JMN · Workflow Automation & AI

Custom AI agents, built around
your exact workflow.

Your workflow · scoped to your desk Any function · research, finance, ops End-to-end · a whole process We stay · built, deployed, supported

Every desk works differently. JMN builds automation around the process you already run — pulling news from your platforms to your desk, turning inputs into first-draft outputs, and taking the recurring manual steps end-to-end. Not one tool for everyone; an agent scoped to how you work.

See what we automate
Any workflow
Research, finance, accounts
or operations
End-to-end
A whole process automated,
not a single step
Built with you
Scoped, deployed, and
supported after launch
The problem

Too much of the week goes
to repetitive work.

A large share of the desk's time goes to data and production work — triaging news across dozens of names, wading through long documents, running the same processes every cycle. It is necessary, but it is not analysis. These are the areas JMN automates:

News triage
News scanned across your whole coverage from several platforms every morning, and filtered to what is material — not forty raw headlines to read.
Large documents
Long filings, contracts and reports read end to end and turned into structured, usable data — at a scale no desk wants to work through by hand.
Custom AI agents
Bespoke agents built for the specific, repeating steps your desk runs — the process no off-the-shelf tool was ever shaped around.
RAG systems
Thousands of your published reports turned into one knowledge base the AI can answer from — ask a question, get an answer with its source.

And these are only the obvious ones. No two desks run the same process, so we don't work from a template — we customise deeply to the tasks your team actually repeats. Start with a free consultation: we map your research workflow end to end, then design and build the automated pipeline around it.

What we automate

The recurring work —
shaped to your desk.

This is for the bespoke work — the multi-step processes and AI systems that sit outside an off-the-shelf tool. Repeatable model updates run in the Excel Assistant and repeatable report drafts in DraftPilot; here we build everything around them. A few of the systems we build:

Pipelines
Custom research pipelines
Multi-step · end-to-end
Pull the news for every coverage name and organise it in one place
Draft a tailored note per company — the whole process, start to finish
Built for the one-off, bespoke work a repeatable tool won't cover
Document processing
Documents at scale
Unstructured → structured
Read filings, contracts, PDFs and forms — thousands at a time
Extract the fields you need into clean, structured data
Route the outliers to a person, not a black box
Knowledge systems
Search & RAG over your library
Ask your own library
Take the thousands of reports your house has published and make them one searchable knowledge base
Anyone can ask a question in plain English and get an answer drawn from across all of them
Every answer points back to the exact report it came from
Beyond research
Finance & operations
Any repeating process
Reconciliation, reporting and other finance workflows
Accounts and operations tasks that repeat on a schedule
If it is manual and it repeats, it is a candidate for automation
The techniques we use — the right one for the job
Custom AI agents
LLM agents with memory and tool access, wired into your systems to carry a task from start to finish.
Agentic workflows
Several agents orchestrated across steps — retrieve, reason, act — with a person at the checkpoints that matter.
RAG systems
Retrieval over your own documents, so answers stay current and cite the source they came from.
Fine-tuning & adaptation
When a model needs to do one thing very well, we adapt it to your domain — used only where it earns its place over RAG.

RAG, fine-tuning and agents are not a ladder — each solves a different problem, and the right build is often a combination. We pick what fits the task, not the other way round.

How we build it

Modern AI, built
to a finance standard.

The techniques are current — retrieval-grounded models, large-scale document extraction, multi-step agents. The discipline is ours: in research, an output is only useful if you can trust it and trace it. So every pipeline we build is grounded in your data, reviewed by a person, and auditable end to end.

01
Grounded in your data
Outputs are retrieved from your own documents and data — so answers are traceable to a source, not invented. It is the difference between a system you can publish from and one you cannot.
02
Human-in-the-loop
The system proposes; a person approves at the checkpoints that matter. Automation handles the volume; the analyst keeps the judgement and the accountability.
03
Auditable by design
Every step and decision is logged, so a pipeline stands up to a compliance review. No black box — you can always see what happened and why.
04
Scoped to your stack
Built around the data, systems and tools your team already uses. We fit the way you work today, rather than asking you to move onto a new platform.
How we engage

We map it, build it,
and stay with you.

A simple path — start with a conversation, see it work on your own desk, then roll it out. We build, test and deploy, and support it as your workflow changes.

Step 01
Map
A short call to understand how your desk works and the tools you already use.
Step 02
Scope
We tell you what is worth automating — and, just as clearly, what is not.
Step 03
Build & pilot
One workflow, built and tested on your desk — free to evaluate before you commit.
Step 04
Deploy & support
We roll it out and keep it working, evolving as your workflow evolves.

Clear guidance on where AI fits — and where it doesn't. Automation handles the repetitive work; every output reaches the desk for review, and the analyst's judgement stays the analyst's.

How it starts

Tell us your workflow.
We'll show you what to automate.

Start with a short call. We map how your desk works, point out what is worth automating — and what isn't — and build a pilot on your own workflow, free to evaluate.

Email us directly