Aug 302010

Time is Money in Paid Search

My monthly paid search column for SEL in case you missed it.

In my last column for SEL I wrote about 6 common business fallacies that lead paid search managers to spend time unproductively. Unproductive work is costly in that someone internally or externally is paid to do work that adds no value, and that same person is prevented from doing work that is valuable.

Prioritizing tasks is difficult in the abstract: what is most impact-full for one account will be less important for another. Triage experience dictates that almost any weakness taken to an extreme can be devastating. Lousy, un-targeted, un-compelling ad copy can hurt quality score so much that bid management is irrelevant. Landing all the traffic on the homepage will hurt conversion so much that no amount of wizardry elsewhere will save the program. Account structures that prevent ad copy from being targeted can by itself sabotage performance. If there are just a couple of hundred keywords attempting to cover a product catalog in the tens of thousands, you will not be able to make the program hum until that problem is addressed. No question about it, if the account foundation isn't reasonably solid the priorities below won't make sense.

So, for the sake of argument, let's assume we're talking about a program where some time and thought were put into keyword development and the connection between keywords, landing pages and ad copy is reasonable. The question then becomes:

How should the paid search manager prioritize her/his time?

Here is how we think the priorities should go in tiers:


  • Measure value accurately. The wonky data mining and paid search tips and tricks below are irrelevant if the program doesn't accurately measure the value of traffic generated. In lead generation all leads are not created equally. Establishing a back feed of lead valuation info is the first order of business. For eCommerce sites measuring the margin rather than the sales is crucial if the spread varies across different parts of the inventory. For many there may be multiple success metrics that need to be track to understand the full value of traffic, including email sign-ups, catalog requests, and a host of different advertising revenue metrics for the arbitrage advertisers.
  • Measuring spillover and breakage. For many advertisers, spillover to the call center is an essential part of the equation. Some products and services are more likely to have "clicks to bricks" influence than others. Even the online tracking isn't clean. Javascript tracking drops between 10 and 30% of the traffic it's trying to track.
  • Measure cross channel effects. Those crediting only the last touch are likely to be losing sales to cannibalistic channels. Some folks even mistakenly credit brand paid and natural search (navigational searches) that follow visits from truly competitive marketing initiatives, blinding themselves to the real value of each channel.
  • Think long and hard about cookie windows. The higher the ticket and longer the consideration cycle the more important cookie considerations become.
  • Carefully develop efficiency targets. What is the role of paid search in the overall marketing plan? Is it a "cash machine"? Is it a prospecting vehicle focused on acquiring new customers? Or, recognizing that oftentimes "my" customer is also one of my competitor's customers, is it focused as much on wallet share as anything? Have lifetime value considerations been carefully studied and baked into the above assessment?
    • Study performance by keyword and by clusters. The key to smart bid management is to marry the bids to the value of the traffic coming from each ad. A sophisticated bid management system will run the statistical analysis across all the different ways the data can be logically aggregated or disaggregated to determine the net effect of each variable. Absent such a platform, a smart analyst with a spreadsheet can look for important trends in the data that may help inform how they set bids (eg: "That's interesting, when you look at the combined performance of all the terms that include words like "cheap, discount, bargain, wholesale" they perform as a group very differently than other words in the same product categories!"). Product categories, subcategories, geographies, manufacturer brand names, traffic volume level, keyword length...all kinds of ways to cluster the data.
    • Get away from campaign budgets. If you're hitting campaign budgets regularly you are wasting money, guaranteed. If there are legitimate reasons for budgeting use bidding to hit the target spend. Would you prefer to have 100 visitors at $1 each or 200 visitors of the same quality at $0.50 each? Pretty easy question to answer, yet so many folks rely on campaign budgets. As I argued before, campaign budgets are like guardrails: they're a great safety feature, but they're not a substitute for steering.
    • Build out the Keyword list. Keyword list maintenance can be a daunting task at the highest levels, but for folks without tools, focus the build out around the top selling products and product categories. It may seem logical to focus efforts around keywords related to the stuff that isn't selling, but remember that paid search is demand driven. Assuming the basic coverage is decent, build out the winning categories in depth first.
    • Study user search strings. This is tremendously valuable for both negative creation and new keyword development.
    • Hand-write copy for the top volume keywords. Writing keyword-specific ad copy for tens of thousands or hundreds of thousands of keywords is impossible, and not worth the time involved. However, having super-target, tested copy on the head is absolutely worth the time.
    • Hand-pick landing pages for the top volume keywords. Landing traffic on the ideal page is tremendously important for conversion rates. Designing special pages may or may not be cost effective, but taking the bulk of the traffic to the best available page is well worth the effort.
    • Study seasonal impacts on traffic quality. How much does the value of traffic vary with the seasons? If the answer is "significantly" for some categories of keywords then there is a significant opportunity for anticipatory bidding to capture more traffic as the season ramps up and avoid over spending as the season winds down.
    • Study match-type and syndication partner performance. Add layered campaigns with exact match versions bid highest, modified broad match or phrase bid next highest, and broad bid lowest as the data dictates.

