.
- Pre-register experiment variables, observation windows, sample conditions and stop rules.
- Show early intent, sales acceptance and later-stage movement together in recurring reviews.
Investigate definitions before choosing a preferred number
When methods point in different directions, first check date range, time zone, currency, event definition, attribution window, deduplication, consent behavior, channel coverage and reporting delay. If the difference remains unexplained, record it as a hypothesis with an owner and next test. A measurement stack is valuable not because it produces one absolute total, but because each form of evidence supports the decision clock for which it was designed and remains traceable to pages, events, companies and sales records.
Sources
- Google Ads, Build a measurement stack you can rely on to steer your campaigns, September 2, 2026: https://blog.google/products/ads-commerce/ads-decoded-podcast-measurement-stack/

