Tag Archives: store testing

The Myth of the Single KPI for Testing

Continuous Improvement through testing is a simple idea. That’s no surprise. The simplest, most obvious ideas are often the most powerful. And testing is a powerful idea. An idea that forms and shapes the way digital is done by the companies that do it best. And those same companies have changed the world we live in.

If testing and continuous improvement is a process, analytics is the driver of that process; and as any good driver knows, the more powerful the vehicle, the more careful you have to be as a driver. Testing analytics seems so easy. You run a test, you measure which worked better. You choose the winner.

It’s like reading the scoreboard at a football game. It doesn’t take a lot of brains to figure out who’s ahead.

Except it’s usually not that easy.

Sporting events just are decided by the score. Games have rules and a single goal. Life and business mostly don’t. What makes measuring tests surprising tricky is that you rarely have a single unequivocal measure of success.

Suppose you add a merchandising drive to a section of your store or on the product detail page of your website. You test. And you generate more sales of that product.

Success!

Success?

Let’s start with the obvious caveat. You may have generated more sales, but you gave up margin. Was it worth it? Usually, the majority of buyers with a discount would have bought without one. Still, that kind of cannibalization is fairly easy to baseline and measure.

Here’s a trickier problem. What else changed? Because when you add a merchandising drive to a product, you don’t just shift that product’s buying pattern. The customer who buys might have bought something else. Maybe something with a better margin.

To people who don’t run tests, this may come as a bit of surprise. Shouldn’t tests be designed to limit their impact so that the “winner” is clear? ‘

Part of a good experimental design is, indeed, creating a test that limits external impacts. But this isn’t the lab. Limiting the outside impact of a test isn’t easy and you can  never be sure you’ve actually succeeded in doing that unless you carefully measure.

Worse, the most important tests usually have the most macro-impact. Small creative tests can often be isolated to a single win-loss metric. Sadly, that metric usually doesn’t matter or doesn’t move.

If you need proof of that, check out this meta-study by Will Browne & Mike Jones (those names feel like generic test products, right?) that looked at the impact of different types of test. Their finding? UI changes of the color and call-to-action type had, essentially, zero impact. Sadly, that’s what most folks spend all their time testing. (http://www.qubit.com/sites/default/files/pdf/qubit_meta_analysis.pdf)

If your test actually changes shopper behavior, believe me, there will be macro impacts.

It’s usually straightforward to measure the direct results of a store test. It’s often much harder to determine the macro impact. But it’s something you MUST look at. The macro impact can be as or more important than the direct impact. What’s more, it often – I’ll say usually – runs in the opposite direction.

So if you fail to measure the macro impact of a store test and you focus only on the obvious outcome, you’ll often pick the wrong result or grossly overstate the impact. Either way, you’re not using your analytics to drive appropriately.

Of course, one of the very real challenges you’ll face is that many tools don’t measure the macro impact of tests at all. In the digital world, the vast majority of dedicated testing tools require you to focus on a single KPI and provide absolutely no measurement of macro impacts. They simply assume that the test was completely compartmentalized. That works okay for things like email testing, but it’s flat-out wrong when it comes to testing store or website changes.

If your experiment worked well enough to change a shopper’s behavior and got them to buy something, the chances are quite good that it changed more than just that behavior. You may have given up margin. You likely lost some sales elsewhere. You almost certainly changed what else in the store or the site the shopper engaged with. That stuff matters.

In the store world, most tools don’t measure enough to give you even the immediate win-loss results. To heck with the rest of the story. So it can tempting, when you first have real measurement, to focus on the obvious: which test won. Don’t.

In some of my recent posts, I’ve talked about the ways in which DM1 – our store testing and measurement platform – lets you track the full customer journey, segmentfunnel and compare. Those capabilities are key to doing test measurement right. They give you the ability to see the immediate impact of a test AND the ways in which a change affected macro customer behavior.

You can see an example of how this works (and how important that macro behavior is in store layout) in this DM1 video that focuses on the Comparison capabilities of the tool.

https://www.youtube.com/watch?v=lbpaeSmaE74&t=13s

It’s the right way to use all that power a store testing program can provide.

Store Testing & Continuous Improvement

Continuous improvement is what drives the digital world. Whether applied as a specific methodology or simply present as a fundamental part of the background against which we do business, the discipline of change and measure is a fundamental part of the digital environment. A key part of our mission at Digital Mortar is simply this: to take that discipline of continuous improvement via change and measurement and bring it to stores.

