I've written on the importance of a basic search marketing audit every six months. This is a four step process you or your team can accomplish in an afternoon. Early posts covered the PPC sales audit and the PPC cost audit. Today's topic: the pay-per-click bid audit. From the outside, you can consider a paid search bid management algorithm to be a black box.
All posts in the PPC Audit series:
A paid search bid management algorithm receives primary inputs -- historic data on PPC sales, costs, bids, ranks, CTRs, and impressions -- and secondary inputs -- seasonal trends, time-of-day and day-of-week trends, and marketing events like catalog drops, product releases, and stockouts, arrivals and departures of competitors, etc. The algorithm also receives your marketing objectives like maximizing revenue subject to a budget constraint or hitting a target metric like A/S, ROI, ROAS, CPO, or whatever. From these inputs and objectives, the algorithm generates bids, a time-series of smart (one hopes) maximum bid CPCs for every individual ad (versus ad-group), capturing the theoretical true value of each inbound click. A robust bid management system uses economics and statistics to set bids, rather than just maintaining ads in certain positions. Positional bidding destroys value, as my colleague George Michie pointed out last week.
You can spot-check a bid system without seeing the guts of the black box. For example, RKG's bid platform uses statistical optimization to maximize the ad portfolio, an algorithm with roots in my doctoral research in optimization at MIT.
Other folks may take other approaches.
What ultimately matters is performance. Even if you can't inspect the algorithms inside the black box, you can still do some rough checks to gain confidence in your bid management process. The PPC Bid Audit Here are nine steps for a ppc bid audit:
- Ask your search agency or your in-house search team to describe their algorithm for setting bids. While they may not disclose every detail, they should be able to sketch out the general approach, and that approach should make sense to you.
- Ask them if they're bidding at the adgroup or the keyword or the ad level. You want at least keyword-level bidding (eg by-keyword-by-engine). Ad-level tracking and bidding (by-keyword-copy-url-engine) is even better.
- Ask them how they handle ads with costs but no sales.
- Ask how they handle periods of great change -- for example, going into and out of a holiday season.
- Choose two of your highest click, highest cost phrases. Pick another phrase which has incurred a medium level of cost and some sales. And pick a phrase which has incurred a low level of cost and no sales. As much as possible, select 5 phrases which correspond to 5 single ads -- that is, avoid choosing phrases scattered across multiple campaigns and adgroups due to tests or other projects.
- If the same algorithm is handling all the engines, it is likely safe to audit bids on one engine. Pick Google.
- For the last two weeks, for each of the 5 phrases in Step 1, from your search marketing agency or from your in-house team, get daily sales, costs, clicks, impressions, and average ranks for Google.
- If you doubt these numbers from #4, here's how to do a PPC sales audit and a PPC cost audit.
- For the last two weeks, for each of the 5 phrases, obtain a bid history from Google (Tools >> My Change History >> CPC). This is a list of all bid changes, with a date-time stamp, and old bid, and the new bid.
- Graph these five bid trajectories by day.
- Using these graphs,
- Check the frequency of bid changes. If the bids aren't being changed at all, that's a something to investigate.
- If you see intra-day bid changes, check if these changes follow a consistent and logical pattern across the day, or across the week. That's a good sign, suggesting the bid algorithm is exploiting a day-part or week-part effect.
- Check if the bids on the two high click terms fluctuate wildly and unpredictably. That's a bad sign -- bid trajectories shouldn't look like random walks. High click terms provide a good algorithm with sufficient statistical power to make informed estimates of about click value.
- Check if the bids seem reasonably correlated with sales. Caveat -- if you sell more considered products (higher cost, more complex, etc), there can be a significant lag between the paid click and the resulting order, which can lead to the click and the sale occurring on different (but usually adjacent) days.
- Check that bids make marketing sense for the ad phrases during this time period. For example, if one of your five phrases was, for example, "ruby necklace", and if your firm started or ended a ruby sale during this two week window, and if you intended your bidding system to reflect this sale (exogenously, eg. your marketing team wanted that, rather than endogenously, eg. based on changes in "ruby" conversion resulting from the sale), then you should see that desired increase or decrease at the appropriate time.
- Considering the four ads with sales, check the profitability of each of the four by day, by week, and overall. If you are aiming for, say, 25% ad-to-sale ratio (that is, each ad dollar must generate four dollars in tracked sales), the ads should very roughly be accomplishing this 25% A/S goal. Due to statistical variability, the daily or weekly numbers may fluctuate from this target, particularly on lower volume ads.
- Look at how the system is bidding the sample ad with no sales.
- Check if the system is using updates intelligently. It makes no sense to make small changes or overly frequent changes on terms which get effectively no traffic.
- Pick two or three specific bid changes for the high volume ads.
- Ask your agency or in-house team to explain the logic behind those specific changes. ("On April 2nd, 'widget' went from $0.65 max bid to $1.02 max bid, and stayed there until April 4th. Can you walk me through the logic of that change?")
- You should expect a reasonable explanation backed up by specific sales and cost statistics.
All posts in the PPC Audit series:
Join the Discussion