The holidays can wreak havoc on inventory avails and prices, especially during the all-important fourth quarter. Americans look up more recipes in the days leading up to Thanksgiving, and they buy more stuff leading up to Christmas. These patterns in traffic and conversions show up on the open exchanges.
During Q4, the stakes are particularly high for advertisers trying to suss out these patterns. Advertisers up their bids and budgets to attract audiences, just as consumers change when, what and where they consume content.
The first bonanza holiday in Q4 – in terms of traffic and CPMs – is Thanksgiving. For one, there are certain times consumers don’t show up on the Internet.
For Scripps Networks Interactive, with sites like the Food Network, Thanksgiving is the No. 1 traffic day of the year. It’s also highly in demand from advertisers.
Scripps uses its years of historical data to predict how many impressions it will serve in the days leading up to Thanksgiving. Advertisers who want to ensure access to relevant, holiday-themed content will buy direct or set up a priority look at inventory through private marketplaces.
About.com, which has 2 million pieces of content, also sees a bump in food content consumption around turkey day. Some other less obvious content verticals such as parenting articles also trend around this time, as more time spent with the kids inevitably leads to a desperate quest for family advice.
“The shape of the pattern is the same every year,” said About's head of data science and audience development, Jonathan Roberts.
CPMs climb well before Thanksgiving. Starting Nov. 12, Amnet sees a 10% to 25% increase in average CPMs, and a 50% spike above average on Black Friday and Cyber Monday.
“The irony is the largest budgets are at the time when it’s most expensive to buy,” said Amnet platforms head Chris Romano.
But because more consumers are in a shopping mindset the last couple of months of the year, campaigns still perform better.
On AppNexus’ exchange, people are three times more likely to convert during the fourth quarter, with the average conversion rate going from .1% in Q1 to .3% in Q4.
The ratcheting up of ad spend halts on Christmas, when advertiser spend and Internet traffic falls off a cliff.
“On Dec. 25, traffic in the morning was virtually nonexistent in the US,” said Sovrn's Knapp. “We saw a huge spike in traffic at 4 p.m. Presumably that’s when everyone gets bored and goes online.”
CPMs drop on Christmas and remain low through January. But people still buy after Christmas.. AppNexus showed conversion rates ranging from .07% to .09% from Dec. 23 to Jan. 2, when CPMs are at their lowest.
Among other holidays, the Fourth of July and Memorial Day both also bring big traffic dips. According to AppNexus, traffic in the US dipped 25% compared to non-US traffic on July 4.
For these warm holidays, consumers aren’t just away from the Internet. They’re not shopping, either.
“There are holidays like a July 4 or Memorial Day, where the performance is not there unless you have a promotion at that time,” Romano said.
During holidays desirable to brands, advertisers will obviously pay more to get near the hot content. MediaMath observed that private marketplace deals rise 20% to 30% around special deal packages for holidays like the Super Bowl and Valentine’s Day.
“For something like a March Madness package, we negotiate these packages well in advance and offer them to clients,” Amnet's Romano said.
But not all advertisers buying around holidays want to be next to timely content. They just want users before the buying window closes.
“Competition becomes fierce to reach the right consumers on Black Friday or Cyber Monday because the sales cycle is very short," said Tanuj Joshi, senior director of strategic media enablement for MediaMath. "That’s when top brands are willing to pay top CPMs for securing those users.”
Holidays aside, daily digital rhythms factor into how advertisers reach consumers.
Traffic day by day looks like a smooth sine wave, dipping when people sleep and rising as they wake.
Many advertisers like to reach users as they’re logging on first thing in the morning, when traffic volume is rising.
“The assumption is that advertisers are trying to grab users’ attention first thing, to get that first impression of the day,” said Bill Murray, VP of sales planning and strategy for Scripps Networks Interactive. On Scripps’ family of sites, the heaviest programmatic activity occurs between 8 a.m. and noon Eastern.
MediaMath sees two peaks in spend. During the first, between 10 a.m. and noon, traffic and CPMs peak at 60% to 80% above the early morning hours (2 a.m to 5 p.m.). That spend tapers off until the evening, Joshi said, when there’s a second peak of people going on the Internet from home, from 7 p.m. to 11 p.m.
Weekends see lower traffic, and a higher percentage of mobile traffic, than weekdays. Some advertisers avoid the weekends because they don’t see as many conversions.
“Weekends are never great performers for hotels, because people are in the hotels on the weekend,” Romano said.
Budget Dumps And Late Creative
Media agencies have money to spend and deadlines to spend it. The end of months and quarters usually bring a spike in CPMs and spend as agencies try to unload budget.
Likewise, the first days of a new quarter can also be quiet as campaigns get off the ground.
Publishers like Scripps let in more programmatic demand at the beginning of a quarter, and less at the end. That’s in part because some direct-sold campaigns don’t always go live at a campaign's start date.
“Late creative influences a lot of programmatic spending,” Murray said. “If a direct deal is set to start at the beginning of the month, and the creative comes in four days late, those were impressions that you didn’t sell.”
Programmatic, in this scenario, provides an improvement from the past for publishers. “At least you now have an effective way to monetize those impressions,” Murray said.
AppNexus data shows that CPMs rose 12% from the beginning to the end of the month in April 2014, 4% from the beginning to the end of May and 22% from the beginning to the end of June, the end of most Q2s, proving out Scripps’ observations.
Advertisers wanting value should increase programmatic buying at the beginning of the month to take advantage of these volume and price benefits
Slave To The (Tech) Rhythm
To buy impressions programmatically, DSPs have to follow some kind of logic. Enough bidding platforms are following the same rules that patterns emerge, creating inefficiencies. For instance, advertisers will pay more for impressions served at 3:01 p.m. than 2:59 p.m.
“Yield on impressions tends to be higher top-of-the-hour, and decrease through the hour,” said Sovrn's Knapp.
The same pattern plays out on a daily basis. Impressions tend to clear at higher prices at 12:01 a.m. than they do at 11:59 p.m., because 24-hour frequency caps reset then.
Over time these dynamics will fade as bidding technology improves.
“Two years ago, something as simple as dayparting to exclude certain hours, like 1 a.m. to 6 p.m., you would have to do that every day,” Romano said. “Now, we have customized dayparts. Not only can I run different hours of the day each day or the week, but even within those hours, I can exclude certain inventory.”
The idea of resetting frequency caps at midnight is falling out of favor, for example. Some advertisers now reset frequency caps in the early morning, to ensure that they won’t waste impressions late at night reaching someone who isn’t in a buying mindset.
And ideally, technology should be able to make smart decisioning no matter the time of day. “If [the DSP] has to bid really high at 3 a.m. because that user is about to convert, then why not?” Joshi said.
But the humans manning the algorithms will always want to find new ways to make them even smarter and take advantage of consumer traffic and buying patterns. “The more tools and levers we have to pull, the better performance we see,” Romano said.
Romano already has his eye on the next frontier: patterns related to geofencing. “Where you are now, yesterday, tomorrow – that’s a lot of information and intent, and that’s the next data set we’re going to have trends on.”