THE RKGBLOG

Geographic Impact of PPC Part 2: Retail Chains

BACKGROUND

A month ago or so, Digital Element was kind enough to lend us some of their highly accurate ip to zip code mapping information for use in a study. We combined this with census data about each zip code, and our own post-click tracking of sales.

We mapped the traffic from 3 million visitors to one of our online pure-play client’s site, and 2 million visitors to the site of one of our clients with a national chain of retail stores.

We hoped and expected to find a correlation between geographic isolation (low population density) and conversion rates, and a similar impact from wealth. We studied many many other factors as well, but none showed the strong correlation we expected relating to the value of the traffic. We did find a stronger correlation with respect to the per capita search volume from each zip code.

THE IMPACT OF RETAIL PRESENCE

The second piece of our study sought to determine whether the proximity to a brick and mortar store had an impact on perceived PPC traffic value and likelihood to search. We overlaid the data from above with the locations of our client’s stores. We separated traffic into two buckets: brand/trademark search and competitive/non-brand search.

We hypothesized that in zip codes with a physical store presence online conversion rates and sales dollars per click would be lower than in demographically similar zip codes without a store. We suspected that people shopping for product online would be more likely to abandon their online shopping in favor of an in person visit to the store if a store was geographically convenient.

Indeed, we hoped that by calculating the difference in measured traffic value online, we would get meaningful insight into the elusive spillover effect to stores.

FINDINGS

The value of online PPC traffic, measured as sales dollars per click, is the same or better in zip codes with stores as it is without!

The propensity to search is significantly higher in zip codes with stores than it is in demographically similar zip codes without a store!

“This is the big one, Elizabeth! I’m comin’ to ya!”

– Fred G. Sanford

Valid statistical analysis is complex. For example, a cursory view of the data would conclude that poverty boosts conversion rates. The traffic from zip codes with average home values less than $75K appears to be much better than the traffic from wealthier areas, but that’s an illusion. We have to mathematically tease out the effect of home value from the effect caused by geography; the fact that most of those low home value zip codes are in Puerto Rico. Teasing out these effects requires some heavy stats.

It is often difficult to present complex findings in a way that doesn’t sound like gibberish to the desired audience, so rather than turning this into a stats paper, we’ll just present the data sliced a few ways to give a sense of what we see.

[Note: as we take apart the data by specific attributes of the zip codes, we see tremendous volatility in the "with store" data as the signal to noise ratio gets worse. This is a by-product of thin data. This company has stores in 800 of the 30,000 + active zip codes, and when you then take traffic from those 800 zips and parse them into smaller buckets the data gets very noisy. Interesting nevertheless!]

Aggregate Data

Aggregate Data

By Average Home Value

By Average Home Value

By Population Density

By Population Density

By Geographic Region

By Geographic Region

LIMITATIONS

There are many limitations to this study.

  1. N = 1. We studied one brick and mortar chain. More cases are needed to reach definitive conclusions.
  2. Product categories may influence results. We may find that the results vary with the product category. SKU-based vs non-SKU based, apparel vs durable goods, high-ticket vs low-ticket, in-store pickup vs not all may produce different outcomes.
  3. Proximity proxy. We defined the concept of a store being convenient as in the same zip code, but by definition parts of an adjoining zip will be much closer than some points in the same zip code as the store. A better definition would measure proximity by driving time or mileage between the searcher and the physical store
  4. Competitor’s stores. It’s possible that having a direct competitor’s store nearby results in similar bleed effects from online to offline.
  5. Zip codes of the ISP, not the user. We really don’t know the location of the user, we know the location of the IP address. We had to cull data from zip codes that demographic data suggested were actually PO Box zip codes with no population.

CONCLUSIONS

A wiser person than me would consider the above limitations and politely decline to draw any conclusions, but what’s the fun of that?!?

This data suggests that the presence of a retail store doesn’t negatively affect the perception of online advertisements; in fact it has the opposite effect. Conversion rates are the same or better, and the volume of search is much higher when a store is present than when not.

The good folks at Forrester produce periodic studies showing that for every order placed online, 6 or 7 or 10 (depending on the study) orders are place offline that were “influenced by” online marketing. At one conference, I asked the presenter from Forrester: “Has anyone put serious thought into these statistics? Lots of folks “go online” before placing an order, okay, but do they actually see an ad, or do they go straight to the online store? Are they just visiting the site for the store locator? Do they even go to a store, or just to review sites? If they do interact with an ad, is it a brand ad or a “competitive ad?” Is it even an ad for the retailer from which they make the offline purchase? What good are these “influence” numbers if we don’t know the answers to these questions?”

Her response was: “They’re directionally valid.”

This research suggests that even that claim may be off base. From what we see, online marketing efforts benefit much more from the effects of offline advertisements (circulars, print ads, local TV ads, store signage, etc.) than vice-versa.

