Better decisions through Integrated Tracking

Just before the fall 2000 season, I had a conversation with a client about collecting source codes from customers. “If you’re not able to capture more than 90% of the source codes, it’s a problem,” I said. Back then, I was right. Because back then, phone orders represented roughly two-thirds of all orders, with mail and fax making up most of the balance. So if you weren’t collecting source codes from customers it was likely because your call center staff wasn’t properly trained in collecting the information, the order entry process was set up incorrectly, or the people at that level weren’t committed to — or didn’t understand the importance of — collecting the information. A few training sessions, an incentive or two, and the problem generally would fade.

But those days are over. Sure, low source-code capture rates remain a training issue, but these days it’s the customer who needs the training. And that’s a problem, because at the heart of it, most customers don’t care about helping you market to them better.

For the fall and holiday 2005 mailing seasons, many multichannel merchants experienced their first year of processing 50% or more of orders through the Web. How many of those Web orders came through with a customer-provided source code? In most cases, fewer than 10%. Carry that math out a step or two and look at the impact. Let’s say the Internet accounted for 50% of orders, 90% of which were untracked; phone accounted for another 40%, of which 10% had no source code; mail and fax combined accounted for 7%, of which 5% were untracked; and the remaining 3% of orders came from miscellaneous sources. Given these figures, nearly 50% of all orders taken would be untracked to a specific source code — and I haven’t even mentioned the impact of customer-directed e-mail campaigns that far too many multichannel merchants are pushing according to a plan that has nothing to do with how and when catalogs are mailed.

At a recent direct marketing event, veteran industry analyst Don Libey discussed “monster trends” for 2006 and beyond. One of his leading trends was that the catalog would shift from sales generator to a driver of customers to the Internet. Libey noted several ramifications of this anticipated shift, but what struck me was the logical implication that somehow we might reach a point where catalogs are intended to do what commercials do: Tell consumers about an assortment of products and engage them enough that the consumers would be moved to get up, walk to the computer, log on, and buy.

The fact is paper prices are going up. Postage rates just rose and are likely to rise again sooner rather than later. Shipping and handling expenses keep increasing. Margins in price-competitive industries are eroding. Consumers are skittish about everything from terror threats to high gas bills to natural disasters to geopolitical instability. It is more expensive than ever to mail a catalog and ship an order, and if you can’t track with even a modest level of certainty where your orders are coming from, you will become like every mass-marketed brand in America: hoping that your message, your catalog, got to the right people at the right time with the right product assortment and offer strategy.

It may be harder than ever, but tracking your orders has never been more important. Without an integrated method of tracking orders and building contact strategies, a flawed decision-making process is your destiny.

Integration now

An integrated tracking plan is rooted in an integrated mailing plan. (See “Creating an integrated mail plan,” January issue.) If your company has two to three years of Web order history available, you should be integrating those data into your contact strategy so that your segmentation strategy accounts for Web-only, catalog-only (in other words, phone/fax/mail-only), and multichannel buyers, and your methodology for mailing and e-mailing reflects the customer’s behavior. At this point, most multichannel marketers should be beyond mass-mailing their e-mail file when they’re segmenting, mailing, and measuring the catalog mail file at a detailed segment level.

Once you’ve laid out and executed a detailed, integrated plan, the tracking is a critical issue. Matchback processing continues to be a hot topic at industry events, but while matchbacks are important, they aren’t perfect.

A good integrated tracking plan should include the following steps:

  1. Ask for a source code when taking the order

    This seems like one of those no-brainers that you always hear about. But apparently it is not. Where source-code capture six or seven years ago was about how well a customer service person performed his job, today that onus is falling also on the shoulders of your Web development team, likely a group of people highly skilled at information systems and e-commerce engines but not as skilled at marketing. If you’re expecting a deluge of Web orders for 2006 and haven’t addressed source-code capture at the most basic level online — asking for the source code from a customer — you’re probably already in trouble with respect to getting an accurate read on results.

    At the very least, make sure you ask for a customer’s source code when the order is placed. And mimic the order entry process from the contact center. Most inbound scripts include some form of “May I get the 10-digit number from the back of your catalog, in the yellow box?” as part of the call process, yet many marketers will wait until, oh, the fifth of six checkout steps to ask for the same information online.

    If you have to position the data request as “To receive all applicable promotions, please provide the code from the back of your catalog,” so be it. And get visual. L.L. Bean is an example of a catalog-centric multichannel merchant that actually shows the customer online where to find the source code on the back of the print catalog. This technique, coupled with asking for the code at the beginning of the process instead of the end, has been shown to increase source-code capture rates dramatically.

