Merlin Liquidity
MERLIN Agentic Treasury Pilot
MERLIN is a treasury pilot for finding out, on evidence and safely, where agentic AI adds value, what controls it needs and what production should contain, before any major AI investment or architecture decision is made.
Most companies are asked to believe in AI before they can see it work on their own business. MERLIN proposes the other way round: take one real operation, run it with AI agents beside the people who do it today, and judge the result on evidence.
A treasury desk's day is short, has firm rules and clear numbers, and every decision can be checked against the books: what is coming in, what must go out, what to invest and what to bring home. That makes it a fair place to test whether AI helps, and to see where it does not.
The pilot leaves a tested agent design, controls, code, test cases and a production specification, ready for the AI architecture and platform of your choice. Whatever you decide, your treasury policies and controls come out tested and firmed up.
Results in weeks, not months, because we start from a proven treasury library instead of building infrastructure first. Every outcome moves you forward, including "not yet".
The MERLIN Way → The method extends to adjacent finance operations and working-capital processes.
Every part has been through real daily funding plans and carries its own tests. In a pilot they are set to your policy and your data, then improved on what the AI-generated daily liquidity plans show, over many runs.
And other tools, checks and steps, each added when a real plan showed it was needed.
The same method extends to adjacent finance operations: working capital, receivables and payables, cash forecasting.
Conventional software gives the same answer to the same input, traceable to a rule someone wrote. A model gives the likeliest answer: it can vary between runs, state something false with confidence (a hallucination), and be steered by words planted in what it reads (prompt injection).
In well-designed AI based systems, conventional code prepares the facts and checks every figure the model gives back. Guard rails limit what it can see and do. A person approves before anything consequential happens. The pattern MERLIN runs: AI proposes, code checks, a person decides.
Knowing how a model works makes choices clearer, whether you're placing AI in your company or learning what it is. AI Unboxed shows how a model is made, what happens inside an answer and how a real system wraps it, with a live chat with a raw model.
MERLIN Studio is where we put every pilot together. Your treasury workflow is laid out as a map of steps, each built from the proven treasury library, so the pilot can be adjusted quickly as your team learns what works.
A map describes what your operation does, not the technology underneath, so a design proven here can be taken to LangGraph, CrewAI and the cloud providers' agent platforms, or to your own systems.