How to Launch Web Traffic: A Practical Framework

Web Traffic does not require a replan. It takes one campaign on a small share of the budget, one decision about which optimisation event to open on, and a baseline recorded before the campaign starts. Most tests that produce no conclusion fail on that last point, not on the objective.

Here is the sequence, and the details that decide whether the result is readable.

Step 1: Identify the bottleneck

The entry point depends on what is limiting the account today.

If several apply, pick the one tied to the KPI being defended this quarter. One campaign should answer one question.

Step 2. Test the optimisation event rather than assuming it

TikTok's own recommendation is to run a split test across clicks, landing page views, and engaged sessions, and to match the choice to the customer's consideration cycle: clicks and landing page views for short cycles, engaged sessions for long ones.

Two practical caveats. Engaged Session is a rarer event than a click or a page view, so an account with a small budget and no history may struggle to accumulate enough of them for the model to learn from. And the ten-second baseline can only be moved upward, to a longer duration or a set number of page visits, through your TikTok representative. There is no way to lower it, so a page that cannot hold attention for ten seconds has to be fixed rather than accommodated.

On bidding, start with Maximum delivery. Manual bid adjustment at this stage tends to worsen CTR and raise CPC.

Step 3. Settle measurement before launch

Engaged Session can run without a pixel, but pixel-free availability is limited to selected regions, and TikTok recommends connecting an existing pixel for the most accurate measurement and reporting. Confirm availability for your geo before the plan depends on it.

Step 4. Check the landing page

An Engaged Session is ten seconds on the page or an action on it, so the page is part of the optimisation:

• Slow load costs the visit entirely. A visitor who leaves before the page renders produces no engaged session and no landing page view either.
• The visitor needs something to act on within reach: a second-page link, a calculator, a video, a form. Without it, every qualifying session has to come from the full ten seconds.
• Judge the page on mobile only. Desktop behaviour is irrelevant to this buy.

Step 5. Launch as a separate layer and freeze the rest

Run one new campaign on roughly 10–15% of the account budget, which is the allocation recommended when cannibalisation is the concern. The exact figure should follow account economics, audience size, funnel maturity, and available budget.

Traffic is designed to run alongside lower-funnel campaigns rather than instead of them, and the combination is what drives the return. Match the campaign duration to any promotion you are running, and coordinate its timing with the conversion campaigns.

Everything already running stays as it is. During the test window, change nothing else: no budget reallocation, no new creative in the conversion campaigns, no audience edits. Decide the audience overlap policy up front and keep it fixed.

Step 6. Record the baseline, judge at account level

Capture the same list for a comparable period before and after:

• Conversion volume and CPA across the whole account
• Budget utilisation: planned spend against actual spend
• Retargeting pool size
• Assisted conversions
• Session duration and bounce rate from your own analytics
• Cost per engaged session and engaged session rate

For reference, TikTok's figures for Traffic and Performance running together against Performance alone are +23% lower-funnel CVR, −36% CPA across the account, and ×5 budget utilisation.

Not CPC. Budget that finally gets spent at the target CPA, a retargeting pool that is filling again, and search performance lifted by users who engaged on the site are all invisible in that metric.

Four things that invalidate the test

1. Evaluating the Traffic campaign on its own CPA, in isolation from the account.
2. Editing the conversion campaigns during the test window.
3. Running Engaged Sessions on a budget too small to accumulate a rare event.
4. Reading the result before the campaign has exited learning, or against a baseline that was never recorded.