“Data Driven Thinking” is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is by Andrew Casale, VP of Strategy at Casale Media.
Plenty of ad exchange buying patterns just make sense. There’s increased demand over the holidays, for instance, when more consumers are shopping, and retail – with its access to rich audience data – is the leading category overall. But there’s one less obvious pattern in the real-time bidding marketplace that recurs month after month like clockwork, but that isn’t a response to the way consumers shop. Instead, it’s an inefficient product of human intervention in digital marketing and the media planning calendar. And it’s costing marketers money.
Across the RTB marketplace, demand regularly ramps up throughout any given month, typically peaking in the last few days of a month, with an even bigger peak at the end of each quarter. On the first day of every month, bidding tends to scale back by approximately 15-20% from the previous day.
Considering the sophistication behind the tools and platforms that execute programmatic buys, it’s clear that this significant decrease in activity is the result of external intervention over what should otherwise be a fairly automated buying channel. When a marketer manually stops or reduces bidding based on an arbitrary date – the first of the month – it means that much of the learning established through this buying model are effectively being disregarded. That’s concerning given that the success of programmatic buying relies heavily on machine learning based on targeted data.
Because of all of this, campaigns regularly start later than they should, then spend faster than desirable in order to deliver on time, only to repeat the process again the next month.
To share some more specific data, in the first quarter of this year, the peak average winning bid price – the highest price that won an impression in an RTB auction – occurred on Jan. 25, Feb. 25, and March 29. This pattern is entirely due to the greater budget availability toward the latter half of the month, ratcheting up impression prices when demand is arbitrarily highest.
March was fairly typical of what happens month after month. Here’s a snapshot of how clear prices rose week over week in March:
|Week||Clear Price (Indexed)|
|March 1-7, 2013||94|
|March 8-14, 2013||91|
|March 15-21, 2013||96|
|March 22-28, 2013||104|
|March 29-31, 2013||111|
To be clear, I’m not suggesting that media planning is useless in the programmatic era, nor that the calendar has been rendered irrelevant. There will always be product launches, key seasonal periods deserving of heavier activity and increased marketing budgets during key shopping periods.
But while strategic executions like the above examples warrant a well-orchestrated flight, I would argue that the rest of the time, executions should not be constrained to fit into a monthly, calendar-driven buying model. Instead, machines should be given more autonomy to optimize budgets to deliver the best value. This is, after all, how most search budgets are allocated, and I think it’s about time RTB benefits from the same budget efficiency and follows suit.
Buyers not taking advantage of an “Always On” planning model are missing potential sales opportunities for their brand. As a marketer, if you’re confident that your buying strategies are well-tuned, you should consider blurring the lines between months. A consumer who was pegged as a strong prospect to consider your product on May 31 will be no less interested on June 1, so your message to that consumer shouldn’t arbitrarily change either.
While the marketplace remains in a state of end-of-the-month price ramping, smart marketers who can change their monthly pattern can avoid falling into the end-of-the-month trap and take advantage of better pricing earlier in the month. Relying on the calendar to govern the intelligence of programmatic technology is clunky and costly. It’s a prime example of where human decision making should be tested against machines, and where machines are likely to do better.