For the first 20 years of my working life, I was in an advertising or media agency. In our interactions with clients, very early on, there would be a defined target audience that we set out to create communication for or buy media in order to reach. Usually, the definition of that audience was based on an extensive U&A (usage and attitude) study or other forms of market research.
At one point a few years ago, I was working in a mobile ad-tech company that was a DSP, Ad Network and performance agency rolled into one – not to mention being in the business of developing and licensing a DSP platform. One of our biggest clients was Uber, focusing on acquiring users in China. We also had other clients who were focused on user acquisition in different markets – Taiwan, Japan, South East Asia and so on.
When the data science team first sat down with me to explain how they started a campaign I was fascinated to find that they didn’t begin with any defined target audience at all. Instead, they’d attempt to deliver impressions evenly across every single audience segment they could define – and in an ADN environment where we were installing our own SDKs onto mobile apps and therefore reading a lot of data from consumer phones we could define segments using a very large number of variables including phone hardware and software specs, apps on the phone, frequency of usage of those apps and so forth. Typically we had several thousand segments in any initial campaign.
The approach was that we’d deliver enough impressions in each segment to start getting a read of consumer response – click through rates, app download, app install, user registration and first usage of the app. Then, based on where the segment size and response rates were most promising, our system would start optimizing bidding for and buying impressions in those segments so that we could deliver the client KPI at the lowest cost. Once we’d exhausted a segment and the response rates in it came down, the system would automatically move to the next best on the list and so on.
Across clients, while I could often “predict” the target audience, I often found to my surprise that the user profile which emerged from doing this kind of exercise led to a counter-intuitive audience definition, but one that was delivering results.
Of course, we had more than our fair share of clients who just wanted to deliver impressions against a pre-defined audience and we didn’t get to do this kind of experimentation with them but I’d often wonder if they’d missed something in defining their audience.
For one thing, brands often define their target audience using fairly theoretical criteria, or with research that’s done under very unreal conditions. I’ve seen lots of audiences defined in terms of psychographic or other variables that don’t show up as filters when you’re buying media, so they’re essentially useless. Digital media buying, especially once we get to a post-cookie world, will be entirely predicated on observing and reacting to a consumer behavior without knowing much more about them, so this obsession with defining a segment with lots of variables that can’t be used as filters will become even more useless over time.
We used to talk about “spray and pray” media strategies where clients bought broad reach media and then hoped that sufficient spending and share of voice would deliver results. We now have the option of “spray and learn where to focus” media where you can observe which consumer behaviors lead to the responses we want, identify a buyable audience on that basis and focus subsequent buys on that definition.
That process works only if you have a hard KPI, of course. You need a KPI that is a specific consumer action specific to a brand or to the ad that they’ve just been seen. Unfortunately many marketers have not evolved their KPI definition as digital media has evolved its capability of delivering them.
So, what would I recommend?
Start with a campaign that has a specific purpose – not just reach / frequency / awareness, but an action relative to a brand.
Work with a set of DSPs/ADNs/ performance marketing agencies and invest some initial money on delivering impressions across a broad audience, segmented by behavioral and other data in as much detail as possible, reading the results in terms of whatever hard KPIs you’ve set.
Analyze the results and identify the (behavioral) segments that give you the best results.
Focus on those going forward until the results start to dry up – then move to the next best tier of segments.
Two advantages of that:
- If you’ve defined a suboptimal audience, you’ll learn very quickly who your audience really is
- You’ll be defining a target audience in terms that actually make sense when buying media, instead of using theoretical filters that are hard / impossible to apply to real life media
If you’re launching a brand that’s never existed before and has no history or taking a brand to a new market with different consumers and culture, this approach makes even more sense.
Don’t spray and pray – spray, learn and focus.