Huawei Ads Automation: What to Hand to AI and What to Keep

Bidding is the main candidate. Huawei Ads offers optimization toward goals such as installs, activations or in-app events, with automated bidding toward a target cost. When conversion data is flowing correctly, automated bidding usually responds to auction changes faster than manual bids.

Audience expansion and placement selection are the second. Letting the system find users similar to converters across AppGallery and partner inventory saves manual testing. Dynamic creative combination, where the platform mixes titles, images and descriptions, is the third.

• Bid adjustments toward a target cost per action.
• Audience expansion based on converting users.
• Placement distribution across available inventory.
• Creative asset combination and rotation.
• Scheduled reports sent by email.

Tracking and goals. The platform optimizes toward whatever event you choose, so a poor choice, such as optimizing for installs when you care about paying users, leads automation to cheap users who never return. Choose the deepest event that still has enough volume.

Budgets and caps. Set daily budgets and campaign caps yourself, and do not let automation raise them. Exclusions too: placements that send poor-quality traffic, countries you do not serve, and audiences of existing users if you only want new ones.

• Conversion event selection and attribution setup.
• Daily and total budgets.
• Country and language targeting.
• Placement and audience exclusions.
• Creative concepts and brand rules.

Integrate the HMS Core Analytics Kit or connect a supported mobile measurement partner to send install and in-app events to Huawei Ads. Confirm in the platform’s conversion tracking section that events appear with correct counts before starting automated bidding.

Test with a real Huawei device if possible. Many integration problems only show up on HMS devices: an SDK initializes on Google devices but fails silently on Huawei ones, or events fire on Android but not on HarmonyOS builds. Compare counts between the MMP and Huawei Ads after the first days.

Choosing the optimization event

For example, a food delivery app might optimize first for registration, because purchases are too few in the first week to train the algorithm, then switch to first order once the campaign records enough of them. The exact threshold depends on the platform’s guidance in your account; ask the account manager if it is not shown.

Automated campaigns need a learning period. Changing the target cost, budget or creative every day restarts learning and makes results look unstable. A practical rule is one meaningful change per campaign per week, with notes on what changed and why.

For targets, adjust in steps. If cost per action is above goal, lower the target gradually rather than halving it, because a sharp cut can stop delivery. If delivery is weak, raise the target slightly or broaden the audience before increasing the budget.

Compare automated campaigns against your own goal, not against the platform’s estimate. Pull cost per install, cost per key event and day-7 retention by campaign from your MMP or analytics, and check that the cost per valuable user is falling or stable as spend grows.

Look at placement and country reports weekly. Automation may push spend toward cheap inventory with poor retention. When a placement or region shows many installs and few retained users, exclude it.

• Weekly: cost per action, delivery, placement report, country split.
• Bi-weekly: day-7 retention by campaign.
• Monthly: revenue per user by campaign compared with acquisition cost.

Supply the system with enough variety: several images in the required sizes, a few videos, multiple titles and descriptions. Automation can only combine what you give it. Rotate in new assets every few weeks and remove the ones with consistently poor click-through.

Keep creative concepts under human control. The platform finds which combination works, but it does not know which message fits your brand or a seasonal campaign. Localize for each market; Huawei’s user base is large in regions such as the Middle East, Africa, Southeast Asia and parts of Europe, and each needs its own language and visuals.

Optimizing for installs indefinitely is the most common. The second is running one campaign across many countries with very different costs, which lets the algorithm spend in the cheapest market regardless of value. The third is ignoring HMS-specific testing, so events break on Huawei devices and nobody notices for weeks.

Splitting campaigns by country tier and checking events on a real device prevent most of these.

Reporting rhythm and documentation

Write a one-page playbook for each campaign: the optimization event, the target cost and how it was calculated, the countries included, the exclusions and the creative set in use. When results change, the playbook shows what the automation was told to do, which speeds up diagnosis.

Export platform reports weekly to a shared sheet alongside MMP data. Platform dashboards and attribution windows can differ from your own analytics, and keeping both side by side makes gaps visible early. Agree on which source is the reference for decisions, usually the MMP for cost per retained user, and use the platform view mainly for delivery and creative performance.

Share the playbook with whoever builds creative and whoever manages the app release cycle. A new app version that changes event names can break tracking overnight, and automation will keep spending on stale signals until someone notices.