Agentic AI and Hyper-Personalisation
Robo-advice and the new advisory
This module covers what changes when an AI system moves from producing answers to carrying out tasks. It explains how agentic systems plan, use tools and act, and then applies that to advisory work: robo-advice, hyper-personalised recommendations, suitability and mis-selling risk, the competitive position of the human adviser, and the supervision duties that arise when work is handed to an agent. The module works throughout from one principle: a task may be delegated to an agent, but accountability for it stays with the institution and the professional. In South Africa the FAIS conduct regime provides the legal framework for that.
- Explain how an AI agent plans, uses tools and acts, and where autonomy creates new risk.
- Assess robo-advice and hyper-personalisation for suitability, manipulation and conduct risk.
- Design the supervision and accountability arrangements required when work is delegated to an autonomous agent.
- Agentic-system literacy
- Suitability and conduct analysis of automated advice
- Supervision design for delegated AI
- Identifying where human advice adds value that an automated product cannot
Lessons in this module
How it lands across the four desks
This module applies most directly to your work. You learn how robo-advice and hyper-personalisation work, where they create suitability and manipulation risk, how FAIS applies to them, and which parts of an advisory relationship an automated product cannot replicate.
You maintain the suitability and supervision framework for automated advice: the FAIS Fit and Proper requirements that apply to it, the oversight design for delegated agents, and the allocation of accountability when an agent acts.
You look at the agentic automation now arriving in fraud and AML workflows, and apply Module 8's containment approach to it. An agent that can act across your systems widens the damage a single error or compromise can cause, so its access rights need to be assessed alongside any productivity benefit it offers.
You look at agentic automation of underwriting workflows, and apply this module's supervision model, in which the work is delegated but accountability is retained, to any agent that touches a credit decision.
Key literature · 6 sources
Every module rests on a verified scholarly and institutional evidence base. The full core and further reading lists open with the module.
- Chan, A. et al. (2023) 'Harms from Increasingly Agentic Algorithmic Systems.' FAccT '23: the risk vocabulary and the agency-as-spectrum framing.
- Shavit, Y. et al. (2023) 'Practices for Governing Agentic AI Systems.' OpenAI: the governing principle and the supervision practices this module operationalises.
- D'Acunto, F., Prabhala, N. & Rossi, A. G. (2019) 'The Promises and Pitfalls of Robo-Advising.' Review of Financial Studies 32(5): the empirical benefits and limits of automated advice.
- IOSCO (2016) 'Update to the Report on the IOSCO Automated Advice Tools Survey': the international supervisory frame for automated-advice suitability.
- Susser, D., Roessler, B. & Nissenbaum, H. (2019) 'Online Manipulation: Hidden Influences in a Digital World.' Georgetown Law Technology Review 4(1): the service-versus-manipulation distinction.
- SARB Prudential Authority & FSCA (2025) 'Artificial Intelligence in the South African Financial Sector': Box 1 on automated advice and the FAIS Fit and Proper requirements; adoption and governance findings.