9 Keys to Successful Day-Parting
Taking advantage of intraday fluctuations in traffic value can be important to maximizing the efficiency of your pay-per-click (PPC) ad spend. We see statistically significant variances in traffic value on the order of 5% to 25% for many retailers, and much more than that for some. However, measuring and acting on these value differences requires […]
Taking advantage of intraday fluctuations in traffic value can be important to maximizing the efficiency of your pay-per-click (PPC) ad spend. We see statistically significant variances in traffic value on the order of 5% to 25% for many retailers, and much more than that for some.
However, measuring and acting on these value differences requires careful thought and proper execution.
Value, not volume: As you do the research, make sure you’re looking at the quality rather than quantity of traffic. The fact that more traffic is available at certain times during the week doesn’t mean you can afford to pay more for it. Ideally you should analyze margin dollars per click as a function of day of week and time of day. Sales per click or conversion rates (orders per click) may be acceptable proxies for value.
Analyze the right traffic: You want to study the value of the non-brand PPC traffic coming into your site isolated from all other traffic. Direct load and brand traffic behave very differently than competitive PPC traffic, and since the latter is all your PPC program controls that’s what you need to study. Also, make sure your click times and order times are from the same time zones!
Credit sales appropriately: Remember your cookie windows! Many of the conversions, hence much of the value, comes long after the click. It is a mistake to divide the PPC margin that flows in from, say, 9AM – 10 AM Monday morning by the clicks that came in during that period. For most retailers half of the orders placed during that period were driven by clicks that happened more than a day earlier. The proper way to do this is to credit the value to the time of the click that got the credit.
Balance data volume and data recency: You’ll need to pull in data over a period of weeks to smooth out statistical noise and have enough data to parse usefully. How many weeks? This is tricky and depends on your business. Usually 6 to 12 weeks makes sense, but the farther back you go the less relevant the data is to today, particularly if your product offerings are seasonal and/or your clientelle varies by season. Go back too far and there is trouble, too little and you may not have enough data to act upon rationally.
Carve out the exceptions: Understand that sale events and holidays can totally wash out the normal time of day and day of week dependencies. Exclude data from these periods from your analysis so that “Black Friday” doesn’t get your systems too excited about the average Friday. Also, as these events approach, think about suspending your dayparting rules to avoid zigging when you should be zagging.
Don’t slice too thin: It’s tempting to try to gild this lily by parsing the data more finely than you really can. Each time-bucket should have at least 100 conversions to have a true sense of the traffic’s value. It may sound cool to say that your system tweaks the bids 24 hours a day 7 days a week associated with day-parting. Very few advertisers have enough data to accurately assess traffic value in 168 buckets without going so far back in time that the data may be irrelevant.
Creative bucketing: One strategy for addressing sparse data is to assume independence of time of day and day of week. It’s not a great assumption, but from our experience it’s not a bad approximation of the truth. What it means is you can calculate the day of week effect, parsing data into 7 buckets, and separately calculate a time of day effect parsing data into 24, or 12, or 8 buckets, and then overlay those effects. This assumption allows you to get much closer to the truth with much less data than you’d need to treat the variables as dependent. Also, there’s no reason for the buckets to all be the same size. You might, for example, parse by the hour during peak traffic, but glom the “wee hours” of the morning all together.
Execution: Make sure the bid changes are actually pushed out to the engines when they should be. This is not trivial. For most systems there is latency between calculation of bids and implementation of them. Make sure you understand that latency and factor it into your process.
Wade in, don’t dive in: Bear in mind, we don’t see all the value we’re driving. People search at home at night and then buy on their work machine’s T1 line. The tracking is on the home machine which didn’t place the order, so the day-parting analysis may think nights are worse than they are. There can be similar impacts on the weekends. There are times during the week when people are more likely to pick up the phone and call an 800 number after getting to a website, which again can make evening and/or weekends look artificially depressed. We recommend ramping up the day-parting factors slowly and watching for “surprising” effects.
Day-parting should not be first and foremost on the minds of SEM managers. Building comprehensive and smart keyword lists tied to the right landing pages, and having a robust, smart bidding system are far more important in the great scheme of things. But when your program is ready to go beyond blocking and tackling, following these keys can generate incremental top line within your efficiency needs.
Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.