Adobe Advertising

vijayj_kumar · 1/17/18

How to check if a portfolio is ready to be optimized?

To understand this, you should understand some basic requirements of launching a portfolio.

Why do we need to collect data before launch?

The Ad Cloud technology predicts performance and set bids of keywords based on their historical performance. The more the data the Ad Cloud technology has, the more accurate the forecasts will be. Hence it is important to allow the Ad Cloud technology to collect data before launch.

What is model coverage?

The model coverage will provide statistics on how much click and revenue data has been collected on the bid units in the portfolio. With accurate models, the system will be able to make better predictions on how to bid the keywords for optimal performance. Once you know the quality of the models for your Portfolio (the click and revenue coverage), you will be able to determine which Portfolio Parameters should be modified to influence & improve the models. Select Portfolio Settings and check the model coverage to see how many of the bid units have ‘models’ (i.e. the optimizer has forecasts and hence can make educated bidding decisions). More the coverage, the better the bidding decisions will be.

Simulations

Simulations can be used to determine the baseline performance at the current budget and for the current objectives. The simulations are also a good indicator which tells whether model coverage is good and accuracy is high.

Simulation Checks

If the simulation data is accurate for current spend level then it indicates good model accuracy.
In Reports > Report settings, create a calculated column for Weighted Revenue. This should be same as the weighted revenue defined in portfolio objective function.
Generate and download the report.
Open the report and find out the Daily Average Clicks, Daily Average Cost and Daily Average Weighted Revenues by dividing the total numbers of Clicks and Weighted Revenue by the number of days for which the report was generated.
Compare these values against predicted values returned by simulation

Launch Best Practices

Before optimizing a portfolio, turn off any other type of bid management system that was being used on the portfolio.
Do not manually bid on keywords (unless absolutely needed) after launch.
Expect to see a period of volatility the first 2 to 3 weeks after launch.
Initially set portfolio spend strategy to Daily and enter spend target which is close to your average daily spend prior to optimization.
You may see an increase in costs or decreased ROI (the number of conversions, etc.) or both.

The primary reasons for volatility are as follows:

Very few keywords may have been productive historically. Adobe Advertising Cloud will be required to learn on all keywords to determine which keywords at which positions will contribute to the ROI goal.
Keyword history may be sparse and tracked for a very short time period so the data used by Adobe Advertising Cloud is not statistically broad.
Productive keywords may not have been very efficiently bid in the past. These keywords will be moved around to find alternate, more efficient positions resulting in some fluctuation in results for a short duration.

Portfolio Launch – Top Tips

DO not optimise a portfolio near the end of the week.
Before optimising, take a download of the account so that you have a record of previous bids in case anything goes wrong.
Once bids have been set by the technology, check on spend levels during the day, in comparison to previous week to ensure nothing is majorly out.

Model Accuracy

Model accuracy gives information on how accurate the forecasted revenue and cost were compared to the actual achieved value. This is given on a daily level, but should be aggregated to the last 7 days to judge accuracy as day-to-day fluctuations are normal.

Evaluating Portfolio Spend Strategy

Daily: This strategy is used when client has fairly even traffic trend throughout the week. 7 day average will be close to daily budget point. This strategy is easy to manage and cannot exploit day of the week trend.
Day-of-Week: This strategy is used when there is inherent DOW trend. It is used when client has specific business constraints (especially over the weekends).
Monthly: This spend strategy is used when client has monthly budgets for optimization. It is easy to use for monthly budget use cases and spend may be higher or lower at the end of the month.
Day of Month: This strategy is used when clients have a pronounced monthly trend. It has Strong weekly trend and enough historical data to allow Adobe Advertising Cloud to create day of the month factors. It can leverage intra-month trend. It has more budget control but difficult to determine appropriate budgets.
ROI: It has no specific budget constraint but only an ROI target. In this, Client needs to provide revenue information. Its objective function only contains properties pertaining to revenue but the feed should be reliable. In this, the revenue is in the same currency as configured for SE account
CPT: Its Objective is to achieve a certain Cost per Transaction (only orders are in objective function). Else it is similar to ROI.
Marginal CPT: It is the incremental cost that needs to be spent in order to get an additional $1 of revenue.

Portfolio settings:

Portfolio Status: when the portfolio status is Inactive then Advertising Cloud will gather cost/click/impression data for the relevant campaigns for reporting purposes, but won't model the data nor set bids for keywords and ads.

