Understand your Search Lift measurement data

If lift is detected, you’ll get access to the following Search Lift metrics:

• Relative Lift
• Normalized Incremental Searches Per Impression
• Normalized Incremental Searches Per Cost 

Search Lift studies last for 28 days, however you can get Search Lift results earlier depending on the amount of impressions in your measured insertion orders. The more impressions running, the quicker you will get results and the highest chance you will have to measure statistically significantly positive lift results. Search Lift results are only released if a study meets privacy thresholds, which require minimum impressions and search volume. You have a higher likelihood of meeting the privacy thresholds by following the budget recommendations and keyword recommendations. 

Progress	What you will see in Display & Video 360
Less than privacy minimums	“Not enough data”
Greater than privacy minimums	If there’s statistically significant, positive lift, we’ll report it. If not, we’ll report “No lift detected”.

Statuses

“X% Relative Lift”

An X% relative lift indicates that we've collected enough data to detect lift with high confidence. For example, a 25% increase in relative lift indicates that users exposed to your ads are 25% more likely to search for your brand or product. 

“Not enough data”

“Not enough data” means that, based on the date range you’ve selected, the amount of ad impressions and/or the number of searches made in that date range is below the minimum threshold required to surface results.

Troubleshooting “not enough data”

There could be multiple reasons for not getting enough data for your study or an individual slice. To fix it, make sure that you wait until the end of your measurement and make sure to spend your budget in full. Campaigns spending less than their budgets are a common issue leading to “not enough data” but if you’re spending enough but are still not getting Search Lift results, check for the following: 

Is your bid too low?

Low traffic can indicate that you're getting outbid. Raise your bid to win more impressions and generate traffic. However, keep in mind that, if you raise your bid, you’ll spend your budget faster.

If you have an unlimited budget, “bid too high” isn’t an issue.

Recommendation: If your traffic is low despite broad targeting, consider raising your bid. If raising your bid means you are hitting your budget cap, consider raising your budget to accommodate the higher bid.

Is the campaign targeting too narrow?

The following study and campaign setup configurations may sometimes reduce the number of impressions your study will be able to gather. The extent to which they slow down impression count varies depending on the degree to which they’re narrowing your targeting reach.

Audiences (particularly retargeting), placements, keywords, and topics

More restrictive targeting types, such as placements, keywords, and retargeting, reduce the number of eligible viewers and can lead to less impressions. Less impressions and less viewers in turn mean that there's less potential for users to search for your search terms.

Small geography

 Too small of a geography might limit unique viewers, which reduces your odds of getting enough impressions. Ideally, studies are run at the country level, but you can also target smaller geographies, as long as there’s a large enough population.

Recommendation: Monitor your traffic closely as the study is progressing. If you aren’t spending in full, broaden any overly restrictive targeting by expanding geography or removing overly restrictive targeting types like placements or keywords.

Have you selected the right search terms?

Search term selection is a critical part of Search Lift study configuration, and poorly chosen search terms are a common cause of unsuccessful measurement. The goal is to capture the searches relevant to the ad campaign, while excluding unrelated queries. It is therefore recommended to prioritize specific and focused keywords which are related to your product or brand or ad campaign—they tend to outperform broad search terms. However if you select a search term that is too specific it might not reach enough searches to meet the minimum amount of searches that we require to share results. In that case, try a more broad keyword next time, but be careful to not go too broad either as you risk getting “no lift detected”

For more details on how to select your search terms, learn more about setting up Search Lift measurement or reach out to your account representative. 

“No lift detected”

Sometimes a study with enough impressions will show “No lift detected”. First of all, we recommend waiting until the end of the study to get a chance to find results as the resolution of the measurement will improve over time. If there still isn’t lift by the end of the study, it means that we couldn’t observe a statistically significant difference between the search behavior from users who saw your video ad and those who didn’t. If you don't have lift at the study level, check if you have lift in specific segments, for example, search term, age, gender, campaign or device. Consider focusing on those segments with positive lift. 

As with any media channel, some metrics are more difficult to move than others. Some audiences are more difficult to reach than others. It’s normal for studies to have no lift on certain metrics and audiences.

In addition, consider retesting with an improved setup. Retesting can also help confirm slice level results, which can have lower reliability due to cherry picking.

---

 Search Lift metrics

Metrics:

• Relative Lift
• Incremental searches (indexed) per impression
• Incremental searches (indexed) per cost
• Incremental searches (indexed) when slicing by search property
• Incremental searches (indexed) when slicing by search term
• Confidence interval

Relative Lift

Relative lift reports the percentage increase in search volume caused by your ads. This is calculated as (absolute increase in searches) / (baseline number of searches). For example, a relative lift value of 50% would mean that users exposed to your campaign were 50% more likely to search for your product or brand than those who weren’t. 