The question: how much credit should online ads get for traffic driven to the physical stores? may be the wrong question. The question might be how much are our online ads being over-credited for sales driven by traditional media?

We’ve long known that catalog mailings drive sales through PPC ads, both brand and competitive. This is not surprising and the more we think about it, neither would it be surprising that online ads cannibalize sales that should rightly be credited to other forms of traditional media.

Certainly, far more money is spent on offline ads than online, so it makes sense that we online marketers are working with a tail wind at our backs. However, there sure isn’t much talk about this in the industry.

It’s incumbent on those of us in position to offer good advice to be mindful of this possibility. In the rush to turn off “unprofitable” offline ads in favorable of “more profitable” online ads, advertisers may end up throwing out the baby with the wash water, and that would be bad for everyone at the end of the day.

Technorati Tags: , ,

Comments
9 Responses to “Geographic Impact of PPC Part 2: Retail Chains”
  1. George, first off, I’m imagining 10-15 years down the line you’ll be the inaugural Larry Page & Sergey Brin Chair of Search at Stanford, Harvard or Cambridge, and I’ll come visit you. I love this research and think it’s incredibly important stuff you’re doing.

    The results you found are exactly what I would have expected, and one of the conclusions you draw is one I always suspected, namely that offline advertising impacts search more than search impacts offline conversion.

    Search engines and industry analyst firms have been selling the myth of a one-way, online-to-offline value street for many, many years now, and it’s high time people realize that waiting at the bottom of the sales funnel – paid search – for customers is a recipe for disaster. Multichannel attribution measurement/optimization is the key opportunity for revenue and marketing gains from CMO orgs, as it alone will allow for a conscious, strategic shift from dependency on lower value, recurring cost paid search channels over to higher value, higher growth direct & referral traffic sources.

  2. Chris, thank you for the kind words. Hopefully they’ll endow a chair at the University of Virginia so I can stay near my donkeys :-)

    I think you’re right, though I suspect we’re never going to be able to measure everything. I think it’s important for advertisers to keep the totality of their marketing expenses within some reasonable fraction of their top line revenue, and do some experimentation within those limits to find the right mixture of marketing vehicles.

    The ability to track online media is a powerful, powerful asset, but we need to be mindful of the untrackable effects of marketing.

  3. Jim Novo says:

    Nice job. “Marketing Bands” at work:

    http://blog.jimnovo.com/marketing-bands-series/

    Offine media are better than online media at generating Attention and Interest. Online media are better than offline media at generating Desire and Action. Put them together and you get a really nice glide path into a Sale.

    I don’t think it’s too much of a stretch to realize a very simple idea: when people go to a search engine, they already have Interest – the very nature of doing a search begs Interest. Many times, likely much more often than not, this Interest starts with offline Awareness.

  4. Jim, I agree 100%. I often give talks on Paid Search and use the metaphor of a rain barrel. Paid search is a rain barrel, it’s not a Native American rain dance. Search is great for capturing existing demand, but doesn’t work for creating demand where none exists.

  5. Rachel Schoenewald says:

    You may find it useful to work with a product called MapInfo Professional. It includes things like demographics, indicators of which zip codes are “points” instead of geographic boundaries and has amazing visualization tools. I’ve used it in the past and found it invaluable for understanding the relationships between retail locations, population density, competitor locations, income and home values, etc. You can load your data onto the map, analyze it with the built-in demographics and see the results.

  6. This need to at least *try* to understand and react to the interplay between online and offline is why I think SEM platforms, analytics, CRM, SCM and back-office systems are starting to form natural points of technical integration.

    This will be followed by increasing use of API’s, vendor consolidation, and ultimately the emergence of an enterprise technology industry that can say ‘Yes’ to companies that want to do multichannel marketing as efficiently as they perform manufacturing, SEM and other tasks.

    Exciting times ahead…

  7. Rachel, thanks for the tip!

    Chris, I concur that exciting times are ahead! Consolidated information warehousing and processing will help advertisers connect the dots. But having data and having the expertise to interpret and use the data wisely are different issues.

    As always, the wise advertiser must understand both what the numbers say AND the limitations of the numbers they study. Those folks who lose site of the holes in their tracking can be badly misled by false precision.

    Something as seemingly trivial as uniquely identifying customers is by no means a solved problem.

    You’re right, we have to try to measure everything, but the power of word of mouth is huge, and very difficult to track and that’s just one. Knowing what you don’t know is just as important as knowing what you know.

Trackbacks
Check out what others are saying...
  1. [...] the other piece of the study we looked at the impact of having a physical store in the zip code of searchers on both their [...]

  2. [...] try to be different at RKG. We’ve done lots of tests that didn’t go the way we expected, and didn’t produce the results that would help us increase our revenue. We [...]