If you want to apply personalisation, you need to understand your customer. Every interaction in the customer journey helps you do that. By taking a customer-centric approach, you’ll be able to make data-based decisions and improve your customer experience.
In the Visser & Van Baars webinar 'Organising technology to monetise on customer experience,’ on July 1, 2021, data experts from Crystalloids, ANWB and Google Cloud got together to discuss the question of how best to organise your marketing technology and analytics to better serve your customer. Below, a brief summary.
Jan Hendrik Fleury: flexible, future-proof marketing technology
First off was Jan Hendrik Fleury. As commercial director at Crystalloids, he helps clients such as ANWB, Calvin Klein and Rituals get more out of their data by building customer data platforms (CDPs) for data, analytics and engagement, based mainly on Google Cloud. Fleury is also a member of the Data, Decisions & Engagement (DDE) committee of the Data Driven Marketing Association (DDMA), an industry association that helps 335 organisations take their data-driven marketing to the next level.
Fleury first emphasised the importance of customer centricity and taking a customer-centric approach: “If you give customers a seamless customer experience, they will be willing to share their data. As a result, you can be more relevant throughout the customer journey, make better decisions and gain a competitive advantage.” In his online lecture, Fleury explains how to make sure you have sufficient data, how to automate your data flows, perform quality checks and realise a flexible and future-proof cloud and marketing technology stack.
Everything starts with data
The crucial first step of course is to collect data. Fleury: “That might be your own data or external data, collected via a data-management platform. Then you have to have the right privacy permission and be able to manage it.” Processes such as identity management, tag management and metadata management all take place in the enterprise data layer, facilitated by platforms such as Google BigQuery or Amazon Redshift.
Once you have all the ingredients to support customer-centricity, such as demographics and behaviour, you can start modelling; examples might be lookalike modelling, propensity modelling, engagement modeling or churn prediction. You can then start micro-targeting – distributing the content via owned, paid or earned channels.
A logical architecture for data and analytics
What is a logical architecture for data and analytics? And what do you need for an analytical data store such as Google BigQuery, Snowflake and Amazon Redshift? As became obvious during Fleury’s explanation of the architecture of Body&Fit, which uses real-time serverless data and an analytical CDP in Google Cloud, that is something that is can’t be easily summed up in just a few sentences.
So how does it actually work in practice, earning money through personalisation? To put it very simply – you have your data, you match it with the customer transaction ID and you score with an AI platform. You can apply two strategies to this: audience and bidding, making data-based decisions depending on your data and your chosen strategy.
Fleury closes his lecture with six takeaways:
1. Start with a clear vision and data strategy and seek support from management for them.
2. Prioritise the most important user stories and coordinate them with programme management.
3. Don't wait until you have 'everything', because you can achieve a lot without all data being available.
4. Start fast, dare to fail and be agile.
5. Avoid getting into a vendor lock-in because you’ve chosen an all-in package for your CDP. Rather connect applications on a loose basis and store your data and analytics in a central place.
6. Apply privacy and security by design and be transparent to the customer. Because of the control they have over their data, they will trust you and be willing to share their data.
Arnold Moeken: omni-channel personalisation at ANWB
A good example of personalisation in practice is ANWB's omni-channel personalisation programme, which earned the organisation the DDMA Customer Data Award. Arnold Moeken, manager of the Customer Analytics Department at ANWB and member of the DDMA’s DDE committee, talked about this programme during the webinar.
With almost 5 million members, ANWB has a wealth of customer data. “Previously we only had a number of individual initiatives, but in 2016 we started an omni-channel personalisation programme to make better use of customer contacts across all channels,” Moeken recalled.
Right propositions, moment, channels and method
“The aim of the programme is to offer the right propositions, at the right time, through the right channels and in the right way,” said Moeken. It all starts, of course, with data. After collecting and extracting data, you add the intelligence; by combining this with the right content, marketing proposition engine and dashboards, ANWB can present the best propositions to its members and customers.
This is a continuous loop, explains Moeken. “We constantly monitor and optimise our propositions via our dashboard.” The models are also constantly being refined and ANWB has a model factory specifically for this purpose, with more than 200 algorithms that predict, daily, the likelihood of customers buying certain services or products.
Ambition and vision are everything
Moeken stressed that ambition and vision are everything; the organisation must arrive at one single vision around personalisation, and have the ambition to make its personalisation programme a success. Moeken: “Previously, all initiatives were in silos within our organisation. Fortunately, we collectively tuned in to the one single ambition and were able to connect and optimise all channels.”
