01 ·IN PRODUCTION
ANALYTICAL CLASSIFICATION
The agent assigns each transaction to the correct cost centre, line of business and entity. This works today, in production.
The Territory
01 — Eight process families
The range of finance work an agent can now carry.
Not everything in finance is ready for an agent. A great deal of it now is — wherever the work is evidence-bound and the answer is checkable. This is the ground we work on.
PROVEN
Record to Report
The books, kept and closed.
Ingestion, classification, reconciliation, accruals, month-end close, consolidation, management reporting.
IN PRODUCTION
Order to Cash
Getting paid, cleanly.
Cash application, AR aging, collections and dunning, credit files.
IN PRODUCTION
Procure to Pay
Spend, controlled.
Invoice capture, three-way match, duplicate detection, payment-run preparation, expense audit.
IN PRODUCTION
Treasury & Cash
The cash position, live.
Balance consolidation, short-horizon forecasting, liquidity and covenant monitoring.
IN PRODUCTION
Planning & Analysis
The numbers behind the decision.
Board packs, variance analysis, margin and profitability intelligence, rolling forecasts, cost allocation.
IN PRODUCTION
Tax & Compliance
Filed on time, tied to source.
VAT/return preparation, statutory file generation, obligation tracking, provision support.
BUILDABLE
Investment & Funds
The fund's finance machine.
Portfolio-company reporting, diligence document analysis, fee and waterfall calculation, LP reporting, position reconciliation.
BUILDABLE
Controls, Audit & Risk
The trail, watched.
Anomaly detection, evidence tie-out, continuous close monitoring, parallel-run reconciliation.
PROVEN — parallel-run against a human baseline · IN PRODUCTION — live in production delivery · IN DESIGN — specified, shown as intended behaviour · BUILDABLE — within the proven architecture, not yet built · EXPLORATORY — under investigation, no claim made
We colour our own work honestly. We will always tell you which is which — and we will never promote a claim before it has earned it.
Beneath all of it sits one engineering discipline — the patterns we build every system on.
Sooner
02 — Time-to-knowing
Not just correct. Sooner.
Getting the entry right is the floor, not the ceiling. Once the work is posted correctly and allocated correctly, something else becomes possible: the numbers can be interrogated.
The agent doesn't out-think your team. It compresses the time between a problem existing and you knowing about it — from months to days. Not smarter. Sooner.
Treasury
03 — Ask what-if
IN PRODUCTIONANONYMISED DATA
A thirteen-week cash forecast with a P10–P90 range, base and downside scenarios, and suggested collection actions. Ask a what-if question and the projection recomposes.
Board Pack
04 — Drafted for the board
IN PRODUCTIONANONYMISED DATA
Drafted commentary, highlights, points of attention and decisions for the board — with the variance narrator shown in design.
11,808
Settlements calibrating the forecast
2,000
Simulations per 13-week fan
62%
Of allocation patterns resolve to one centre
1,900
Allocation patterns discovered in one client
Anonymised client data · real ledger history.
The Allocation Layer
05 — Where the margin went
A correct entry doesn't tell you where the margin went.
Beneath both of those sits the allocation layer. A correct entry tells you the books are right. It does not tell you where the margin went. That takes a second act: assigning every transaction to the right cost centre, the right line of business, the right entity — inside a structure that is unique to your business and changes from one year to the next.
This is the layer that makes margin visible. It is what feeds the forecast, the variance, and every business-specific view built on top. Without it, everything a CFO wants to ask is guesswork.
02 ·IN PRODUCTION
THE CONVENTION CONFIGURED TO YOUR MODEL
There is no universal allocation model — every business allocates differently. The convention is the rule structure discovered from your own history: by centre, by suffix, by document type. It is what tells the classifier how you think about your analytics — which is why it adapts to a structure that changes year to year.
03 ·IN DESIGN
DRIFT DETECTION CONFIGURED TO YOUR MODEL
As the year progresses and the structure evolves, the system detects the deviation and raises it for review. It does not rewrite the rules behind your back. The convention stays alive, with a human confirming every change.
Why this matters
The lookup gives you the data. The convention gives you the rule.
Of roughly 1,900 allocation patterns discovered in one client's history, 62% always resolve to the same centre — for those, a query against history is enough. But the patterns that split across two to seven or more centres carry most of the volume. That is where the convention earns its keep: it learns the stable rule behind each split, with its proportion and its variance.
Stable splits are applied automatically. Dispersed ones are raised for review. Never decided in silence.
That is the work. Here is the discipline underneath it.