How to get human insights out of digital metrics if "all models are wrong and only some are useful"?
Digital Analytics Melbourne: 13 excellent speakers in a day, virtually every angle covered to lift any data analyst to a higher level, with Lawrence Leung, “most noticeable for being of Asian descent yet able to grow facial hair” as a perfect MC. It’s been a really, really good day.
One thread throughout the day was the imperfectness of whatever model is applied: “All models are wrong, some models are useful”.
I would pose that the challenge for digital data is not only to provide a good proxy model of consumer behaviour, but to turn digital data into real human insight.
Authenticity of communications, speaking to “people” or even better to ‘persons’ is the best way to achieve click throughs, engagement and resonance – it requires real insight. Also, because these real human insights provide a common understanding of the target groups, based on which each individual in a company can act in sync. The social media team, the marketing team, the customer service team or the analytic team; A common consumer understanding keeps a brand together.
For that reason one of my favourites of the day – must have been Tim Wilson, with the topic: Digging for buried Analytics Treasures.
Unlike the title of his talk suggests, he didn’t delve into technical possibilities of Google Analytics, or, in tips and tricks. He went upstream. He provided a practical way of hypotheses driven analysis:
The formula which he proposed not only states the hypothesis but also its potential impact when verified:
I believe …..(some idea)…..
If am right, then (some action).
This is in effect “Starting with the End in Mind”.
It offers a great opportunity to bring in the “Human”. e.g., I believe that we’re attracting mostly price sensitive buyers. If I am right, we may have to review our value proposition.
This way of starting the problem solving process, also helps to realise that digital is but one source of human data and may need to work together with other sources. Formulating hypotheses may raise questions that cannot be answered with digital data alone.
Hypothesis development is hard to do well, McKinsey does it well and has probably set the bar for this approach, but for instance very few traditional consumer research agencies actually practice this approach. Clients often don’t invest the time in it either: “Didn’t I give you the questions I want answered?” which is clearly not the same. Ironically, clients will spent tons of time once the results are in, dicing and slicing to find what they’re looking for.
(Here is a template that Tim Wilson has kindly provided via Twitter)
Framework For Omni-Channel Consumer Behaviour
Carey Wilkens offered a very useful model for looking at modern consumer behaviour, as modern consumers do not fit well in funnels, such as awareness, consideration, preference and purchase.
So, she proposed to look at consumers in the below lifecycle metrics, placed in a circle:
- Favour (including the payment) and
- Engage (which includes recommending)
From any stage in the cycle, consumer may go to “Favour” – in order to drive revenues. These lifecyle data can consequently be aligned with business objectives.
This is another great step forward to bring back human colours into the digital data analysis.
Digital can explain the “what” better and better, even though much of the conference dealt with the challenges just to gather a set of data that accurately captures the “what”.
To mention just two quotes: “If you’re not thinking about how to keep your data clean from the beginning you’re f*_@3d!. And: “Are you sure that the week starts Sunday midnight in all of your data sources?”
‘The why’ holds the real learning; It has some longevity and cross applications. Once the ‘why’ is clear, lessons learned can be applied elsewhere - not only in digital platforms, but also in other forms of engagement. This helps to build that common understanding of the target consumer, a common culture so that each individual intuitively acts in sync with the rest.
Good UX design will provide those true colours, just as good marketing campaigns are based on an in depth understanding of how people actually make their purchase decision and it is up to the digital data-analysts to provide, not just the performance data, but, as Tim argues, the verification of the hypotheses and assumptions about human behaviour: the tangible results of those true colours.
Let me put Tim’s method in practice:
I believe that, to get ‘truer’ models for consumer behaviour, we will see an unlikely blend emerge of anthropologists, psychologists, sociologists, statisticians and data analysts. As ’machine learning’ and data analysts are growing evermore sophisticated, the focus of the analysis will shift from the "technical" challenges, back to the “human” challenges.
The company that understands its customers’ true colours, and is able to respond accordingly, will win. So nothing really changes - except this is now happens at the speed of light at unprecedented scale.
Read More On Digital Doughnut