TikTok for Business published a Marketing Mix Modeling analysis on July 24 to discuss advertising efficiency beyond a single visible conversion path. The study concerns a specific market and should not be projected onto an exporter as a forecast. Its practical value is methodological: a business buyer may discover a supplier on TikTok, search the brand later, inspect the website, email a technical question and continue on WhatsApp. Last-click reporting usually records only the final observable step.

Attribution and incrementality answer different questions

Attribution asks which channel a measurement system credited. Incrementality asks whether an outcome would have happened without the activity. A pixel, analytics tool and CRM can document parts of a journey, but device changes, privacy controls, offline sales work and long buying cycles create gaps. A mix model, conversion-lift study or geographic test attempts a different inference, and its usefulness depends on the data, assumptions and execution.

There is no permanently correct universal number. A decision brief should show each evidence source with its definition, window, market coverage and uncertainty. Agreement across several imperfect measures is usually more useful than forcing them into one artificial total.

Build a B2B evidence ladder

The first level is operational availability: did the campaign deliver, did the landing page load, and did the form or messaging route work? The second level is buyer behavior: video completion, specification-page visits, document views, qualified conversations and request preparation. The third level is sales quality: did the inquiry include market, product, quantity, application and timing? Only then should a team add an incrementality experiment or model.

Keep the original meaning of each event. A short video view is not purchase intent. Opening WhatsApp is not a qualified inquiry. Sales qualification should enrich the record with reasons and context without overwriting the original event.

Do not transplant an MMM result across markets

Marketing mix models typically use time series for media investment and business outcomes. A B2B exporter may have too few outcomes, simultaneous market changes or a sales cycle too long to identify one channel cleanly. A study in Poland is evidence about a measurement approach, not evidence of the same effect for a Chinese industrial supplier in Germany, Mexico or the Gulf.

A smaller team can start with a bounded geographic or time-based comparison. Define the spend, audience, creative set, landing page, sales-response rule and observation window in advance. Avoid changing pricing, page design and follow-up procedures during the same test; otherwise a movement cannot be attributed to a clear operational cause.

The measurement window should match the buying motion. A low-consideration sample request and a capital-equipment project should not share the same conversion deadline. Track leading evidence during the test, but wait for the agreed commercial window before judging downstream quality. If sales capacity differs by market, record response time and follow-up completion as covariates. Otherwise, a channel may appear weak when the actual constraint is an unanswered inquiry or missing technical documentation.

Use a pre-test memo to prevent retrospective storytelling. It should state the decision that will be made, the minimum usable sample, the events considered diagnostic and the conditions that invalidate the comparison. After the test, retain results that contradict the preferred narrative. A measurement system is valuable only if it can change a budget or process decision, not if it is used to decorate a conclusion already chosen.

What this means for Chinese exporters

TikTok B2B activity often contributes discovery, education and early trust. The website and sales process then carry technical verification and commercial progression. Last-click reporting may understate upstream influence, while platform attribution may overstate the part it can see.

Management should ask for an evidence portfolio rather than a channel claim: what was directly observed, what was modeled, what sales confirmed, and what remains unknown. This supports better content, landing-page and follow-up decisions without turning an external study into a performance promise.

Action checklist

Sources