In our previous post, Getting to know AI and how it can benefit your campaigns, we gave you some insight on what artificial intelligence (AI) is, and why it’s important for advertisers to have access to AI for their digital campaigns. On April 29, StackAdapt was at Programmatic I/O, where I shared 10 questions to ask your DSP—so you can be sure your campaigns can leverage AI effectively. In case you didn’t attend, I want to share those questions, and why it’s important you ask them.
1. How accurate is your AI?
Algorithms are not all made the same. For example, in the context of CTR prediction, a human based logic that layers on historical site domain data, on top of device data, exchange data, and geo data will still work well. However, it’s prediction potential is limited compared to the AI algorithm, which is able to find hidden patterns and signals in thousands of other attributes that the non AI model cannot. See the graph below for the comparison of the actual results versus the predicted results of the two models.
For this reason, you want to be sure your DSP is using an AI model that is accurate.
2. Is AI optimizing your bid prices?
Bid price optimization is a key algorithm regardless of the KPIs of your campaign. Whether it be brand awareness or direct response, you should always be bidding to the perceived value of your user—and each individual user is unique—so AI is important. If you bid a flat rate, both publishers and exchanges will be able to take advantage of extracting revenue from you with their yield optimization. With sophisticated price optimization (using AI), the auction environment (first or second price), has no effect on the outcome as we are always bidding to the exact value.
3. Can your AI optimize multiple KPIs?
It’s one thing to know that the AI is optimizing toward a single goal—but if the AI can optimize toward multiple goals in tandem, you can really reach the best campaign outcomes. If your DSP can optimize using a single KPI, that’s great. However, watch out for your DSP negatively affecting other KPIs while doing so. Ask your DSP about layering KPIs during optimization—that will begin to demonstrate how strong their AI capabilities are.
4. Does your AI adapt to market conditions?
Market conditions are always changing, and to be accurate, the AI you’re using needs to adapt to those changes. For instance, if there are changes to the dollar, and conversion rates drop, you may see an influx of online shoppers. The AI should adapt to the increase in traffic, resulting in changes to bidding and frequency of ad placements.
5. How often do you train your models?
Practice makes perfect. If you want to be a master chef, you need to get in the kitchen and cook often. AI is no different. AI models need to be trained often, to make them more adaptable to changes in traffic and inventory, and ensure higher accuracy. Your DSP should be training their models frequently so you get the best results.
6. How quickly are they trained?
Timing is of the essence. AI training for months at a time and hoping it grasps the material is not helpful when the industry and the programmatic landscape is constantly changing. The AI needs to be quick enough to pick up on changes as they happen—otherwise, they will be obsolete and no longer relevant. If the information is already out of date, the AI will just need to be retrained. You need an AI that is trained quickly.
7. Do you train more often for special occasions?
During special occasions traffic can increase drastically. Consider Thanksgiving, and one of the biggest online shopping days, Black Friday. Your AI needs to be prepared for that change. You want to know that your DSP is taking these special occasions into consideration, and ensuring the AI is trained enough during these high traffic events.
8. Does your AI detect fraud or poor quality traffic?
To be sure that your campaign is actually performing well, your AI needs to detect fraud and poor quality traffic. It may appear as though your ad is getting a ton of clicks, but really, 73% could be poor quality, fraudulent, or duplicate requests. Without the AI detecting this, your metrics will be skewed and you will not be able to achieve the KPIs you’ve set for the campaign.
9. Do you have an A/B testing system in place?
Because AI models must update themselves constantly, it is dangerous to put them into practice immediately after having trained in a historical environment. Your AI must prove themselves gradually through a robust A/B testing system. This will ensure optimization will actually increase performance over time, rather than the other way around. A strong engineering backbone to incorporate a learning cycle is critical to a mature AI system.
10. How do you see AI evolving in your DSP in the future?
This question, although interesting, does not really dictate the effectiveness of AI. Rather, it’s more of a question of how active or proactive the DSP is in the evolution of AI. There is no doubt that AI technology is improving year over year, with less need for manual hand-holding of campaigns as we go along. Hopefully, this is a shift that you can take advantage of, and eventually focus on more of the creative and strategic aspects of marketing.
I hope these questions shed some light on how your DSP is using AI—or should be using AI—for your benefit. Explore more about how StackAdapt is using AI to provide contextual targeting for your campaigns.