    • Product-level keyword build out. This is hugely valuable for folks with makes and model numbers. Should be done once for every advertiser; must be done regularly for others. This can be less valuable for others, it just depends on the importance of the tail for that particular advertiser.
    • Build and track geo-targeted campaigns. Folks with brick and mortar footprints, and lead-gen advertisers may find this most beneficial.
    • Tie in Google product extensions if possible. The click-through-rate benefit is real and worth going after for folks with clean Google base feeds.
    • Test landing page templates. Do sub-category pages convert as well as search results pages? Does sorting, or limiting the view of search results pages help/hurt conversions? These tests often need to be done across a wide portfolio of keywords to generate actionable volumes of data.
    • Test "unique selling propositions", static versus dynamic headlines and calls to action in the ad copy. It gets hard to raise the bar on copy after a while, but these tests are worth doing particularly in the early stages of a program.
    • Test offer copy in the ads. Well-written promotional copy improves CTR and QS, but often brings in lower quality traffic as well. Find out early on whether touting promotions helps or hurts and apply the findings of those tests ongoing if the promo copy typically helps.
    • Study time of day and day of week effects on traffic quality. Day-parting can have material benefits for some companies and not much for others. It's worth investigating and making adjustments as needed.
    • Build out content campaigns. Constructed properly and bid properly, content ads whether text or display can be cost effective. They're not likely to be huge, and we recommend wading in slowly.
    • Study keyword-level data over longer and shorter time windows. Sometimes interesting and actionable trends appear when you look at different windows than "normal". We've found, for example, that keywords tied to particularly high-ticket items often have very low conversion rates. That can make them bounce from being "goats" to "heros" depending on the window of time studied; that fluctuation can result in irrational bidding and lost opportunity if not spotted. Keep statistical significance in mind.
    • Play with Google's Bid Simulator Data! You may find that there are opportunities to bid more aggressively and make up the lost efficiency in volume; and, you may find opportunities to drop bids and save significant amounts of money without losing much traffic.
    • When available, add Product Listing ads, and non-brand site-links. We anticipate both will be material wins for well-managed programs.
    • This list is not intended to be complete, but a mixture of tried and true essentials and perhaps a few new ideas that folks haven't tried yet.

      I'm speaking at SMX East on the Industrial Strength Paid Search panel. Y'all come!

  • Without a clear understanding of, and execution around the fundamentals above, the tactical issues below are merely window dressing. The above determines the size of the opportunity; what's below can help determine how much of that opportunity is realized.

    Sidebar rant: There are folks who don't know how much value paid search traffic brings to the table because they haven't set up the tracking and understood its limitations; haven't thought about channel interaction; or even how much they should spend to generate a lead or a dollar in margin, and yet they still want to talk about account structure -- it blows my mind!



7 Responses to "Time is Money in Paid Search"
Great post! Looks like I’ve got some work to do. Willing to share more about your bid-management predictive analytics? I've done studies similar to what you've described using decision trees, etc. and kept coming up with the keyword being most predictive of a conversion. Of course that keyword has a lot of information content associated with it (where it lives in the account for one). Geography was another variable that was highly predictive of a conversion for us (we are a moving company). It is also interesting to note that what is most predictive of a conversion (lead) is not necessarily predictive of a sale. We found that variables surrounding a lead, and perhaps their post click activity were much more predictive of a sale (great for lead-scoring, value-based call routing). This doesn't necessarily help you target more leads as you don't learn these variables until post-click. Is there a particular algorithm or method you prefer for bid-management data-mining/predictive analytics? It's tough to find good info on this topic.
Thanks Chad, I'll answer these questions soon as you come to work for us! :-) No question that the historical data for the ad (the combination of keyword, copy, landing page, geo-targeting, etc.) is the best predictor of traffic value on that ad, with data about that keyword globally being the next best predictor. The hard part of bidding is how to handle the mid-low traffic ads and keywords that don't have sufficient data by themselves to make sensible bids. You won't find much information online about specific data modeling methodology for bidding because those are hard won trade secrets that agencies aren't willing to share. We've found that there are many many different ways to do the modeling and the quality of the resulting predictions vary widely. Sorry to be opaque!
No worries, I wouldn't tell me either. But I've never been shy about asking smart people questions--has made a huge impact on my career so far.
Keri Morgret says:
First, I think this is a great post that helps people decide where to focus their attention for PPC. Regarding measuring spillage over to call centers, tracking has improved since the post from January of 2009. There are ways to use javascript to give a unique phone number for PPC (and organic and social) that is on every page of the site, and stays cookied for 30 days. It's particularly useful for someone managing PPC for a car dealership to show how many leads PPC is driving.
Thanks for stopping by, Keri. We find the tracking methodology described in our post gives us much more precise, actionable data than the phone number game. We can track to the exact ad that triggered the call, where most phone number systems have a limited bank of numbers. That's fine for gauging the overall performance, but for paid search bidding more granular is better.


Check out what others are saying...
[...] This post was mentioned on Twitter by Rimm-Kaufman Group, Keyword Grouper Guy, Long Tail, SEM Software, Meta Keywords and others. Meta Keywords said: Time is Money in Paid Search: My monthly paid search column for SEL in case you missed it. In my last column for S... [...]
[...] any appearance of link spamming, below are a list of “more information” links for my “Time is Money” Post going live at SEL on Monday 8-30-2010. If you’re a regular reader of RKG Blog, [...]

Leave A Comment