Data to Audience conundrum – using a DMP – An approach
A data led approach
The data to audience process can be broken down into stages as below where each stage works on the dimensions of data values and customer id.
1 – Data Identification
The first stage is to ensure that you only bring in quality and relevant data into DMP with associated Ids.
Remember GIGO (Garbage In > Garbage Out), bad quality data will give you poor performing audiences.
Some of the basic data identifications steps are:
- Filter out PII (personally identifiable information), primary key, foreign keys, audit keys, sparse and duplicate fields
- Convert date to duration values, dependent values to concatenated values, description to codes
- Consult marketing teams with the identified list.
Tip:
Even if there is no identified use case for a data value, if marketing feels data may be useful for personalisation in future, bring it in.
2 – Signal Transformation aka Trait taxonomy
The next step is to apply transformation logic (comparison, string, regex, logical) on data signals to build traits.
Remember traits are the basic building blocks for audience.
A good trait taxonomy should be:
- Granular to provide maximum flexibility while combining them to build audiences
- Structured for easier navigation, avoiding duplicates and providing for scalability
- Comprehensive because they are not populated retroactively
- Marketing friendly so it is easy for marketing to understand what they mean by just looking at their names
Tip:
Trait structure should be generally aligned to the inbound data sources, and resource managing the trait taxonomy should have good understanding of enterprise data and marketing function (unicorn!!)
3 – Trait Aggregation aka Audience taxonomy
If the first two steps have been done properly, this final step is a breeze for marketing and this is the moment marketing falls in love with data team. Otherwise, you will notice first signs of a failing DMP project here.
In this step, marketing aggregates traits into audiences using logical operators. Audiences get populated retroactively based on constituent traits. Simple!
Remember audience rules may need to be often optimised to achieve the minimum viable volume.
A good audience taxonomy should be:
- Marketing aligned, example aligned to various marketing channels, or customer life stages, or product portfolios
- Democratised, audience creation should be democratized within the marketing team with required governance using data export controls/ RBAC
- Managed, you may be trying out multiple versions (rules) of an audience to get the best definition but ensure to remove the trial versions to avoid confusion
- Exclusive, this is the most tricky aspect of audience taxonomy where you need to build exclusions when a visitor qualifies for both a lower value and a higher value audience, or a visitor needs to be unsegmented from an audience because they qualified for a mutually exclusive audience (example – prospect, logged in)
Tip:
Un-segmenting a visitor from DMP segment does not automatically inform the outbound marketing channels. Make sure you are passing the segment disqualification information to outbound channels, where required.
Wrapping Up
There are more details to each step, but hopefully this quick overview provides some clarity on the data to audience approach. This approach requires extensive involvement and collaboration from data and marketing teams. I recommend this approach since it helps to put maximum data at your disposal to explore and experiment audience building, instead of only building audiences based on existing (and sometimes limited) understanding of enterprise data.
As mentioned earlier, some companies prefer to take a use case led approach where you come up with use cases, identify relevant audiences and on-board corresponding data into DMP. This approach works well for MVP where you want to show quick results with first few campaigns, but may not be the best approach for scaling the use of DMP. It is also marred by process delays, hence huge lead times from use case identification to audience activation. However, it may work well for some clients like a publisher working with multiple advertisers to build audiences specific to advertiser’s use cases. Another approach is where companies hope to bring their 6-8 brand personas to life using DMP, but persona traits can be widely different from available data points (feasibility issue) and persona audience may be too broad for personalisation (targeting issue). Again, this may suit well in some cases, example you may build persona audiences around life moments where your products are aligned to such moments.
The audience building approach will differ for companies based on data capability and business need. But whichever approach you take, planning out your audience strategy is core and critical to DMP success and achieving effective personalisation.