When it is active then Advertising Cloud will gather cost/click/impression data and revenue data for the relevant campaigns and model the data, but it won't set bids for keywords and ads.

When the portfolio status is optimized then Advertising Cloud will gather cost/click/impression data and revenue data for the relevant campaigns, model the data to optimize bids, and set bids for keywords and ads. You can't manually select this option; it is selected automatically after you launch an active or inactive portfolio.

Note: If you remove all the campaigns from an optimized portfolio, the portfolio state is automatically changed to "inactive."

Auto Optimize Bid Adjustment Values: It allows Adobe Advertising Cloud to automatically change bid adjustments in optimized portfolios.

Currently supported bid adjustments are:

Mobile devices for Google, Bing, and Yahoo! Japan
Computer and tablet devices for Google and Bing
Google remarketing lists at the ad group level and
Google location targets at the campaign level.

It defines the preferred range within which Advertising Cloud will set individual bid adjustments for the campaigns or ad groups in the portfolio, for each target. These adjustments are updated daily and recommended to opt in for portfolios that are targeting different devices, remarketing lists and location targets.

Learning Budget: In this, portion of portfolio spend target is used on testing keywords with zero or little historical data. It uses a higher value in this range if there are a lot of keywords with no traffic data and a lower value, in a promotional or tight budgeting period.
Learning Bids: This setting defines the percentage by which to increase bids when no impressions are received for a specified number of days. It is recommended to increase the percentage and reduce no of days during promotion period.
Learning Maximum Bid: For a bid unit that is being learned. The maximum percentage above the portfolio's average cost per click for which a bid can be placed. When you use the default "Conservative | Aggressive" position, the default value is 0%, which allows bids of up to 100% of the average cost per click.

Adjusting the Learning Settings:

When to Increase Learning Options

When you add new bid units.
When the portfolio's cost model coverage or revenue model coverage is low. In particular, bid units with low bids but no cost models are good candidates for zero impression exposure.

When to Decrease the Learning Budget

When the portfolio's performance declines because a large learning budget is diverting more of the portfolio budget to new bid units that don't lead to conversions.

Portfolio settings- Modeling Recency

Modeling Start Date: The first date for which the portfolio's revenue and the click data is modeled. Advertising Cloud can model and optimize past data, but it won't begin placing bids for keywords in the portfolio until the portfolio is launched. If you change the start date to a more recent date, data for the previous dates won't be considered by the cost and revenue models but will still be available in reports.
Model and optimize search on an intraday basis: It allows Advertising Cloud to retrieve click and cost data up to four extra times per day; update cost and revenue models based on detected historical trends; and then set new bids (if required) based on the updated models.
Cost Model Half-life: The number of days before the current date for which cost data is relevant for cost models; the cost model weighs individual cost events, by how recently they occurred according to this value. Generally, use seven days or more unless you expect volatile user behavior. A shorter half-life makes the model more responsive to market trends, but performance will still be more erratic than with a longer half-life.
Revenue Model Half-life: The number of days before the current date for which revenue data is relevant for revenue models; the revenue model weighs individual conversions by how recently they occurred according to this value. This helps control how quickly Advertising Cloud reacts to changes in market conditions. It is also known as "Revenue Recency."

When to adjust Cost and Revenue Model Half-lives

Cost Half Life:

The competitive landscape or market trends change.
The portfolio budget has changed significantly or frequently.
Campaigns keep hitting their budget caps.
Ads or landing pages change.
Seasonal changes occur.
The pricing model changes (such as during sales and promotions).
Abnormal search activity occurs.
The search engine or ad network algorithms change.

Revenue half life:

Ads or landing pages change.
The pricing model changes (such as during sales and promotions).
The competitive landscape or market trends change.
Revenue data is delayed.
Revenue data is missing because of feed data or parsing issues.

Portfolio Settings - Excluding and Including Dates in Forecasts

Exclude data from cost models: You can exclude cost data for specific dates from the cost forecasts used to optimize keyword bids. By default, no dates are excluded from exclude data from cost models:
Exclude data from revenue models: By default, no dates are excluded from exclude data from revenue models:

Portfolio Settings – Limits

Min Bid: The minimum bid for bid units in the portfolio.
Max Bid: The maximum bid for bid units in the portfolio.