Enterprise AI & Decision Systems
Ahmed Muzammil
Every great customer experience starts with a good decision.
I build the systems behind those decisions, using data, AI and human judgement.
Most of my work is the plumbing around the model: what it is allowed to see, where it fits in someone's day, what success looks like, who keeps checking it, and who signs off on the risk.
My work is shaped by more than 15 years across technology, analytics and regulated financial services.
Vice President, Data Science & Advanced Analytics · Bank of Singapore · Singapore
- Experience15+ years across technology, data and AI
- Teaching2,000+ practitioners trained
- AwardTop Trainer Award, 2024 and 2025
- RecognitionVisionary Leader to Watch in AI and in Fintech, 2026
The system around intelligence
Beyond The Model
Most AI conversations stop at the model. Beyond The Model is for leaders who have to make AI work inside a company. I write about the part that decides whether it does: the data it can see, the decision it feeds and who signs off.
Subscribe to Beyond The ModelFree. I write it from the work I do every week.
- Data & Signals
- Context
- Intelligence
- Decision
- Workflow
- Outcome
- Learning
Control + Human Accountability
Questions
The questions I keep coming back to
Most of what I build and write starts with one of these.
What context does the system need, and what is it permitted to know and do?
Who remains accountable when intelligence becomes action?
Work
What I work on
Enterprise AI & Decision Systems
How intelligence moves from models into decisions, workflows and reusable organisational capability.
Agents & Governed Autonomy
How to design systems that can reason and act while remaining bounded, observable, permissioned and accountable.
Data, Analytics & Customer Intelligence
How behavioural and analytical signals become useful context for better decisions, interventions and conversations.
AI Adoption & Operating Models
How organisations move from isolated experiments to repeatable capability through workflow design, controls, ownership, feedback and adoption.
In all four, a person stays accountable for what happens next.
Frameworks
Models I use to make the logic visible
The models I actually use with teams, from writing a good prompt to designing an agent.
Work specification
PINPOINT
A practical framework for specifying work clearly enough for AI to perform it with the right level of context, verification and control.
PublishedAgent design
BLUEPRINT
A structured way to define an agent's purpose, responsibilities, context, tools, constraints, workflow, checks and hand-offs.
PublishedEnterprise capability
EVOLVE
A model for moving from isolated AI experimentation towards embedded organisational intelligence with context, controls, adoption and learning loops.
Published thesisDecision quality
The Fractal Leader
A decision lens for seeing what averages and aggregates hide, understanding context at the edges and deciding what changes when you zoom in.
PublishedBusiness systems
Business Maximizer®
An original systems framework for examining where value is created, where friction sits, what constrains performance and where a change would make the biggest difference, including when evaluating AI, automation and agent opportunities.
PublishedOpen source
What I'm building
Kognita
Alpha · active developmentProve an AI answer was permitted, and evidence it.
Kognita is my open-source alpha for one question: was the AI allowed to see this, and can you prove it?
Authorise first
Permission is checked before retrieval.
Fail closed
Unclear or missing permission stops the request.
Show evidence
Answers retain provenance and supporting evidence.
MIT licence
pip install kognitaBanking
Customer intelligence
Most of my customer analytics work has been in banking. A model can spot a behavioural signal. The harder question is what that signal means in the context of a real relationship, what decision should follow and how the system learns from the response.
- Signal
- Context
- Judgement
- Action
- Response
- Learning
Now
Questions I'm exploring now
When does a workflow actually need an agent?
How much autonomy should we delegate before evidence catches up?
What context improves a decision, and what context simply adds noise?
What does human accountability mean when an AI system can take action?
Ideas
Latest thinking
Builder note
Kognita 0.5.1 and 0.5.2: hardening released and documented
Kognita alpha 0.5.1 shipped the review hardening with a migration guide, and 0.5.2 documented every 0.5 feature in the README.
Read the noteBuilder note
Closing egress gaps in the AI gateway
Kognita alpha's gateway now redacts decoded JSON, screens every provider-read field and JSON key, forwards only known endpoints and refuses attachments it cannot inspect.
Read the noteBuilder note
Closing the library path: credentials and tier limits
Kognita alpha now requires a live workload credential on the library path and keeps tier 2 and above use cases gateway-only at runtime.
Read the note
Speaking
Selected speaking
Four recent talks, from agentic AI to customer analytics in banking.
Agentic AI in the Enterprise: Accelerating Transformation Through Data Modernization, Intelligent Operations, and Customer 360
Ortus Club · Coriander Leaf Group
Moderator
DEPLOY: From AI Pilots to Production – What Really Works?
GSDC Global AI Adoption Virtual Summit
Panellist
Leveraging Predictive Analytics to Optimise Customer Experience in Banking
Singapore Data & AI Conference (VDAC)
Speaker
Wealth Management in the Digital Era: Balancing Technology with Human Expertise
Fintech Revolution Summit Singapore · Traicon · Orchard Hotel
Speaker
About
I build things, then talk about what I learned building them.
I've spent more than 15 years working across technology, data, analytics and intelligent systems. These days I spend less time on models and more on what happens around them: getting the right data in, fitting into how people already work, and getting people to actually use it.
Regulated, relationship-led financial services has been one of my main proving grounds because the decisions are contextual, the information is sensitive and the human relationship still matters.