According to Moeken, a number of preconditions are important. “You need structural and relevant propositions for example, but for us the most important thing was governance.” Moeken explained that the programme they drew up, based on their ambition and vision, was multidisciplinary, including marketing, IT and business line departments.
Five principles and a seven-layer model
ANWB designed five architectural principles:
1. Centralised data
2. Centralised analytics
3. Centralised campaign management
4. Centralised CDP/identity management
5. Centralised content management
These principles have been translated into a seven-layer personalisation model, comprising touchpoints, content delivery, content management, CDP, campaign management, analytics and data. In turn, this model has been sub-divided into an analytical and an operational CRM approach.
Analytical CRM Approach: technology
Moeken: “We used to have BlueConic, Portrait and a few custom-made things. However, we faced a lot of problems, such as not being able to scale. Then we started with a new central data platform, AWS.”
ANWB has also introduced a new analytics platform with KNIME, as well as embracing a new campaign management platform with IBM Watson, which has since been exchanged for Unica Campaign. Gradually, ANWB has increasingly centralised all its marketing technology.
The most important lessons
So has the personalisation programme paid off? The facts and figures speak for themselves: for example, both the share of impressions and the share of clicks on the website, as well as the CTR in the app, have increased by more than 200% and even the printed magazine Kampioen is now being personalised.
Moeken ended his talk by listing a few important lessons for organisations that want to personalise:
1. Maintain a ‘holy belief’ in personalisation taking precedence over a business case.
2. Make sure you have the right knowledge and expertise. If you don't have it in-house, outsource it – as ANWB did with Crystalloids.
3. Architecture is the basic ingredient, think about where you want to go in advance. This prevents a vendor lock-in.
4. Collaboration is crucial: instead of working in silos, ANWB now works as one single company.
Rokesh Jankie: an Architecture for Marketing Analytics
Finally, a member of the DDMA’s AI committee took to the floor – Rokesh Jankie, Google Cloud Sales Engineering Manager at Google Cloud in Amsterdam. Jankie addressed questions such as: how do you set up an architecture for marketing analytics, and how do you extract value from data?
Jankie listed three use cases for marketing analytics that are currently hot topics: understanding the customer journey based on trend-spotting; self-service analytics and customer segmentation; predicting marketing outcomes using lifetime value (LTV) prediction and purchase prediction; and personalising the customer experience through sentiment monitoring, data-driven segmentation, and a personalisation engine.
LTV prediction: get to work yourself
Jankie explains that data professionals who are only SQL masters can work very well independently with lifetime value (LTV) prediction. After you’ve collected data, you enter it into a cloud trainer, using a pre-trained or own model, transform and analyse it and can then visualise it using a tool like Google Data Studio or Tableau.
Then you activate the data. In other words, you cash in on its value by converting insights into actions that improve the customer experience. This can be done, for example, via advertising platforms, e-mail platforms, content optimisation, CRM platforms and social platforms.
“The interesting thing is, that in a world where the cloud plays such a central role in business operations, it’s easy to consult multiple sources of information to get a 360-degree view of your customer,” says Jankie.
Step-by-step plan for customer LTV calculation
To calculate the customer lifetime value yourself, you can use a data warehouse such as Google BigQuery or Amazon Redshift as a basis. “As a data scientist you have to deal with people who copy data sets, often terabytes in size, so you need a subset. But which data do you choose? And is that a good representation of the entire dataset? At Google we said: why not bring the modeling – the AI part – to the data instead of the data to the model?” So says the Google specialist.
Jankie’s employer did just that with BigQuery ML, which allows you to create and run machine learning models via standard SQL queries. Jankie sums up the advantages: building predictive models without making assumptions based on the data; quickly implementing iterations; plenty of options for post-optimisation; and the possibility of using algorithms from TensorFlow, K-means Clustering and Matrix Factorization, among others. In addition, you can scale without operational hassle, there are minimal IT requirements, users need less machine learning (ML) knowledge and centralised governance and monitoring tools are available.
Jankie's conclusion: BigQuery ML ensures the democratisation of machine learning, partly because data analysts can build models with existing BI tools and spreadsheets and they only need SQL knowledge.
Jankie concludes his lecture by briefly discussing Google's AutoML, a tool to train custom ML models with minimal effort and ML expertise. “You can use AutoML for things like customer lifetime value and predicting customer conversion and churn. Thanks to the graphical user interface, you don't need to have programming skills in order to use ML applications, you just upload a sheet and select the correct column, then AutoML generates the model that best fits your data.”