Europe/US Roundtable

Guest: Julian Berry, Founder, UniFida

• Different ways to use CDPs as repository of customer behavior, in addition to activation
• Analyze long term customer value
  • Track cohorts of customers over time, show number of customers and value per customer by cohort
  • Look at retention rate and value  by source channel
  • Could further analyze cohort members by demographics, locations, etc.
• Customer value planning
  • understand value brought by customers by time period (e.g. show seasonality)
  • show value from new customers vs existing customers by period
  • use to build business plan, based on how many new customers are needed (and when) to meet revenue targets
  • can estimate necessary spending for recruitment and CRM
  • Need to include all channels in analysis; CDP unifies data from different sources to do this
• Multi-touch attribution
  • captures direct channels, not indirect (TV, sponsorship, outdoor)
  • based on understanding steps (touches by channel) during customer journey
  • each channel has a role in making a sale: initialize, hold, close
  • assign fractional credit to each channel for revenue from each sale
  • look at results across all sales to estimate value and ROI by channel and campaign
  • can also calculate number and value of impacted sales (without fractional allocation)
  • problem is setting weights; need to avoid skewing results to favor a particular channel, should be transparent so weights are clear
  • can factor in timing of touches so less recent touch gets less credit
  • difficult to set weights, look at patterns of journeys to assess impact of different channels
  • fallbacks are to set even weight for each touch, to assign intuitive weights; best is to apply algorithms to calculate more precisely
  • basic model is to assign weights and have decay function based on time
  • even if weights are not entirely accurate is much better than last touch, which is what most people use
  • MTA answers only one part of the question; also need econometrics for complete impact for MTA
  • need data scientists to dig into data to answer more detailed questions
  • see AI used to optimize tactically e.g. pick best content to optimize click rate or conversion rate, but not at strategic level such as channel allocation for budget planning
  • asked three AI companies if they could do MTA better, and they couldn’t; do see AI used for defining cohorts, optimizing channel sequence within journey
• Using AI to develop propensity models
  • CDP is perfect data source for building predictive models
  • existing analysis techniques work well, if use right technique for each application;
  • genAI isn’t necessarily improvement, but has other applications:
    • genAI can help improve some data e.g. extract meaning from unstructured data
    • genAI can be starting point to build predictive model e.g. help to write Python code, but not actual model
    • genAI can give human interface to data, can interpret data for non-tech users
• Getting the most from your CDP data
  • is division between business analysts (in financial team) and marketing analysts (further divided into paid media team vs owned media team). 
    • financial team/business analysts look at high level customer revenue buy often misses marketing data such as detail by channel.
    • owned media team looks at bottom of funnel, e.g. ad clicks and opens. 
  • financial team uses cloud data warehouse, marketing analysts generally don’t.  Data team uses their own tools
  • comprehensive analysis requires senior person to get different analysts in the same room to work on
  • is growing use of direct mail (in UK) for catalogs, brand profile

APAC Roundtable

Guest: Sriram Sitaraman, Director of Technology, Material+

• Businesses increasingly integrate AI into CDPs for customer engagement and personalization
• Different companies treat CDP differently:
  • companies with existing activation tools don’t need that in CDP;
  • marketing team will want campaigns included so can have one platform
  • data team only wants unification and activation capabilities
  • IT team looks at supplying data to marketing
• Companies with existing data warehouse, often don’t need to put data in CDP; but do need CDP for calculations and real-time access using data they read directly from the warehouse.  Many companies worry about the security implications of copying data into CDP, if it will be read human or AI agents.
• Maybe will be shift between default assumption that most data resides in CDP and access some data on demand, to assumption that access most data on demand with some residing in CDP; is good way to avoid fighting with data lake/warehouse
• See CDP as tool for marketing department to access data, which is stored elsewhere.  Makes CDP less technical, having more users
• CDP applications for customer engagement:
  • Hyper personalization (messages tailored to individual customer preferences) is not always necessary or better than segment-based messaging; effective in retail;
  • Giving real-time view of data to agents to drive personalized interactions is very valuable (distinguish from hyper personalization, of automated interactions)
• Risks and challenges:
  • Data privacy and compliance: must tie CDP into over-all privacy strategy
  • Integration: some connectors will be available out of the box, building custom connectors for other sources and applications can take significant work, must understand details to assess labor
  • AI models: be careful to assess bias and accuracy
• Practical challenges:
  • Understanding unique individuals; what’s required depends on use case
  • Cold start: how to target new users when don’t have much data; initially can target on general information such as location and source, then switch to hypersonalization as collect data
  • Often have relatively little data about many users
  • Are also issues with users understanding what CDP does and how to use it
    • To mitigate, start with defining target use cases, and which parts of the use cases the CDP should support; this determines what’s desired and what’s practical for CDP
• Presence of data doesn’t guarantee success:
  • Need to look at whether have right data for target use cases
  • 3 core CDP use cases:
    • insights to guide strategic decisions
    • improve efficiency of digital media
    • use behavior data to improve message targeting
  • Must examine at key assumptions that justify CDP: that CDP can meet needs, is lower cost than self-built solution, can integrate with existing systems