Connecticut is famous for a long stretch of congested highway that’s perpetually under construction. Anyone from Rhode Island who takes a holiday weekend in New York drives through Connecticut unless he or she is seeking the fall foliage of western Massachusetts – or is otherwise directionally challenged.
So how many Rhode Islanders visit New York each weekend? Not many (most of us stay put). You might find out by counting all the Rhode Island license plates you find across New York State. But if you’re pressed for time, simply count the cars entering Connecticut from the eastern edge. If you subtract the cars that exit the highway just beyond the border for Foxwoods Casino, there aren’t many other reasons to enter Connecticut. You’ve discovered a reasonably precise proxy variable – but do you trust this number?
Most companies initially focus on optimizing the “top of the funnel” because it’s easy to observe and measure. (Possibilities include page views, registrations, signups, downloads, and new subscribers.) These metrics provide the largest sample sizes; however, they suffer the accuracy problem. They don’t necessarily indicate actual value. Have you ever turned on a campaign that generated a flood of leads with zero purchases? This happens frequently with push marketing, co-registration, and broad campaigns.
On the other hand, some companies shift to the “bottom of the funnel” and endure a different problem. (Sample metrics include transactions, customers, revenue, and bookings.) While these are the ultimate objectives of your customer acquisition campaigns, they typically suffer the precision problem. You may notice blocky or clustered data. For example, suppose your hardware site purchases an AdWords phrase: “Phillips screwdriver for real cheap.” Your 20 monthly clicks may result in 0 purchases, but does that mean you should bid $0.00/click? And when Boring Barry or Average Ann purchase 500 screwdrivers off this link tomorrow, should you suddenly bid a thousand dollars per click? “Long-tail campaigns” with limited data are particularly susceptible to clustering. Furthermore, many bottom-of-funnel metrics suffer time lag – marketers must wait through sales cycles to observe enough data to make optimization decisions. Especially for B2B companies, these sales cycles can last several months.
So what’s the solution to this tradeoff between accuracy and precision? You must identify midfunnel metrics. These are metrics that yield large sample sizes and are quick to be observed, yet are reasonable indicators of actual value or future purchase intent. At LogMeIn, we created a “quality trial” metric to describe someone who downloaded the free service and subsequently performed one remote session. 15Five studies trial users who file that first report. A hardware manufacturer may measure campaign allocations against resulting “purchase page” views. A nonprofit may examine the impact of programs on volunteer rosters. What are some other examples of midfunnel metrics?
In our weekend escape example, you don’t need to count every Rhode Island license plate in New York State, but perhaps you can do better than staking out Connecticut’s eastern edge. For a suitable midfunnel metric, I propose you count the Rhode Island license plates crossing the George Washington and Tappan Zee Bridges (just past the entrance to New York State).
If you’re from northern California, you may find it easier to place African countries than New England states on a map. I’ve got you covered: substitute Lake Tahoe for New York and Sacramento for Connecticut. Count license plates entering the El Dorado and Tahoe National Forests. Everything else follows.