Editor's note: The following is a guest post from Ed Ramsey, vice president of data and analytics at Valassis.
In advertising, the question of efficacy is always difficult. On one hand, it's well understood that promotion is a necessary ingredient of introducing products and services. After all, who would know about a better mousetrap if the inventor tells no one? On the other hand, it's also understood that promotion saturation is a real issue. At a certain point, everyone who needs a mousetrap has already bought one and will continue buying the brand without constant reminders. At least for a while.
Much of the art of promotion lies in finding the correct balance of creative content, timing, channel and cadence of advertisements with the twin goals of keeping existing customers engaged and acquiring new ones.
When a quantitative measure of a campaign is needed, the promotion industry has typically relied on three components: first-party, second-party and third-party measurements. However, within the last decade a new source of information has become available: mobility data.
Mobility data as a proxy for second party
Mobility data is certainly a third-party product, but unlike traditional survey-based information, it's essentially in real time. As smartphones become ubiquitous — approximately 80% penetration in the U.S. — a strong majority of actual and potential consumers can be tracked to a good degree of accuracy via the embedded A-GPS function. It's tempting to think of such data as a substitute for second-party; after all, if we can count customers who enter a quick service restaurant, we can treat that as a close proxy of that QSR's total sales. In other words, we may not know the exact product mix, but our guess would be intelligent.
Recently, we engaged with several mobility data providers to measure the effects of advertising campaigns. In the process of collecting, processing and interpreting mobility data, we learned a few lessons and potential pitfalls, shared below.
Penetration of location technologies
From a non-lawyer perspective, the Telecommunication Act requires consumer consent to use their location (i.e. their cell phone's location) for commercial purposes. While many consumers are aware of this, most will download and enable an app that allows GPS tracking. There are certainly benefits to this decision — local searches, social media check-ins, navigation — but few consumers read the fine print of location-tracking agreements that interpret the "yes" as permission to essentially resell their location to other business subscribers of that app's API. While timely penetration numbers are hard to nail down, we would estimate that about 75% of GPS-enabled phones can, in principle, be tracked to a mobility data provider based on permission settings alone.
Mobility vendors tend to claim very high numbers of mobile devices, some in the high tens of millions, but are reluctant to quantify what percentage of those devices can be seen at a given time or even on a given day.
It's important to speak to a potential vendor about this because the difference between what has ever been seen and what can be seen right now can be dramatic. Ask, for example, how many devices they saw yesterday for more than four continuous hours.
Basic GPS accuracy of about 10 meters seems more than adequate for geolocation. There are, however, limitations:
- GPS is not accurate in dense urban environments like urban canyons.
- GPS is not particularly accurate inside of buildings such as malls. It's particularly terrible inside enclosed parking lots and simply doesn't work underground.
- GPS (as implemented in most cell phones) cannot use its altitude information. In a multi-story mall, GPS will not distinguish between stores that are "above and below" each other.
- 10-meter (30 ft.) latitude/longitude accuracy is not sufficient to distinguish smaller, closely-spaced stores from each.
Geofencing and the Costco effect
Different vendors will report vastly different visitation counts simply due to geofencing standards. One vendor may include the parking lot, while one may not. It's often not clear, just from aerial or satellite photography, where one store ends and another begins. The shared parking lot issues can be a nightmare. Accidently including a portion of an always-crowded and popular store can cause huge errors or completely bury the signal.
- Ask the vendor to explain their geofencing procedures.
- Consider requesting a geofencing map for the study locations.
- Consider excluding locations with a likely spill-over effect from popular nearby attractions.
Beware of temporary outliers
We already mentioned the Costco effect above, but more subtle issues may affect mobile counts. For example, local construction projects will create artificially high or low counts due to road closures or presence of construction personnel who carry cell phones. It's easy to confuse the opening of a nearby playground with the success of a QSR campaign. National holidays and various religious celebrations can create peaks and valleys in the data that will not be visible in an aggregated view.
- Confirm with the customer that all stores or locations under study were indeed operating normally during each day of the study.
- Always plot the daily counts and look for the normal weekly cadences as well as holiday outliers. This is particularly important with unfamiliar geographies, such as outside of the U.S.
Aggregation by vendors
There are two types of aggregation — location-based and temporal. Vendors are likely to pre-aggregate the data by either combining multiple stores or multiple dates. This makes the counts look larger and fine-grained analysis difficult. As it often happens, when the test and control store groups are created, mistakes are often made, or new evidence shows up that requires groups are modified. Having the vendor reshuffle the data is expensive and unnecessary.
- Ask the vendor to provide by-store, by-date counts. If practical and if needed for analysis, ask them to provide the by-hour counts as well. If personally identifiable information becomes an issue, ask for data broken down into "morning," "afternoon," etc.
Mobility data is a new, rich source of information. Within its limitations, it can be used instead of or in addition to second- and third-party engagement data. Above, we outlined some of the common pitfalls specific to mobility and mobility vendors. These pitfalls are avoidable, and with proper vendor management, mobility data can make a robust, near-real-time contribution to traditional efficacy calculations.