Every part of DM1 – from store visualizations to segmentation to funnel analytics – is there to help measure and illuminate the in-store customer journey. You can’t build an effective strategy or process for continuous improvement without having that basic measurement environment. It provides the context that let’s decision-makers talk intelligently about what’s working, what isn’t and what change might accomplish.

But as I pointed out in my last post, some analytic techniques are particularly useful for the role they play in shaping strategy and action. Funnel Analysis, I argued, is particularly good at focusing optimization efforts and making them easily measurable. Funnels help shape decisions about what to change. Equally important, they provide clear guidance about what to measure to judge the success of that change. After all, if you made a change to improve the funnel, you’re going to measure the impact of the change using that same funnel.

That’s a good thing.

One of the biggest mistakes in enterprise measurement (and – surprisingly – even in broader scientific contexts) is failing to commit to your measurement of success when you start an experiment. It turns out that you can nearly always find some measure that improved after an experiment. It just may not be the right measure. If folks are looking for a way to prove success, they’ll surely find it.

Since we expect our clients to use DM1 to drive store testing, we’ve tried to make it easy on both ends of the process. Tools like funnel analysis help analysts find and target areas for improvement. At the other end of the process, analysts need to be able to easily see whether changes actually generated improvement.

This isn’t just for experimentation. As an analyst, I find that one of the most common tasks I have do is compare numbers. By store. By page. By time-period. By customer segment. Comparison provides basic measurement of change and context on that change.

Which makes comparison the core capability necessary for analyzing store tests but also applicable to many analytics exercises.

Though comparison is a fundamental part of the analytic process, it’s surprising how often it’s poorly supported in bespoke analytics tools. It took many years for tools like Adobe’s Workspace to evolve – providing comprehensive comparison capabilities. Until quite recently in digital analytics, you had to export reports to Excel if you wanted to lay key digital analytic data points from different reports side-by-side.

DM1’s Comparison tool is simple. It’s not a completely flexible canvas for analysis. It just takes any analytic view DM1 provides and allows you to use it in a side-by-side comparison. Simple. But it turns out to be quite powerful in practice.

Suppose you’re running a test in Store A with Store B as a control. DM1’s comparison view lets you lay those two Stores side-by-side during the testing period and see exactly what’s different. In this view, I’ve compared two similar stores by area looking at which areas drove the most shopper conversions:

Retail Analytics and Store Testing: Store Comparison in DM1

You can use ANY DM1 visualization in the Comparison. The funnel, the Store Viz or traditional reports and charts. In this view, I’ve compared the Shopper Funnel around a single merchandising category at two different stores. Not only can I see which store is more effective, I can see exactly where in the funnel the performance differences occur:

Retail Analytics and Store Testing: Time Comparison

Don’t have a control store? If you’re only measuring the customer journeys in a single store or if your store is a concept store, you won’t have another store to use as a control. No problem, DM1’s comparison view lets you compare the same store across two different time periods. You can compare season over season or consecutive time periods. You don’t even have to evenly match time periods. Here I’ve compared the October Funnel to Pre-Holiday November:

Retail Analytics and Funnels: Store Testing

Store and Date/Time are the most common type of comparison. But DM1’s comparison tool lets you compare on Segments and Metrics as well. I often want to understand how a single segment is different than other groups of visitors. By setting up a segmentation visualization, I can quickly page through a set of comparison segments while holding my target group constant. In the first screen, I’ve compared shoppers interested in Backpacks with shoppers focused on Team Gear in terms of how effective interactions with Associates are. With one click, I can do the same comparison between Women’s Jacket shoppers and Team Gear:

Funnel Analytics and Store TestingStore Analytics Comparison: Store Testing Segments

The ability to do this kind of comparison in the context of the visualizations is unusual AND powerful. The Comparison tool isn’t the only part of DM1 that supports comparison and contextualization. The Dashboard capability is surprisingly flexible and allows the analyst to put all sorts of different views side by side. And, of course, standard reporting tools like Charts and Table provide significant ways to do comparisons. But particularly when you want to use bespoke visualizations like Funnels and DM1’s store visualizations, having the ability to lay them side by side and quickly adjust metrics and view parameters is extraordinarily useful.

If you want to create a process of continuous improvement in the store, having measurement is THE essential component. Measurement that can help you identify and drive potential store testing opportunities. And measurement that can make understanding the real-world impact of change in all its complexity.

DM1 does both.

Click here to sign-up for a Demo of DM1.