How To Manage Really Really Large Paid Search Campaigns
To answer this question, I commissioned my friend George Michie from the Rimm-Kaufman Group (RKG), one of the most respected paid search agencies in the SEM world, working with many advertisers of this size. George very graciously gave such an in-depth answer, we decided to turn it into a blog post and share it with you all. Thank you, George!
Question: "How do you maintain a keyword list when the SKU count is large and the product turnover is high?"
The answer is: "Very carefully."
We manage campaigns for a number of companies with over 100,000 SKUs and huge product turnover rates. These folks end up with programs with millions of unique keywords and it is incumbent upon the paid search manager to make sure that the keyword list:
- Has coverage of new products;
- Ties keywords to the optimal landing pages;
- Has targeted ad copy; and
- Is purged of keywords as the products they reference drop out of stock
This is no small feat and requires smart tools. We've developed our own tools and systems for helping RKG's analysts do some of the heavy lifting using a clean product feed.
Core Keywords: category, subcategory, non-product specific head, torso and tail.
The first and most pressing order of business is to make sure coverage of the head, torso and the top of the tail is well done.* We advocate a careful, machine-aided but human-driven process that starts with landing pages and works backwards. Building out keywords, assigning landing pages and writing copy for these core terms is of paramount importance and this piece needs to be separate from, and to an extent protected from, the ongoing product-level keyword process that follows.
*Linda's note: Head, torso and tail refer to the pieces of the bell curve. In the context of paid search, the "head" consists of the keywords that send the highest volume of PPC traffic to your site, individually. The "tail" is made up of less popular terms that might occur only a few times per year but collectively can add up to a big chunk of your keyword referrals. The "torso" is all the good stuff in between. For example, Nike might have the head term "nike shox," the tail term "D width mens cross trainer gray suede upper" and the torso term "mens d width cross trainer."
Product-Level Keywords: SKUs, product names with and without manufacturer names, etc.
Building out huge lists of permutations on product names, SKUs, manufacturer names, feature lists, etc. is trivially easy, but doing it well requires a more sophisticated tool set than most folks can easily access. The tools must:
- Control for requiring modifiers. Imagine you sell Halloween costumes. If you bid on keywords like Harry Potter, Mickey Mouse and Playboy Bunny without key modifiers like "costume," "suit," "outfit" etc. would be disastrous - your ads would show whenever these keywords are searched, regardless of the context, diluting your click through rate and attracting irrelevant clicks. The tool must provide the ability to automatically weed out dangerous phrases that are missing those modifiers.
- De-dupe against core keywords. Because we've put more time and oversight into building the core keyword list than a simple machine can match, we should give preference to those keyword-landing page-copy combinations than to phrases generated by a product feed. Keywords generated through this process that match existing keywords should be discarded.
- Make smart landing page choices. Most product specific search terms should land on product specific pages, however the process described above will create many keywords that aren't product specific. Because similar products may have like feature sets, keyword development tools will create many duplicate phrases. Most of the tools on the market will de-dupe internally but will simply grab one of the product URLs at random for the landing page. This is a problem. It tells the user that you only have one product that responds to their search when in fact you have many. By not showing the user the full-range of choices you lose opportunity every time. Whether the search results page on the advertiser's site or a category or sub-category page is better will depend, but a process needs to exist for a smart analyst to provide guidance. A machine will get this wrong every single time.
- Assign copy, and cluster appropriately. A dynamic process for clustering related phrases and gluing on appropriate ad copy is also important. More targeted ad copy yields better Quality Scores (which is largely influenced by click-through rate), and smart clustering provides important clues for a sophisticated bid management system.
- Identify other potentially high-traffic, ergo dangerous, keywords. The system should output potential new ads in a format that ranks the terms by likely traffic volume. This helps prevent overly broad keywords, studiously avoided in the core buildout, from creeping in through the back door.
Hygiene: Now the keywords are out there, how do we know when they should be dropped?
First, it's important to note that in some categories, advertising on out-of-date models can actually be quite profitable, provided that the landing pages handle those "upgrades" well. If products have been replaced by newer models, you may benefit from advertising on both as long as the user ends up on a page with a good alternative product. Testing can reveal whether this works for your business.
Second, recognize that because the process for generating keywords generates both product specific and more general keywords the connection between products that drop out of inventory and the keywords generated from those products isn't an exact match.
Finally: consider whether "Out of Stock" is generally a temporary condition or a more permanent reality. If the process pauses keywords that are out of stock, but two days later the products are back in stock, is there a way to identify those and flip them back "on"?
RKG uses a system to flag products that drop out of stock or out of the product feed. We also have built custom flags for certain clients where inventory becomes thin in ways that impact conversion rates, e.g.: "we still have men's swimsuits but only for waist sizes less than 24" and greater than 52"." Much depends on the quality of data provided by the advertiser in the feed.
We couple this with a URL-checking system that flags landing pages tied to search keywords that contain page-load error messages or "out of stock" messages.
What to do with the output.
This is tricky business. It is tempting to automate the whole process from keyword creation to posted ads to paused or deleted ads as the products drop out of inventory, but experience has taught us the value of human review.
We recommend baking in some level of analyst oversight before any ads go live and before any ads are paused. While this does create a lag time between inventory shifts and keyword list changes, the cost of a single big mistake can greatly overshadow lost opportunities in the deepest regions of the "tail."
What if we don't have these kinds of tools?
Doing all of this work manually adds greatly to the cost of the keyword list maintenance. This can change the cost/benefit calculation quite a bit and may mean that this level of housekeeping can only be done every other month, or quarterly. Much depends on how important the tail is to your business.
High product turnover is a paid search management challenge, which, through well-regulated use of smart tools can be handled effectively.
You can read more of George's blog posts on the Rimm-Kaufman Blog.