Incremental searches (indexed) per impression

Incremental searches (indexed) per impression illustrates whether impressions from a given slice, for example, age group or gender, performed above or below your study’s average in driving searches. This indicates the effectiveness of a given slice. Incremental searches (Indexed) per impression greater than 1 is above average. Incremental searches (Indexed) per impression below one is below average.*

Example

Campaign A = 1.44

Campaign B = 0.72

Overall = 1

Interpretation

Campaign A performed 1.44x better at increasing a user's propensity to search per impression than campaign A and B together (overall result)

Campaign B underperformed when compared to your study average in terms of driving search behavior per impression since the incremental searches (indexed) per impression is below the overall result.

The “Overall” Slice will always show “1” as the incremental searches (indexed) per impression metric, as this is the baseline value used to index and compare all other slices.

Incremental searches (indexed) per cost

Incremental searches (indexed) per cost illustrates whether ad spend from a given slice (for example, age group or gender) performed above or below your study’s average in driving searches. This indicates the cost-effectiveness of a given slice. Incremental searches (indexed) per cost greater than 1 is above average. Incremental searches (indexed) per cost below one is below average.*

Example

Campaign A = 1.05

Campaign B = 0.97

Overall = 1

Interpretation

Campaign A performed 1.05x better at increasing a user's propensity to search per unit of cost than campaign A and B together (overall result). 

Campaign B slightly underperformed when compared to your study average in terms of driving search behavior per unit of cost.

The “Overall” Slice will show “1” as the normalized incremental searches metric, as this is the baseline value all other slices are then compared to.

Incremental searches (indexed) when slicing by search property

Incremental searches (indexed) illustrate the contribution of a search property (for example, YouTube or Google Search) to the overall lift of a specific search term. This indicates the effectiveness of a given property. Incremental searches (indexed) by search property is expected to total to 1.

Example

Google.com = 0.55

YouTube.com = 0.45

Interpretation

Google.com contributed 55% of the total incremental searches for the selected search term.

YouTube.com contributed 45% of the total incremental searches for the selected search term.

Incremental searches (indexed) when slicing by search term

Incremental searches (indexed) illustrate the contribution of a search term to the overall lift of the search term group. This indicates the effectiveness of a given search term. Incremental searches (indexed) by search property may add up to 1, but a search by a user in the study may apply to multiple search terms. For example, an ad for the Google Pixel which included references to the camera and a feature called Magic Eraser. A search for Google Pixel Camera would apply to the search term Google Pixel and Pixel Camera. Therefore, both of those terms will get credit for the search.

Example

Google Pixel  = 0.91

Pixel Camera = 0.55

Magic Eraser = 0.22

Interpretation

Google Pixel was included in 91% of the total incremental searches for the search term group.

Pixel Camera was included in 55% of the total incremental searches for the search term group.

Magic Eraser was included in 22% of the total incremental searches for the search term group.

One can know from these results that this ad drove the most incremental searches for the brand name “Google Pixel”. Between the 2 features highlighted, the ad drove more incremental searches for the Pixel Camera when compared with Magic Eraser.

Confidence interval

Based on your experiment data, we build a range of possible values for your lift metric that we find likely; this is called the confidence interval. For example, if you notice a relative lift of 35%, this is our best estimate of your lift and is called the point estimate. We also might share that our confidence interval for your relative lift goes from 30% to 40%. This means that there is a good chance that your relative lift is anywhere between 30% and 40%. As a lift study collects more data, the confidence intervals become smaller, reflecting the fact that your estimates are getting more precise and reliable.

---

Troubleshooting

Below are a few things you can do to improve your campaign’s set up, creative or targeting to increase the chances of noticing lift.

Set up your study correctly

The most important thing is to properly set up your study. If your creative was focused on a specific product, and you measured the impact by using the brand name, you’ll likely have “No lift detected”. It’s common when a brand is very well known that it’s unfortunately hard to detect lift on the brand term. While you should wait for your next campaign to measure again, we encourage you to use more specific terms in addition to the brand term, if possible, as part of another search group, to increase your chance of lift.

Improve your creative

Quality of the creative plays a huge role in getting lift. Check if your ad is following the ABCDs of effective YouTube creative. Contact your account manager for detailed guidance on improving your creatives.

For light-branded ads, if your brand or product name isn’t present, appears late in the ad, or is too subtle, the audience won’t attribute the creative back to the brand or product advertised. To correct this, consider adding branding, like an icon, watermark, or banner, earlier in the ad. You can also change the script to integrate the brand or product more clearly.

Limit exposure to your creative outside the study

If a creative has been noticed by viewers before the Search Lift study has been launched, it’s possible that the control group or a group that doesn’t view the ad, has been contaminated, resulting in no lift. This will make the control group respond similarly to how your exposed users respond, thus reducing “lift”. To minimize creative contamination:

• Avoid running YouTube campaigns with non-YouTube channels like TV and other ad platforms at the same time as your test.
• Avoid running other video campaigns with a similar creative unless you add these campaigns to the lift study too.
• Set up and activate your study before campaigns have started.

Target the right viewers for your campaign

Sometimes a campaign may not be tailored to the target audience. Watch the creative and ask yourself: is this message relevant to the target audience? If you have lift on certain segments, consider targeting them only with that creative. For the other segments, consider tailoring the creatives.