Written by Dan Blagojevic PhD, Chief Data Scientist
Personalised decisioning saw a huge boom in the early-2010s. With adoption of could technology providing access to far cheaper and flexible compute power, businesses were finally equipped with the tools to unlock the vast potential from terabytes of data previously simply “hoarded”.
Increased personalisation was achieved by applying predictive models and machine learning to vast amount of behavioural data. Businesses began defining ever more granular customer segments, with accompanying granular decisioning strategy. The notions of pre-approved offers, sitting ready to be deployed within segment-level customer journeys became common-place.
However, there is an asymptotic ceiling to be achieved simply by throwing more compute power and more data to existing methods, tech and operating models. This is, in large part, due to continued use of rules-based decision strategy design (as distinct from decision orchestration).
The old adage of ‘a system is only as strong as its weakest link’ comes to the fore. No matter how granular, accurate and timely the insight provided by AI for strategy design, most organisations still simplify strategy to something manageable in an Excel spreadsheet. One reason is justifiable: transparency, auditability and minimising risks of inadvertent bias. Afterall it is not that long ago (the Credit Crunch) that automated credit line decreases pushed many households over the edge (which we will cover in a subsequent post). However, two other reasons for this ‘simplicity’ are:
- Business mindset: the misplaced notion that advanced AI is a ‘black box’ (even though the only unexplainable black box is the human brain).
- The (limited) capacity of any one individual to ‘join the dots’: to reason through ultra-high-dimension cause-effect insight provided by sophisticated ML, and define the best decision strategy for a customer.
Within these constraints, a highly granular view of customer behaviour, and future needs, is often collapsed to a standard set of products and services. Personalisation kicks in when it comes to how these are prioritised and positioned in the customer journey, but the underlying choice has already been standardised.
Enter the reasoning agentic AI system. Recent impressive progress in reasoning ability of AI agents is opening a whole new way in which strategy design can be done. Bringing together the power of knowledge graphs, large language models and prompt engineering, the prospect of agentic AI systems co-developing personalised strategy with their human counterparts is a realistic prospect. These systems stitch together very diverse data types, and reason through high-dimensionality cause-effect pathways much faster than the human brain. More importantly, they find connections and paths which we may miss altogether.
In the context of designing the best borrowing offer for Mr & Mrs Smith looking for a loan, we might very well conclude it is not a loan at all. Instead, they are better off using part of their easy access savings to reduce the borrowing amount and consolidate their existing borrowing in a money transfer credit card with an interest free period.
Reasoning AI systems act like an orchestra: a collection of versatile musicians, coordinated by a conductor who ensures each member is performing within their defined role, at the right time. In our context, dedicated AI agents scanning the full range of products, cross-referencing Terms & Conditions, executing what-if scenarios for income, cost, losses. Working in tandem to provide a well-synchronised fully personalised strategy for each customer.
Retaining the human in the loop and establishing strong guardrails and back-stops remains paramount (we reported on this in our previous blog). Just as they can find connections we might miss, we need to avoid inappropriate proposition furnished by a hallucinating agent. Instead, they can act as very powerful assistants to the decision strategist.
Reasoning AI systems are not, and will not be for the foreseeable future, autonomous decision-makers. A tool to deliver personalised strategy, beyond personalised decision orchestration.
Key Takeaways:
- The current trajectory in personalisation is reaching its limits. The limiting factor is the (decision) strategy design approach. Lagging behind sophisticated downstream decisioning, strategy design is still stuck in the world of simpler rules.
- Emerging agentic AI systems with advanced reasoning capability offer a real prospect to transform strategy design from static rules to a dynamic, customer-responsive strategy.
- Evaluate incrementally: agentic AI systems should never execute autonomous strategy design: allow them enough freedom to design customer-tailored strategies in controlled tests. Evaluate, iterate, roll out incrementally.
- Retain the Human in the Loop: as well as setting up AI guardrails, retain robust human oversight of final strategy execution.
Contact Optima Partners today to find out more.