  2. Account for e-mail sales at the segment level

    As suggested earlier, if you account for purchase channel among your customers as part of your contact strategy, you can identify segments that are receiving only catalogs, only e-mails, or both. A comprehensive mail plan will allow you to calculate contribution per order based on the total advertising expense to a customer from a given segment.

    For instance, if you e-mail a customer and mail him a catalog, his “ad expense” would be the cost to e-mail plus the cost to send a catalog, while the ad expense of the customer who gets only the catalog would simply be the cost to send a catalog. This method allows each segment to be measured independently and based on the contacts each received.

    You might try embedding segment-level identifiers in links. Some service providers can embed data into the links of your e-mail programs that can populate fields in your database through your shopping cart, including source codes. And even if your system can’t populate source codes at the shopping-cart level, your Web analytics software should be able to tell you which orders are being driven by e-mail campaigns. With some segmentation, those reports can help you accurately account for those orders by segment.

  3. Correlate the response between channels to account for shared variability

    Performing a correlation analysis allows you to determine whether two groups of data change in relation to one another and to what degree. As one variable (or channel) increases, does the other increase? If so, the correlation is positive. But if one increases as the other decreases, the correlation is negative.

    A correlation coefficient will be between 0 and 1, positive or negative. The closer to 1 the value is, the more tightly the two variables are related.

    Squaring the correlation coefficient (multiplying it by itself) yields the coefficient of determination, or the percentage of the variability that is shared between the variables.

    In English, this means that if the correlation coefficient (R) is 0.95, then the two variables are highly correlated with one another. If R is squared (0.95 × 0.95) we can determine that 90.25% of the variability between the two variables is shared. If I were applying this to a catalog environment and I assumed that my phone orders were driven by the catalog, and the Web orders shared 90.25% of their variability with phone orders, I’d have to say that 90.25% of my Web orders were also driven by the catalog.

    If you have the Analysis ToolPak installed with Microsoft Excel, you already have a statistical package to do this analysis. You can plot your orders by channel, by day, or by week. Select the range of data for your phone channel (generally accepted as being almost 100% catalog-driven) and your Web channel orders for the same period and run the correlation analysis through the Data Analysis option on your Tools menu.

    Square the correlation coefficient to calculate the coefficient of determination; you’ll use that figure to apply uncoded Web orders to the allocation process in step 5. But first…

  4. Perform order matchbacks to mail files using a logical and acceptable match key

    A matchback is not a canned process. Build a match key that works for your business. A match key will allow you to match records from one file (say, a mail file) to another file (for example, a response file). A consumer match key, for example, will typically incorporate elements of name, address, and zip code. Trollinger at 1234 Main St., 66202 might have the match key TRLLN123466202 — a fairly “loose” key, but one that would match only someone with a last name whose first 5 consonants were TRLLN and who lived at an address beginning with 1234 in the zip code 66202. We’ve built matching processes that cascade across multiple business units and time periods to acquire the greatest hit rate, but simplicity also can be very effective.

    The goal is to incorporate as many elements as necessary to do the most with the least: get the greatest number of hits with the fewest overmatches. While a good starting point for consumer mailers is a key that pulls in elements of last name, street number, and zip code, for business mailers site-level match keys generally work the best.

  5. Allocate remaining uncoded orders based on the shared variability between channels

    By completing the correlation analysis in step 3, you can say with some certainty just how many of your uncoded orders and dollars — particularly from Web sales — should be attributed to your catalog mailings. The big issue is that you are attributing those sales to the catalog, if for no other reason than the catalog was the driver.

A prorated allocation of the remaining uncoded orders, where orders and dollars are allocated to segments according to the number of tracked orders from the segments, is not a perfect answer. Prospect segments are generally underreported, and customer segments are generally overreported. Still, it is important to make your decisions based on as complete a data set as possible.

Beyond the steps: integrating the data makes the difference

If your merchandise analysis says your season was one of the best ever but your marketing analysis says your mailings were only mediocre, there’s obviously a disconnect.

Sure, some customers are actually “just showing up” at your Website. That’s expected, especially when you put effort into search engine marketing programs and pay-for-click campaigns. But those sales can be isolated, and what’s left needs to be accounted for in as clean and accurate a way as possible.

With a little effort, you can map out a process for accounting for all your orders across all your channels and making sure that your decisions, your future marketing plans, are as strong — and integrated — as possible.

Steve Trollinger is executive vice president, client marketing for J. Schmid & Associates, a multichannel marketing agency based in Mission, KS.