Dynamic Pricing
Let CloudX learn prices for an ad unit, or compare Dynamic Pricing with your existing line items.
Dynamic Pricing explores prices and uses the results to train a model. CloudX then uses the model to price eligible auctions. In Manual mode, your line items set priorities and price floors.
New Dynamic Pricing setups default to 4 auction rounds. The auction stops as soon as a round produces a valid winning bid. Existing setups can retain fewer rounds.
Check eligibility
To start or re-enable Dynamic Pricing, an ad unit must average at least one eligible auction per day over the last seven complete days. Ad units in First Look apps cannot start or re-enable Dynamic Pricing. Other apps in the account can still qualify.
Check Dynamic pricing eligibility in the Ad Units list. On an ad unit’s Monetization tab, Change mode explains any setup requirement that prevents switching. Finish an existing line-item A/B test before changing modes. An active managed priority or a conflicting pricing configuration can also prevent setup.
If eligibility cannot be checked, retry when the error appears. Missing traffic data does not mean the ad unit has too little traffic.
Choose a rollout
- Open Ad Units, select an ad unit, and open Monetization.
- Select Change mode on the Monetization mode card.
- Select Dynamic pricing, then choose a rollout.
- Confirm with Change mode.
| Rollout | What you need | What happens |
|---|---|---|
| Run an A/B test | Applicable active line items for the control | Choose how much traffic to use for first-model training. Once a qualified model is ready, the comparison uses 50% Dynamic Pricing and 50% line items. |
| Roll out to all traffic | An eligible ad unit; line items are optional | CloudX starts with 100% exploration. Once the first qualified model is ready, CloudX sends 85% of traffic to it and reserves 15% for exploration. |
The dialog recommends an A/B test when that rollout is available. You can also select Create A/B test from the ad unit’s A/B tests tab or A/B Tests in the sidebar. Choose an ad unit if prompted, select Dynamic pricing (ML), and confirm with Create A/B test. The test name is Dynamic pricing vs. line items. A new setup starts learning immediately; it does not create a line-item draft.
Complete any required partner mappings on the ad unit’s Networks tab.
Allow time for training
The first model needs at least one day of observations. Collecting enough data, training, and validation can take longer. This does not promise a ready model after 24 hours.
The dashboard shows a best-case estimate for collecting training data. Some exploration auctions yield no usable data, so collection can take longer. The model stays in training until there is enough usable data from the past 30 days.
CloudX recommends training on 100% of eligible traffic to collect data sooner. A lower exploration share leaves more traffic on existing line items during training. A/B results start accumulating once the model is ready and the 50/50 comparison begins.
Read results and end a test
Open A/B Tests in the sidebar for tests across your account. The ad unit’s A/B tests tab shows its current test and completed tests under History. An all-traffic rollout has no A/B report. Check Model status on the Monetization tab to see whether its model is Learning or Active.
The main Dynamic Pricing comparison metric is CloudX revenue per million requests. For a test against existing line items, it compares all Dynamic Pricing traffic, including model and exploration traffic, with the line-item control. Revenue totals and impression counts alone cannot establish which group performed better when traffic shares differ.
Preliminary results can start appearing about six hours after the request hour closes. They can change as more revenue arrives. A confidence interval can remain unavailable until enough valid data arrives. CloudX does not choose the winning group automatically when results become significant.
Once the trained-model comparison against existing line items is active, select End test, choose an outcome, and confirm with End test:
- Use Dynamic pricing applies Dynamic Pricing to all eligible traffic, including its continued exploration.
- Use Manual returns traffic to your existing line items.
During learning or model readiness checks, select Cancel test on the test page and confirm with Cancel test. This stops the test and returns all traffic to your existing line items. You cannot select Dynamic Pricing as the winner before the model is ready.
The ad unit’s Monetization → Change mode dialog blocks mode changes while an A/B test is in progress. Select View test to open the test page, then cancel or end the test there. Cancel only closes this dialog.
Completed comparisons retain their saved results. CloudX evaluates later model updates in the background and selects models automatically. Publishers do not choose between model versions or receive separate model-versus-model A/B tests.
If a test has multiple periods, select one under Measurement periods on its detail page to review those saved results. Opening the test normally shows its current period.
Change modes or run another comparison
To stop an all-traffic rollout, open Monetization → Change mode, choose Manual, and confirm.
From Manual, choose Resume previous model when a previously adopted model is available. To train a new model, choose Start fresh or Roll out to all traffic. You can also choose Run an A/B test when active line items are available for the control.
After a stopped Dynamic Pricing A/B setup, switch to Dynamic pricing and choose:
- Run an A/B test to restart training with your chosen exploration share. Active line items are required.
- Roll out to all traffic to enable Dynamic Pricing without a comparison.
Both choices must pass the current traffic and app eligibility checks.
To compare an all-traffic rollout with existing line items, select Create A/B test from the ad unit’s A/B tests tab or A/B Tests in the sidebar. The dialog offers Dynamic pricing (ML). An active line item is required for the control. If a model is active, the 50/50 comparison starts immediately. If the ad unit is still exploring, choose an exploration share for training. The 50/50 comparison starts when the model is ready. Wait for any model check in progress to finish before starting.
After choosing a trained model in an A/B test, open the ad unit’s A/B tests tab and select Create A/B test to test new line items. Review the draft, then select Start test. Dynamic Pricing continues serving while you prepare the draft. The new test splits traffic equally between Dynamic Pricing and the new line items. In this comparison, Promote test chooses the new line items; Keep control keeps Dynamic Pricing.
Bidder settings and fallback
While Dynamic Pricing is running, select Edit beside Bidders on the Monetization mode card. Use Excluded bidders in Bidder settings to remove bidders, then select Save changes. At least one app-enabled bidder must remain available. These settings affect Dynamic Pricing traffic and do not edit your line items.
Changing bidder exclusions preserves the traffic split, and devices stay in their group. A model trained on one set of bidders is never evaluated or promoted on another, so during a running test the change starts a new revision of the same test:
- Model under test against your line items: the comparison ends and Dynamic Pricing goes back to training on the new bidders. The next model activation starts the revision after.
- Learning: any model in training is dropped and training restarts on the new bidders.
- Weekly model cycle: the cycle ends and its candidate is dropped. The current model keeps serving, and the next cycle trains on the new bidders.
- Line items against Dynamic Pricing: the test ends.
Activating a model also starts a new revision of the same test, with a fresh results and promotion window. Earlier results remain available per revision. Tests created before revisions were available keep the previous behavior: a bidder change starts a new measurement period.
CloudX continues some exploration after a model starts to collect training data. The amount depends on the saved allocation and available evidence. Publishers do not set that percentage in the rollout dialog.
If a model is unavailable or returns an unusable price, CloudX can continue with price exploration.