Tracking web lift in retail cross-sales


I (“Wandering” Dave) moderate the Web Analytics Forum, and a few days ago, Tim M posted a question about cross-sell activities, and determining the effectiveness of some changes they recently made. In particular, he notes that, “The difficulty I’m finding is in determining the attributable growth in store sales overall to cross-sell.”
Meshed G_e_ars
This is indeed a difficult problem that I seldom see addressed, so I wanted to note that there are a couple of ways to “buy” some measurements in situations like these. Of course, they are dependent on a few attributes of the product or service you’re selling, such as the average unit price. If you’re selling $1000 items, then you can probably afford a short 3-day coupon promotion where a web-printed coupon is brought into the store for a $25 discount. That will cost you $25 per response, but if you’re already doing something like a $20 discount promo, then it’s only an incremental $5 to collect some cross-sell data. Of course, if your average unit price is a $2.49 bottle of aspirin, you’ll need some other kind of incentive, like, “Print this coupon out and bring it in for a free buyer’s guide.”

Now, if your retail sales aren’t through your own stores, but through some other retailer (e.g., you make aspirin, but sell it through chain stores you can’t control), then you’ll have to try something else. Maybe a packaging insert that has a link to a vanity-redirect URL (e.g., –>, which is wrapped in some kind of metrics tag to track the downloads.

As with all response, you’re always sampling from your entire customer base, so what you’ll have to do is establish a baseline now, and then watch for changes that match your new activity, with a time lag taken into account. (How long does it take to get a box of aspirin with a new package insert onto a retail store shelf, given the inventory restocking cycle?) So it may well be too late to gather data for the last set of changes, unfortunately.

Aaron King had a few other suggestions, such as looking at how average order value (AOV) (and margin) increased after the recent changes, and suggests looking at orders with and without a cross-sell attribute. I’m assuming that such an attribute is missing in the first place — since if it were there, measuring deltas would be fairly straightforward.

Another suggestion for Tim would be to look and see if, even though the AOV may not have changed, the product mix might have. For example, retail customers are still spending $12.57 per purchase, but instead of buying acetaminophen, they’re buying aspirin for the same price.

One further suggestion would be to ask the retail team about the distribution of stores — if there is a discrepancy between the DMA marketing ZIP codes for store layout, compared to your web visitor distribution, then you might see a change in cross-sell for retail stores in areas which have strong internet penetration, compared to those without. As a rough example, suppose you had stores on the east coast and midwest, but your web visitors come from the midwest and the west coast. Then your east coast retail stores are your control, and your midwest stores would show your lift, if any. West coast data might show potential demand, or inform your retail expansion team where good target audiences might already be, for focusing future growth. This seems like a long-shot though — very indirect data, and so it may not be actionable with the persuasive strength I’d like to see from better-grounded web analytics recommendations.

I’ll be curious to see how Tim moves forward with this, and what decisions, if any, actually get executed based on the data, analysis, and insights generated. And of course, if you have other ideas or suggestions that you’ve successfully integrated into your multi-channel campaigns, we’d appreciate hearing about them.


Tags: , ,

%d bloggers like this: