There’s a stat in our latest research that doesn’t surprise me. Over 80% of senior technology leaders across financial services, retail and health don’t have confidence in using their data for personalisation.
Being able to do this at scale, across a customer base that may total millions, with a genuinely personalised touch that doesn’t leave any particular segment of customers feeling ignored or potentially discriminated against, is no easy thing.
For industries such as financial services, personalisation is even more complex due to meeting existing and incoming regulations.
What form that personalisation takes varies by industry, but can include:
No doubt you’ll have specific use cases in your organisation. Almost all use cases rely on one thing: secure and trustworthy data foundations that ensure data is high-quality, readily available, representative of reality, and governed appropriately. This is the cornerstone for creating any personalisation for end users. Without it, results may be unreliable and, in fact, detract from the customer experience.
Getting our foundations right means we have core data availability - whether materialised through a data mesh, data lakehouse or other solutions. That data needs to be readily understandable and factually up to date, especially in the case of personal information (e.g. within the scope of GDPR). This creates significant requirements around things like data lineage, data ownership, and data cataloguing, topics that I’ve covered here.
It's critical to ensure the data we use to train the models that drive personalisation is fundamentally correct - otherwise the predictions and outputs of the models will be incorrect and unhelpful.
With the right data governance and data quality processes in place, alongside the right technical platform, we can start to create a unified view of the customer. It’s this that creates the bedrock for most personalisation programmes.
You might be wondering what I mean by the ‘right’ governance, and so on. To me, it’s about making it easy to find high-quality data and having no ‘shadow’ data quality processes, with the appropriate amount of security and process. If you’ve got the right framework in place for people to use data, support in ensuring compliance with regulation, etc., and you can trust the data, you’re in a good position.
The data model is also critical, especially in scenarios where we are joining data from different systems into one view of a customer. Poorly modelled data will make it significantly harder to join, relying on fuzzy matching or multiple mapping tables, and will inevitably lead to errors that corrupt the quality of results.
We must also consider identity resolution: how do we unify multiple fragmented identifiers (email addresses, account IDs, loyalty cards, app logins) into a single trusted record of an individual? Without this, personalisation risks targeting the same customer inconsistently across different channels.
It is also likely that we will need significant technical capability with real-time data infrastructure. This is to absorb both new context (e.g. AI assistants learning from the current customer call and adapting to their requirements on the fly) and ensure the most recent behavioural patterns are being considered during the inference process.
This step requires event streaming (Kafka, Pulsar), feature stores to provide models with the right signals, and scalable model-serving platforms that can work in both batch and real-time modes.
Finally, resilience and security cannot be overlooked. If we take an industry like financial services, personalisation engines must have failover mechanisms if models underperform or pipelines fail. They must also adhere to strict security standards to protect sensitive customer data.
Personalisation isn’t a one-size-fits-all journey. Many use cases may be common, but the method of execution and order of execution will change significantly with the business context.
For example, a wealth manager focused on retention may prioritise churn prediction and personalised client re-engagement, whereas a retail bank aiming to reduce cost to serve might prioritise AI assistants that give call centre agents rapid insights.
Business priorities should determine focus areas
The business challenge must dictate the personalisation use case. For instance, in retail/e-commerce, personalisation can be triggered at multiple points:
None of these are wrong; the focus must be on which point in the journey aligns with the business outcome that will deliver the biggest return at this point in time. This reinforces the need for early stakeholder alignment and agreement on strategic priorities.
Increasingly, retail and retail banking customers expect omnichannel personalisation. A recommendation given in-app should be visible when they phone a contact centre or walk into a store or branch. This requires orchestrating journeys across systems and channels, not simply optimising touchpoints in isolation.
As with any ML or AI product, it’s essential to define what success looks like at the outset. Success metrics might include:
There will also be industry-specific metrics; for example, in Asset and Wealth Management, it might be tracking average assets under management (AUM) growth where personalised product recommendations have been used.
Look at the revenue and cost benefits from personalisation
These metrics should be framed in terms of uplift versus control groups, using A/B testing where possible. Success should also translate to financial impact: improving churn prediction accuracy by even 5% (and acting appropriately on it) could add millions in customer lifetime value.
Importantly, ROI should be considered not only in revenue terms, but also in cost efficiency. Personalised self-service journeys can reduce call centre load, while AI assistants can reduce handling time per case. These productivity gains often make personalisation projects financially self-funding.
Personalisation systems must integrate with workflows and systems. Many organisations, especially those in financial services, have legacy platforms that aren’t AI-ready. The challenge is either to modernise those platforms or place abstraction layers in front of them to manage orchestration and consumption of AI outputs.
Test in controlled conditions and consider ecosystem data
Experimentation is vital. Running ML/AI pilots while considering the production architecture allows teams to test models in controlled conditions without prematurely overinvesting in infrastructure.
Some recommendation engines require heavy processing, which may be fine in batch but unsuitable for real-time environments. Experimentation should inform the right balance.
Equally, firms should consider ecosystem data sources such as open banking data, loyalty programmes, or third-party market data. These external inputs can greatly enrich personalisation but bring additional governance and security requirements.
Governance is often the difference between pilots that remain in labs and solutions that scale into production.
Key considerations include:
Ethical considerations call for responsible personalisation
Governance must also address the ethical frontier: How far should firms go in nudging behaviour? Encouraging saving may be positive, but steering specific investment choices could cross into regulated advice territory. Responsible personalisation must respect customer autonomy.
Technology alone will not deliver results. Organisational change is required:
Personalisation succeeds when it is embraced as a cultural capability, not just a technical one.
Metrics and dashboards alone aren’t enough. Organisations should actively gather customer feedback via surveys, review sites, and pilot groups. Negative signals (e.g. upticks in client attrition despite good KPIs) may indicate over-personalisation or a “creepy factor.”
Prototyping with smaller groups allows refinement before scaling. Engaging frontline staff (advisers, account managers) in feedback loops ensures personalisation enhances, rather than detracts from, the human touch.
Beyond today’s use cases, several trends will define the next wave:
The future will not be about standalone personalisation use cases, but about chaining them together into orchestrated, real-time journeys that adapt to a customer’s evolving context.
Personalisation is both a technical challenge and a business opportunity. The path to success involves strong data foundations, a real-time technical architecture, robust governance, and cultural adoption. It also requires clarity on success metrics, linkage to financial outcomes, and a balance between personalisation and customer autonomy.
Much of my advice here is aimed at organisations in retail, asset and wealth management and retail banking. However, it can be applied to virtually any organisation looking to deliver personalisation at scale.
Key takeaways - trustworthy data and modern AI and ML
What’s important is to combine trustworthy data, modern AI/ML, and a clear change programme. Only then can personalisation deliver on building stronger customer relationships, improving efficiency and generating new sources of growth.
If you enjoyed this blog, take a look at my one on building a data strategy.
Follow me on LinkedIn for more data insights.